server : do not default to multiple slots with speculative decoding (#17017)

* server : do not default to multiple slots with speculative decoding

* cont : fix

.devops/s390x.Dockerfile

@@ -24,8 +24,9 @@ RUN --mount=type=cache,target=/root/.ccache \
         -DCMAKE_C_COMPILER_LAUNCHER=ccache \
         -DCMAKE_CXX_COMPILER_LAUNCHER=ccache \
         -DLLAMA_BUILD_TESTS=OFF \
-        -DGGML_BACKEND_DL=OFF \
         -DGGML_NATIVE=OFF \
+        -DGGML_BACKEND_DL=ON \
+        -DGGML_CPU_ALL_VARIANTS=ON \
         -DGGML_BLAS=ON \
         -DGGML_BLAS_VENDOR=OpenBLAS && \
     cmake --build build --config Release -j $(nproc) && \
@@ -103,6 +104,7 @@ FROM base AS light
 WORKDIR /llama.cpp/bin
 
 # Copy llama.cpp binaries and libraries
+COPY --from=collector /llama.cpp/bin/*.so /llama.cpp/bin
 COPY --from=collector /llama.cpp/bin/llama-cli /llama.cpp/bin
 
 ENTRYPOINT [ "/llama.cpp/bin/llama-cli" ]
@@ -116,6 +118,7 @@ ENV LLAMA_ARG_HOST=0.0.0.0
 WORKDIR /llama.cpp/bin
 
 # Copy llama.cpp binaries and libraries
+COPY --from=collector /llama.cpp/bin/*.so /llama.cpp/bin
 COPY --from=collector /llama.cpp/bin/llama-server /llama.cpp/bin
 
 EXPOSE 8080

.github/labeler.yml

@@ -76,6 +76,10 @@ ggml:
     - changed-files:
         - any-glob-to-any-file:
             - ggml/**
+model:
+    - changed-files:
+        - any-glob-to-any-file:
+            - src/models/**
 nix:
     - changed-files:
         - any-glob-to-any-file:

.github/workflows/build-linux-cross.yml

@@ -4,49 +4,49 @@ on:
   workflow_call:
 
 jobs:
-  ubuntu-24-riscv64-cpu-cross:
-    runs-on: ubuntu-24.04
+  # ubuntu-24-riscv64-cpu-cross:
+  #   runs-on: ubuntu-24.04
 
-    steps:
-      - uses: actions/checkout@v4
-      - name: Setup Riscv
-        run: |
-          sudo dpkg --add-architecture riscv64
+  #   steps:
+  #     - uses: actions/checkout@v4
+  #     - name: Setup Riscv
+  #       run: |
+  #         sudo dpkg --add-architecture riscv64
 
-          # Add arch-specific repositories for non-amd64 architectures
-          cat << EOF | sudo tee /etc/apt/sources.list.d/riscv64-ports.list
-          deb [arch=riscv64] http://ports.ubuntu.com/ubuntu-ports/ noble main universe
-          deb [arch=riscv64] http://ports.ubuntu.com/ubuntu-ports/ noble-updates main universe
-          deb [arch=riscv64] http://ports.ubuntu.com/ubuntu-ports/ noble-security main universe
-          deb [arch=riscv64] http://ports.ubuntu.com/ubuntu-ports/ noble-backports main universe
-          EOF
+  #         # Add arch-specific repositories for non-amd64 architectures
+  #         cat << EOF | sudo tee /etc/apt/sources.list.d/riscv64-ports.list
+  #         deb [arch=riscv64] http://ports.ubuntu.com/ubuntu-ports/ noble main universe
+  #         deb [arch=riscv64] http://ports.ubuntu.com/ubuntu-ports/ noble-updates main universe
+  #         deb [arch=riscv64] http://ports.ubuntu.com/ubuntu-ports/ noble-security main universe
+  #         deb [arch=riscv64] http://ports.ubuntu.com/ubuntu-ports/ noble-backports main universe
+  #         EOF
 
-          sudo apt-get update || true    ;# Prevent failure due to missing URLs.
+  #         sudo apt-get update || true    ;# Prevent failure due to missing URLs.
 
-          sudo apt-get install -y --no-install-recommends \
-                  build-essential \
-                  gcc-14-riscv64-linux-gnu \
-                  g++-14-riscv64-linux-gnu
+  #         sudo apt-get install -y --no-install-recommends \
+  #                 build-essential \
+  #                 gcc-14-riscv64-linux-gnu \
+  #                 g++-14-riscv64-linux-gnu
 
-      - name: Build
-        run: |
-          cmake -B build -DLLAMA_CURL=OFF \
-                         -DCMAKE_BUILD_TYPE=Release \
-                         -DGGML_OPENMP=OFF \
-                         -DLLAMA_BUILD_EXAMPLES=ON \
-                         -DLLAMA_BUILD_TOOLS=ON \
-                         -DLLAMA_BUILD_TESTS=OFF \
-                         -DCMAKE_SYSTEM_NAME=Linux \
-                         -DCMAKE_SYSTEM_PROCESSOR=riscv64 \
-                         -DCMAKE_C_COMPILER=riscv64-linux-gnu-gcc-14 \
-                         -DCMAKE_CXX_COMPILER=riscv64-linux-gnu-g++-14 \
-                         -DCMAKE_POSITION_INDEPENDENT_CODE=ON \
-                         -DCMAKE_FIND_ROOT_PATH=/usr/lib/riscv64-linux-gnu \
-                         -DCMAKE_FIND_ROOT_PATH_MODE_PROGRAM=NEVER \
-                         -DCMAKE_FIND_ROOT_PATH_MODE_LIBRARY=ONLY \
-                         -DCMAKE_FIND_ROOT_PATH_MODE_INCLUDE=BOTH
+  #     - name: Build
+  #       run: |
+  #         cmake -B build -DLLAMA_CURL=OFF \
+  #                        -DCMAKE_BUILD_TYPE=Release \
+  #                        -DGGML_OPENMP=OFF \
+  #                        -DLLAMA_BUILD_EXAMPLES=ON \
+  #                        -DLLAMA_BUILD_TOOLS=ON \
+  #                        -DLLAMA_BUILD_TESTS=OFF \
+  #                        -DCMAKE_SYSTEM_NAME=Linux \
+  #                        -DCMAKE_SYSTEM_PROCESSOR=riscv64 \
+  #                        -DCMAKE_C_COMPILER=riscv64-linux-gnu-gcc-14 \
+  #                        -DCMAKE_CXX_COMPILER=riscv64-linux-gnu-g++-14 \
+  #                        -DCMAKE_POSITION_INDEPENDENT_CODE=ON \
+  #                        -DCMAKE_FIND_ROOT_PATH=/usr/lib/riscv64-linux-gnu \
+  #                        -DCMAKE_FIND_ROOT_PATH_MODE_PROGRAM=NEVER \
+  #                        -DCMAKE_FIND_ROOT_PATH_MODE_LIBRARY=ONLY \
+  #                        -DCMAKE_FIND_ROOT_PATH_MODE_INCLUDE=BOTH
 
-          cmake --build build --config Release -j $(nproc)
+  #         cmake --build build --config Release -j $(nproc)
 
   # ubuntu-24-riscv64-vulkan-cross:
   #   runs-on: ubuntu-24.04

.github/workflows/release.yml

@@ -134,8 +134,8 @@ jobs:
         include:
           - build: 'x64'
             os: ubuntu-22.04
-          - build: 's390x-z15' # z15 because our CI runners are on z15
-            os: ubuntu-22.04-s390x
+          - build: 's390x'
+            os: ubuntu-24.04-s390x
           # GGML_BACKEND_DL and GGML_CPU_ALL_VARIANTS are not currently supported on arm
           # - build: 'arm64'
           #   os: ubuntu-22.04-arm

CODEOWNERS

@@ -89,6 +89,7 @@
 /src/llama-model-loader.*               @slaren
 /src/llama-model.*                      @CISC
 /src/llama-vocab.*                      @CISC
+/src/models/                            @CISC
 /tests/                                 @ggerganov
 /tests/test-backend-ops.cpp             @slaren
 /tests/test-thread-safety.cpp           @slaren

README.md

@@ -17,14 +17,13 @@ LLM inference in C/C++
 
 ## Hot topics
 
-- **[guide : running gpt-oss with llama.cpp](https://github.com/ggml-org/llama.cpp/discussions/15396)**
-- **[[FEEDBACK] Better packaging for llama.cpp to support downstream consumers 🤗](https://github.com/ggml-org/llama.cpp/discussions/15313)**
+- **[guide : using the new WebUI of llama.cpp](https://github.com/ggml-org/llama.cpp/discussions/16938)**
+- [guide : running gpt-oss with llama.cpp](https://github.com/ggml-org/llama.cpp/discussions/15396)
+- [[FEEDBACK] Better packaging for llama.cpp to support downstream consumers 🤗](https://github.com/ggml-org/llama.cpp/discussions/15313)
 - Support for the `gpt-oss` model with native MXFP4 format has been added | [PR](https://github.com/ggml-org/llama.cpp/pull/15091) | [Collaboration with NVIDIA](https://blogs.nvidia.com/blog/rtx-ai-garage-openai-oss) | [Comment](https://github.com/ggml-org/llama.cpp/discussions/15095)
-- Hot PRs: [All](https://github.com/ggml-org/llama.cpp/pulls?q=is%3Apr+label%3Ahot+) | [Open](https://github.com/ggml-org/llama.cpp/pulls?q=is%3Apr+label%3Ahot+is%3Aopen)
 - Multimodal support arrived in `llama-server`: [#12898](https://github.com/ggml-org/llama.cpp/pull/12898) | [documentation](./docs/multimodal.md)
 - VS Code extension for FIM completions: https://github.com/ggml-org/llama.vscode
 - Vim/Neovim plugin for FIM completions: https://github.com/ggml-org/llama.vim
-- Introducing GGUF-my-LoRA https://github.com/ggml-org/llama.cpp/discussions/10123
 - Hugging Face Inference Endpoints now support GGUF out of the box! https://github.com/ggml-org/llama.cpp/discussions/9669
 - Hugging Face GGUF editor: [discussion](https://github.com/ggml-org/llama.cpp/discussions/9268) | [tool](https://huggingface.co/spaces/CISCai/gguf-editor)
 

common/arg.cpp

@@ -2030,7 +2030,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
                 params.system_prompt.pop_back();
             }
         }
-    ).set_examples({LLAMA_EXAMPLE_MAIN}));
+    ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_DIFFUSION}));
     add_opt(common_arg(
         {"--in-file"}, "FNAME",
         "an input file (repeat to specify multiple files)",
@@ -2768,6 +2768,20 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
             params.image.emplace_back(value);
         }
     ).set_examples({LLAMA_EXAMPLE_MTMD}));
+    add_opt(common_arg(
+        {"--image-min-tokens"}, "N",
+        "minimum number of tokens each image can take, only used by vision models with dynamic resolution (default: read from model)",
+        [](common_params & params, int value) {
+            params.image_min_tokens = value;
+        }
+    ).set_examples(mmproj_examples).set_env("LLAMA_ARG_IMAGE_MIN_TOKENS"));
+    add_opt(common_arg(
+        {"--image-max-tokens"}, "N",
+        "maximum number of tokens each image can take, only used by vision models with dynamic resolution (default: read from model)",
+        [](common_params & params, int value) {
+            params.image_max_tokens = value;
+        }
+    ).set_examples(mmproj_examples).set_env("LLAMA_ARG_IMAGE_MAX_TOKENS"));
     if (llama_supports_rpc()) {
         add_opt(common_arg(
             {"--rpc"}, "SERVERS",

common/chat.cpp

@@ -313,7 +313,6 @@ json common_chat_msgs_to_json_oaicompat(const std::vector<common_chat_msg> & msg
         }
         if (!msg.reasoning_content.empty()) {
             jmsg["reasoning_content"] = msg.reasoning_content;
-            jmsg["thinking"] = msg.reasoning_content; // gpt-oss
         }
         if (!msg.tool_name.empty()) {
             jmsg["name"] = msg.tool_name;
@@ -1810,7 +1809,23 @@ static void common_chat_parse_deepseek_v3_1(common_chat_msg_parser & builder) {
 
 static common_chat_params common_chat_params_init_gpt_oss(const common_chat_template & tmpl, const struct templates_params & inputs) {
     common_chat_params data;
-    auto prompt = apply(tmpl, inputs);
+
+    // Copy reasoning to the "thinking" field as expected by the gpt-oss template
+    auto adjusted_messages = json::array();
+    for (const auto & msg : inputs.messages) {
+        auto has_reasoning_content = msg.contains("reasoning_content") && msg.at("reasoning_content").is_string();
+        auto has_tool_calls = msg.contains("tool_calls") && msg.at("tool_calls").is_array();
+
+        if (has_reasoning_content && has_tool_calls) {
+            auto adjusted_message = msg;
+            adjusted_message["thinking"] = msg.at("reasoning_content");
+            adjusted_messages.push_back(adjusted_message);
+        } else {
+            adjusted_messages.push_back(msg);
+        }
+    }
+
+    auto prompt = apply(tmpl, inputs, /* messages_override= */ adjusted_messages);
 
     // Check if we need to replace the return token with end token during
     // inference and without generation prompt. For more details see:

common/common.h

@@ -406,6 +406,8 @@ struct common_params {
     bool mmproj_use_gpu = true;     // use GPU for multimodal model
     bool no_mmproj = false;         // explicitly disable multimodal model
     std::vector<std::string> image; // path to image file(s)
+    int image_min_tokens = -1;
+    int image_max_tokens = -1;
 
     // finetune
     struct lr_opt lr;
@@ -505,6 +507,10 @@ struct common_params {
     // return false from callback to abort model loading or true to continue
     llama_progress_callback load_progress_callback = NULL;
     void *                  load_progress_callback_user_data = NULL;
+
+    bool has_speculative() const {
+        return !speculative.model.path.empty() || !speculative.model.hf_repo.empty();
+    }
 };
 
 // call once at the start of a program if it uses libcommon

convert_hf_to_gguf.py

@@ -7187,6 +7187,42 @@ class MiniMaxM2Model(TextModel):
         return super().modify_tensors(data_torch, name, bid)
 
 
+@ModelBase.register("PanguEmbeddedForCausalLM")
+class PanguEmbeddedModel(TextModel):
+    model_arch = gguf.MODEL_ARCH.PANGU_EMBED
+
+    def set_vocab(self):
+        self._set_vocab_sentencepiece()
+
+        tokenizer_config_file = self.dir_model / 'tokenizer_config.json'
+        if tokenizer_config_file.is_file():
+            with open(tokenizer_config_file, "r", encoding="utf-8") as f:
+                tokenizer_config_json = json.load(f)
+                if "add_prefix_space" in tokenizer_config_json:
+                    self.gguf_writer.add_add_space_prefix(tokenizer_config_json["add_prefix_space"])
+
+    def set_gguf_parameters(self):
+        super().set_gguf_parameters()
+        hparams = self.hparams
+        self.gguf_writer.add_vocab_size(hparams["vocab_size"])
+
+        # PanguEmbedded's hparam loaded from config.json without head_dim
+        if (rope_dim := hparams.get("head_dim")) is None:
+            rope_dim = hparams["hidden_size"] // hparams["num_attention_heads"]
+        self.gguf_writer.add_rope_dimension_count(rope_dim)
+
+        if hparams.get("head_dim") is None:
+            self.gguf_writer.add_key_length(rope_dim)
+            self.gguf_writer.add_value_length(rope_dim)
+
+    def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
+        if name == "lm_head.weight":
+            if self.hparams.get("tie_word_embeddings", False):
+                logger.info("Skipping tied output layer 'lm_head.weight'")
+                return []
+        return [(self.map_tensor_name(name), data_torch)]
+
+
 @ModelBase.register("Dots1ForCausalLM")
 class Dots1Model(Qwen2MoeModel):
     model_arch = gguf.MODEL_ARCH.DOTS1
@@ -9802,6 +9838,113 @@ class CogVLMModel(LlamaModel):
 
         return [(self.map_tensor_name(name), data_torch)]
 
+
+@ModelBase.register("JanusForConditionalGeneration")
+class JanusProModel(LlamaModel):
+    model_arch = gguf.MODEL_ARCH.LLAMA  # reuse Llama arch
+
+    def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
+        # Skip vision, aligner, and generation tensors
+        skip_prefixes = (
+            'model.vision_model.',
+            'model.aligner.',
+            'model.vqmodel.',
+            'model.generation_embeddings.',
+            'model.generation_aligner.',
+            'model.generation_head.',
+        )
+        if name.startswith(skip_prefixes):
+            return []
+
+        if name.startswith('model.language_model.'):
+            name = name.replace('model.language_model.', 'model.')
+        elif name.startswith('language_model.'):
+            name = name.replace('language_model.', '')
+
+        return super().modify_tensors(data_torch, name, bid)
+
+
+@ModelBase.register("JanusForConditionalGeneration")
+class JanusProVisionModel(MmprojModel):
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        assert self.hparams_vision is not None
+        if "intermediate_size" not in self.hparams_vision:
+            mlp_ratio = self.hparams_vision.get("mlp_ratio")
+            hidden_size = self.hparams_vision.get("hidden_size")
+            if mlp_ratio is not None and hidden_size is not None:
+                self.hparams_vision["intermediate_size"] = int(round(hidden_size * mlp_ratio))
+
+    def set_gguf_parameters(self):
+        super().set_gguf_parameters()
+        assert self.hparams_vision is not None
+
+        self.gguf_writer.add_clip_projector_type(gguf.VisionProjectorType.JANUS_PRO)
+
+        self.gguf_writer.add_vision_attention_layernorm_eps(self.hparams_vision.get("layer_norm_eps", 1e-6))
+
+        hidden_act = str(self.hparams_vision.get("hidden_act", "")).lower()
+        if hidden_act == "gelu":
+            self.gguf_writer.add_vision_use_gelu(True)
+        elif hidden_act == "silu":
+            self.gguf_writer.add_vision_use_silu(True)
+
+    def _map_aligner_tensor(self, data_torch: Tensor, name: str) -> Iterable[tuple[str, Tensor]]:
+        """Map aligner tensors to projector format"""
+        suffix = ".bias" if name.endswith(".bias") else ".weight"
+
+        if name.startswith("model.aligner."):
+            local_name = name[len("model.aligner."):]
+        elif name.startswith("aligner."):
+            local_name = name[len("aligner."):]
+        else:
+            raise ValueError(f"Unsupported Janus aligner prefix: {name}")
+
+        if local_name.startswith("fc1."):
+            mm_index = 0
+        elif local_name.startswith("hidden_layers."):
+            parts = local_name.split(".", 2)
+            if len(parts) < 3:
+                raise ValueError(f"Unexpected Janus aligner tensor name: {name}")
+            mm_index = int(parts[1]) + 1
+        else:
+            raise ValueError(f"Unsupported Janus aligner tensor: {name}")
+
+        tensor_name = self.format_tensor_name(gguf.MODEL_TENSOR.V_MMPROJ, mm_index, suffix=suffix)
+        return [(tensor_name, data_torch)]
+
+    def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
+        del bid  # unused
+
+        # Skip language model tensors as they will be handled by `JanusProModel`
+        if name.startswith(('model.language_model.', 'language_model.')):
+            return []
+
+        # Skip generation-related components
+        skip_generation_prefixes = (
+            'model.vqmodel.',
+            'vqmodel.',
+            'model.generation_embeddings.',
+            'generation_embeddings.',
+            'model.generation_aligner.',
+            'generation_aligner.',
+            'model.generation_head.',
+            'generation_head.',
+        )
+        if name.startswith(skip_generation_prefixes):
+            return []
+
+        # Handle aligner tensors
+        if name.startswith(('model.aligner.', 'aligner.')):
+            return list(self._map_aligner_tensor(data_torch, name))
+
+        # Handle vision tensors
+        if name.startswith(('model.vision_model.', 'vision_model.')):
+            return [(self.map_tensor_name(name), data_torch)]
+
+        return []
+
+
 ###### CONVERSION LOGIC ######
 
 

docs/backend/OPENCL.md

@@ -39,18 +39,23 @@ The llama.cpp OpenCL backend is designed to enable llama.cpp on **Qualcomm Adren
 | Adreno 830 (Snapdragon 8 Elite)      | Support |
 | Adreno X85 (Snapdragon X Elite)      | Support |
 
+> A6x GPUs with a recent driver and compiler are supported; they are usually found in IoT platforms.
+However, A6x GPUs in phones are likely not supported due to the outdated driver and compiler.
+
 ## DataType Supports
 
 | DataType               | Status                     |
 |:----------------------:|:--------------------------:|
 | Q4_0                   | Support                    |
 | Q6_K                   | Support, but not optimized |
+| Q8_0                   | Support                    |
+| MXFP4                  | Support                    |
 
 ## Model Preparation
 
-You can refer to the general [*Prepare and Quantize*](README.md#prepare-and-quantize) guide for model prepration.
+You can refer to the general [llama-quantize tool](/tools/quantize/README.md) for steps to convert a model in Hugging Face safetensor format to GGUF with quantization.
 
-Currently we support `Q4_0` quantization and have optimize for it. To achieve best performance on Adreno GPU, add `--pure` to `llama-quantize`. For example,
+Currently we support `Q4_0` quantization and have optimized for it. To achieve best performance on Adreno GPU, add `--pure` to `llama-quantize` (i.e., make all weights in `Q4_0`). For example,
 
 ```sh
 ./llama-quantize --pure ggml-model-qwen2.5-3b-f16.gguf ggml-model-qwen-3b-Q4_0.gguf Q4_0
@@ -58,6 +63,17 @@ Currently we support `Q4_0` quantization and have optimize for it. To achieve be
 
 Since `Q6_K` is also supported, `Q4_0` quantization without `--pure` will also work. However, the performance will be worse compared to pure `Q4_0` quantization.
 
+### `MXFP4` MoE Models
+
+OpenAI gpt-oss models are MoE models in `MXFP4`. The quantized model will be in `MXFP4_MOE`, a mixture of `MXFP4` and `Q8_0`.
+For this quantization, there is no need to specify `--pure`.
+For gpt-oss-20b model, you can directly [download](https://huggingface.co/ggml-org/gpt-oss-20b-GGUF) the quantized GGUF file in `MXFP4_MOE` from Hugging Face.
+
+Although it is possible to quantize gpt-oss-20b model in pure `Q4_0` (all weights in `Q4_0`), it is not recommended since `MXFP4` has been optimized for MoE while `Q4_0` is not. In addition, accuracy should degrade with such pure `Q4_0` quantization.
+Hence, using the default `MXFP4_MOE` quantization (see the link above) is recommended for this model.
+
+> Note that the `Q4_0` model found [here](https://huggingface.co/unsloth/gpt-oss-20b-GGUF/blob/main/gpt-oss-20b-Q4_0.gguf) is a mixture of `Q4_0`, `Q8_0` and `MXFP4` and gives better performance than `MXFP4_MOE` quantization.
+
 ## CMake Options
 
 The OpenCL backend has the following CMake options that control the behavior of the backend.
@@ -146,10 +162,13 @@ A Snapdragon X Elite device with Windows 11 Arm64 is used. Make sure the followi
 * Ninja
 * Visual Studio 2022
 * Powershell 7
+* Python
 
 Visual Studio provides necessary headers and libraries although it is not directly used for building.
 Alternatively, Visual Studio Build Tools can be installed instead of the full Visual Studio.
 
+> Note that building using Visual Studio's cl compiler is not supported. Clang must be used. Clang depends on libraries provided by Visual Studio to work. Therefore, Visual Studio must be installed. Alternatively, Visual Studio Build Tools can be installed instead of the full Visual Studio.
+
 Powershell 7 is used for the following commands.
 If an older version of Powershell is used, these commands may not work as they are.
 
@@ -201,9 +220,12 @@ ninja
 
 ## Known Issues
 
-- Currently OpenCL backend does not work on Adreno 6xx GPUs.
+- Flash attention does not always improve performance.
+- Currently OpenCL backend works on A6xx GPUs with recent drivers and compilers (usually found in IoT platforms).
+  However, it does not work on A6xx GPUs found in phones with old drivers and compilers.
 
 ## TODO
 
 - Optimization for Q6_K
 - Support and optimization for Q4_K
+- Improve flash attention

docs/docker.md

@@ -7,9 +7,9 @@
 ## Images
 We have three Docker images available for this project:
 
-1. `ghcr.io/ggml-org/llama.cpp:full`: This image includes both the main executable file and the tools to convert LLaMA models into ggml and convert into 4-bit quantization. (platforms: `linux/amd64`, `linux/arm64`)
-2. `ghcr.io/ggml-org/llama.cpp:light`: This image only includes the main executable file. (platforms: `linux/amd64`, `linux/arm64`)
-3. `ghcr.io/ggml-org/llama.cpp:server`: This image only includes the server executable file. (platforms: `linux/amd64`, `linux/arm64`)
+1. `ghcr.io/ggml-org/llama.cpp:full`: This image includes both the main executable file and the tools to convert LLaMA models into ggml and convert into 4-bit quantization. (platforms: `linux/amd64`, `linux/arm64`, `linux/s390x`)
+2. `ghcr.io/ggml-org/llama.cpp:light`: This image only includes the main executable file. (platforms: `linux/amd64`, `linux/arm64`, `linux/s390x`)
+3. `ghcr.io/ggml-org/llama.cpp:server`: This image only includes the server executable file. (platforms: `linux/amd64`, `linux/arm64`, `linux/s390x`)
 
 Additionally, there the following images, similar to the above:
 

docs/ops.md

@@ -22,11 +22,11 @@ Legend:
 |                           ARANGE | ❌ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ | ❌ | ❌ |
 |                           ARGMAX | ❌ | ✅ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ |
 |                          ARGSORT | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ |
-|                             CEIL | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ |
+|                             CEIL | ❌ | ❌ | ✅ | 🟡 | ❌ | ❌ | ✅ | ❌ | ❌ |
 |                            CLAMP | ❌ | ✅ | ✅ | ✅ | 🟡 | 🟡 | ✅ | 🟡 | ❌ |
 |                           CONCAT | ❌ | ✅ | ✅ | 🟡 | ✅ | 🟡 | 🟡 | ✅ | ❌ |
 |                             CONT | ❌ | 🟡 | ✅ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ❌ |
-|                          CONV_2D | ❌ | ❌ | ✅ | ❌ | ❌ | ✅ | ❌ | ✅ | ❌ |
+|                          CONV_2D | ❌ | ❌ | ✅ | 🟡 | ❌ | ✅ | ❌ | ✅ | ❌ |
 |                       CONV_2D_DW | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ | ✅ | ❌ |
 |                          CONV_3D | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
 |                CONV_TRANSPOSE_1D | ❌ | ✅ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ |
@@ -42,7 +42,7 @@ Legend:
 |                              ELU | ❌ | ✅ | ✅ | 🟡 | 🟡 | ❌ | 🟡 | ❌ | ❌ |
 |                              EXP | ❌ | ✅ | ✅ | 🟡 | 🟡 | ❌ | 🟡 | ❌ | ❌ |
 |                   FLASH_ATTN_EXT | ❌ | 🟡 | ✅ | 🟡 | 🟡 | ❌ | ❌ | 🟡 | ❌ |
-|                            FLOOR | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ |
+|                            FLOOR | ❌ | ❌ | ✅ | 🟡 | ❌ | ❌ | ✅ | ❌ | ❌ |
 |                GATED_LINEAR_ATTN | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ✅ | ❌ | ❌ |
 |                            GEGLU | ❌ | ✅ | ✅ | ✅ | 🟡 | ✅ | ✅ | 🟡 | ❌ |
 |                        GEGLU_ERF | ❌ | ✅ | ✅ | ✅ | 🟡 | ✅ | ✅ | 🟡 | ❌ |
@@ -84,7 +84,7 @@ Legend:
 |                             ROLL | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ |
 |                             ROPE | ❌ | 🟡 | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ |
 |                        ROPE_BACK | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ | ✅ | ❌ |
-|                            ROUND | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ |
+|                            ROUND | ❌ | ❌ | ✅ | 🟡 | ❌ | ❌ | ✅ | ❌ | ❌ |
 |                        RWKV_WKV6 | ❌ | ❌ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ |
 |                        RWKV_WKV7 | ❌ | ❌ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ |
 |                            SCALE | ❌ | 🟡 | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ |
@@ -111,6 +111,6 @@ Legend:
 |                             TANH | ❌ | ✅ | ✅ | 🟡 | 🟡 | ✅ | 🟡 | 🟡 | ❌ |
 |               TIMESTEP_EMBEDDING | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ |
 |                         TOPK_MOE | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ |
-|                            TRUNC | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ |
+|                            TRUNC | ❌ | ❌ | ✅ | 🟡 | ❌ | ❌ | ✅ | ❌ | ❌ |
 |                          UPSCALE | ❌ | 🟡 | ✅ | ✅ | 🟡 | ✅ | 🟡 | ✅ | ❌ |
 |                            XIELU | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |

examples/model-conversion/scripts/causal/run-org-model.py

@@ -138,6 +138,9 @@ if model_path is None:
         "Model path must be specified either via --model-path argument or MODEL_PATH environment variable"
     )
 
+
+print("Loading model and tokenizer using AutoTokenizer:", model_path)
+tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
 config = AutoConfig.from_pretrained(model_path, trust_remote_code=True)
 
 print("Model type:       ", config.model_type)
@@ -147,10 +150,6 @@ print("Number of layers: ", config.num_hidden_layers)
 print("BOS token id:     ", config.bos_token_id)
 print("EOS token id:     ", config.eos_token_id)
 
-print("Loading model and tokenizer using AutoTokenizer:", model_path)
-tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
-config = AutoConfig.from_pretrained(model_path, trust_remote_code=True)
-
 if unreleased_model_name:
     model_name_lower = unreleased_model_name.lower()
     unreleased_module_path = (
@@ -171,7 +170,7 @@ if unreleased_model_name:
         exit(1)
 else:
     model = AutoModelForCausalLM.from_pretrained(
-        model_path, device_map="auto", offload_folder="offload", trust_remote_code=True
+        model_path, device_map="auto", offload_folder="offload", trust_remote_code=True, config=config
     )
 
 for name, module in model.named_modules():

ggml/include/ggml.h

@@ -2108,6 +2108,7 @@ extern "C" {
     enum ggml_scale_mode {
         GGML_SCALE_MODE_NEAREST  = 0,
         GGML_SCALE_MODE_BILINEAR = 1,
+        GGML_SCALE_MODE_BICUBIC  = 2,
 
         GGML_SCALE_MODE_COUNT
     };

ggml/src/CMakeLists.txt

@@ -308,6 +308,10 @@ function(ggml_add_cpu_backend_variant tag_name)
             set(GGML_INTERNAL_${feat} ON)
         endforeach()
     elseif (GGML_SYSTEM_ARCH STREQUAL "s390x")
+        foreach (feat VXE2 NNPA)
+            set(GGML_INTERNAL_${feat} OFF)
+        endforeach()
+
         foreach (feat ${ARGN})
             set(GGML_INTERNAL_${feat} ON)
         endforeach()
@@ -377,9 +381,8 @@ if (GGML_CPU_ALL_VARIANTS)
         endif()
     elseif (GGML_SYSTEM_ARCH STREQUAL "s390x")
         if (CMAKE_SYSTEM_NAME MATCHES "Linux")
-            ggml_add_cpu_backend_variant(s390x_z15  Z15 VXE)
-            # ggml_add_cpu_backend_variant(s390x_z16  Z16 VXE)
-            # ggml_add_cpu_backend_variant(s390x_z17  Z17 VXE)
+            ggml_add_cpu_backend_variant(z15    Z15 VXE2)
+            ggml_add_cpu_backend_variant(z16    Z16 VXE2 NNPA)
         else()
             message(FATAL_ERROR "Unsupported s390x target OS: ${CMAKE_SYSTEM_NAME}")
         endif()

ggml/src/ggml-cpu/CMakeLists.txt

@@ -504,11 +504,18 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
             endforeach()
         endif()
 
-        if (GGML_VXE OR GGML_INTERNAL_VXE)
-            message(STATUS "VX/VXE/VXE2 enabled")
+        if (GGML_VXE OR GGML_INTERNAL_VXE2)
+            message(STATUS "VXE2 enabled")
             list(APPEND ARCH_FLAGS -mvx -mzvector)
-            list(APPEND ARCH_DEFINITIONS GGML_VXE)
+            list(APPEND ARCH_DEFINITIONS GGML_USE_VXE2)
         endif()
+
+        if (GGML_INTERNAL_NNPA)
+            message(STATUS "NNPA enabled")
+            list(APPEND ARCH_DEFINITIONS GGML_USE_NNPA)
+        endif()
+
+        ggml_add_cpu_backend_features(${GGML_CPU_NAME} s390 ${ARCH_DEFINITIONS})
     elseif (CMAKE_SYSTEM_PROCESSOR MATCHES "wasm")
         message(STATUS "Wasm detected")
         list (APPEND GGML_CPU_SOURCES ggml-cpu/arch/wasm/quants.c)

ggml/src/ggml-cpu/arch/loongarch/quants.c

@@ -700,7 +700,8 @@ void ggml_vec_dot_q4_0_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const voi
     for (; ib + 1 < nb; ib += 2) {
 
         // Compute combined scale for the block 0 and 1
-        const __m128 d_0_1 = (__m128)__lsx_vreplgr2vr_w( GGML_CPU_FP16_TO_FP32(x[ib].d) * GGML_CPU_FP16_TO_FP32(y[ib].d) );
+        const float ft0 = GGML_CPU_FP16_TO_FP32(x[ib].d) * GGML_CPU_FP16_TO_FP32(y[ib].d);
+        const __m128 d_0_1 = (__m128)(v4f32){ft0, ft0, ft0, ft0};
 
         const __m128i tmp_0_1 = __lsx_vld((const __m128i *)x[ib].qs, 0);
 
@@ -714,11 +715,9 @@ void ggml_vec_dot_q4_0_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const voi
         bx_1 = __lsx_vsub_b(bx_1, off);
         const __m128i i32_1 = mul_sum_i8_pairs(bx_1, by_1);
 
-        //_mm_prefetch(&x[ib] + 2 * sizeof(block_q4_0), _MM_HINT_T0);
-        //_mm_prefetch(&y[ib] + 2 * sizeof(block_q8_0), _MM_HINT_T0);
-
         // Compute combined scale for the block 2 and 3
-        const __m128 d_2_3 = (__m128)__lsx_vreplgr2vr_w( GGML_CPU_FP16_TO_FP32(x[ib + 1].d) * GGML_CPU_FP16_TO_FP32(y[ib + 1].d) );
+        const float ft1 = GGML_CPU_FP16_TO_FP32(x[ib + 1].d) * GGML_CPU_FP16_TO_FP32(y[ib + 1].d);
+        const __m128 d_2_3 = (__m128)(v4f32){ft1, ft1, ft1, ft1};
 
         const __m128i tmp_2_3 = __lsx_vld((const __m128i *)x[ib + 1].qs, 0);
 

ggml/src/ggml-cpu/arch/s390/cpu-feats.cpp

@@ -0,0 +1,50 @@
+#include "ggml-backend-impl.h"
+
+#if defined(__s390x__)
+#include <sys/auxv.h>
+
+// find hwcap bits in asm/elf.h
+#ifndef HWCAP_VXRS_EXT2
+#define HWCAP_VXRS_EXT2 (1 << 15)
+#endif
+
+#ifndef HWCAP_NNPA
+#define HWCAP_NNPA (1 << 20)
+#endif
+
+struct s390x_features {
+    bool has_vxe2 = false;
+    bool has_nnpa = false;
+
+    s390x_features() {
+        uint32_t hwcap = getauxval(AT_HWCAP);
+        // NOTE: use hwcap2 with DFLT for z17 and later
+        // uint32_t hwcap2 = getauxval(AT_HWCAP2);
+
+        has_vxe2 = !!(hwcap & HWCAP_VXRS_EXT2);
+        has_nnpa = !!(hwcap & HWCAP_NNPA);
+    }
+};
+
+static int ggml_backend_cpu_s390x_score() {
+    int score = 1;
+    s390x_features sf;
+
+// IBM z15 / LinuxONE 3
+#ifdef GGML_USE_VXE2
+    if (!sf.has_vxe2) { return 0; }
+    score += 1 << 1;
+#endif
+
+// IBM z16 / LinuxONE 4 and z17 / LinuxONE 5
+#ifdef GGML_USE_NNPA
+    if (!sf.has_nnpa) { return 0; }
+    score += 1 << 2;
+#endif
+
+    return score;
+}
+
+GGML_BACKEND_DL_SCORE_IMPL(ggml_backend_cpu_s390x_score)
+
+#endif  // __s390x__

ggml/src/ggml-cpu/ggml-cpu-impl.h

@@ -500,13 +500,15 @@ inline static int32x4_t ggml_vec_dot(int32x4_t acc, int8x16_t a, int8x16_t b) {
 
 #endif
 
-#if defined(__loongarch_asx)
+#if defined(__loongarch_sx)
 /* float type data load instructions */
 static __m128 __lsx_vreplfr2vr_s(const float val) {
     v4f32 res = {val, val, val, val};
     return (__m128)res;
 }
+#endif
 
+#if defined(__loongarch_asx)
 static __m256 __lasx_xvreplfr2vr_s(const float val) {
     v8f32 res = {val, val, val, val, val, val, val, val};
     return (__m256)res;

ggml/src/ggml-cpu/ops.cpp

@@ -7084,7 +7084,11 @@ static void ggml_compute_forward_conv_2d_dw_cwhn(
     const int64_t row_end = MIN(row_start + rows_per_thread, rows_total);
 
 #ifdef GGML_SIMD
-    const int64_t pkg_size = GGML_F32_EPR;
+    #if defined(__ARM_FEATURE_SVE)
+        const int64_t pkg_size = svcntw();
+    #else
+        const int64_t pkg_size = GGML_F32_EPR;
+    #endif
     const int64_t pkg_count = c / pkg_size;
     const int64_t c_pkg_end = pkg_count * pkg_size;
 #else
@@ -7507,10 +7511,17 @@ static void ggml_compute_forward_upscale_f32(
     float sf1 = (float)ne1/src0->ne[1];
     float sf2 = (float)ne2/src0->ne[2];
     float sf3 = (float)ne3/src0->ne[3];
+    float pixel_offset = 0.5f;
 
     const int32_t mode_flags = ggml_get_op_params_i32(dst, 0);
     const ggml_scale_mode mode = (ggml_scale_mode) (mode_flags & 0xFF);
 
+    if (mode_flags & GGML_SCALE_FLAG_ALIGN_CORNERS) {
+        pixel_offset = 0.0f;
+        sf0 = ne0 > 1 && ne00 > 1 ? (float)(ne0 - 1) / (ne00 - 1) : sf0;
+        sf1 = ne1 > 1 && ne01 > 1 ? (float)(ne1 - 1) / (ne01 - 1) : sf1;
+    }
+
     if (mode == GGML_SCALE_MODE_NEAREST) {
         for (int64_t i3 = 0; i3 < ne3; i3++) {
             const int64_t i03 = i3 / sf3;
@@ -7530,13 +7541,6 @@ static void ggml_compute_forward_upscale_f32(
             }
         }
     } else if (mode == GGML_SCALE_MODE_BILINEAR) {
-        float pixel_offset = 0.5f;
-        if (mode_flags & GGML_SCALE_FLAG_ALIGN_CORNERS) {
-            pixel_offset = 0.0f;
-            sf0 = ne0 > 1 && ne00 > 1 ? (float)(ne0 - 1) / (ne00 - 1) : sf0;
-            sf1 = ne1 > 1 && ne01 > 1 ? (float)(ne1 - 1) / (ne01 - 1) : sf1;
-        }
-
         for (int64_t i3 = 0; i3 < ne3; i3++) {
             const int64_t i03 = i3 / sf3;
             for (int64_t i2 = ith; i2 < ne2; i2 += nth) {
@@ -7571,6 +7575,51 @@ static void ggml_compute_forward_upscale_f32(
 
                         const float val = a*(1 - dx)*(1 - dy) + b*dx*(1 - dy) + c*(1 - dx)*dy + d*dx*dy;
 
+                        float * y_dst = (float *)((char *)dst->data + i0*nb0 + i1*nb1 + i2*nb2 + i3*nb3);
+                        *y_dst = val;
+                    }
+                }
+            }
+        }
+    } else if (mode == GGML_SCALE_MODE_BICUBIC) {
+        // https://en.wikipedia.org/wiki/Bicubic_interpolation#Bicubic_convolution_algorithm
+        const float a = -0.75f; // use alpha = -0.75 (same as PyTorch)
+        auto weight1 = [a](float x) { return ((a + 2) * x - (a + 3)) * x * x + 1; };
+        auto weight2 = [a](float x) { return ((a * x - 5 * a) * x + 8 * a) * x - 4 * a; };
+        auto bicubic = [=](float p0, float p1, float p2, float p3, float x) {
+            const float w0 = weight2(x + 1);
+            const float w1 = weight1(x + 0);
+            const float w2 = weight1(1 - x);
+            const float w3 = weight2(2 - x);
+            return p0*w0 + p1*w1 + p2*w2 + p3*w3;
+        };
+
+        for (int64_t i3 = 0; i3 < ne3; i3++) {
+            const int64_t i03 = i3 / sf3;
+            for (int64_t i2 = ith; i2 < ne2; i2 += nth) {
+                const int64_t i02 = i2 / sf2;
+                for (int64_t i1 = 0; i1 < ne1; i1++) {
+                    const float y = ((float)i1 + pixel_offset) / sf1 - pixel_offset;
+                    const int64_t y0 = (int64_t)floorf(y);
+                    const float dy = y - (float)y0;
+
+                    for (int64_t i0 = 0; i0 < ne0; i0++) {
+                        const float x = ((float)i0 + pixel_offset) / sf0 - pixel_offset;
+                        const int64_t x0 = (int64_t)floorf(x);
+                        const float dx = x - (float)x0;
+
+                        auto p = [=](int64_t x_off, int64_t y_off) -> float {
+                            int64_t i00 = std::max(int64_t(0), std::min(x0 + x_off, ne00 - 1));
+                            int64_t i01 = std::max(int64_t(0), std::min(y0 + y_off, ne01 - 1));
+                            return *(const float *)((const char *)src0->data + i00*nb00 + i01*nb01 + i02*nb02 + i03*nb03);
+                        };
+
+                        const float val = bicubic(
+                            bicubic(p(-1,-1), p(0,-1), p(1,-1), p(2,-1), dx),
+                            bicubic(p(-1, 0), p(0, 0), p(1, 0), p(2, 0), dx),
+                            bicubic(p(-1, 1), p(0, 1), p(1, 1), p(2, 1), dx),
+                            bicubic(p(-1, 2), p(0, 2), p(1, 2), p(2, 2), dx), dy);
+
                         float * y_dst = (float *)((char *)dst->data + i0*nb0 + i1*nb1 + i2*nb2 + i3*nb3);
                         *y_dst = val;
                     }

ggml/src/ggml-cpu/repack.cpp

@@ -1678,10 +1678,24 @@ template <typename BLOC_TYPE, int64_t INTER_SIZE, int64_t NB_COLS, ggml_type PAR
         int64_t chunk_size = (nr + nth_scaled - 1) / nth_scaled;
         int64_t nchunk     = (nr + chunk_size - 1) / chunk_size;
 
+        // Ensure minimum chunk size to avoid alignment issues with high thread counts
+        // Minimum chunk size should be at least NB_COLS to prevent overlapping chunks after alignment
+        const int64_t min_chunk_size = NB_COLS;
+        if (nchunk > 0 && (nr / nchunk) < min_chunk_size && nr >= min_chunk_size) {
+            nchunk = (nr + min_chunk_size - 1) / min_chunk_size;
+        }
+
         if (nth == 1 || nchunk < nth || disable_chunking) {
             nchunk = nth;
         }
 
+        // Ensure nchunk doesn't exceed the number of rows divided by minimum chunk size
+        // This prevents creating too many tiny chunks that could overlap after alignment
+        const int64_t max_nchunk = (nr + min_chunk_size - 1) / min_chunk_size;
+        if (nchunk > max_nchunk) {
+            nchunk = max_nchunk;
+        }
+
         if (ith == 0) {
             // Every thread starts at ith, so the first unprocessed chunk is nth.  This save a bit of coordination right at the start.
             ggml_threadpool_chunk_set(params->threadpool, nth);
@@ -1695,8 +1709,15 @@ template <typename BLOC_TYPE, int64_t INTER_SIZE, int64_t NB_COLS, ggml_type PAR
         while (current_chunk < nchunk) {
             int64_t src0_start = (current_chunk * ne01) / nchunk;
             int64_t src0_end   = ((current_chunk + 1) * ne01) / nchunk;
+
+            // Align boundaries to NB_COLS - round up to ensure all data is included
+            // The chunk size limiting above ensures chunks are large enough to prevent overlaps
             src0_start = (src0_start % NB_COLS) ? src0_start + NB_COLS - (src0_start % NB_COLS) : src0_start;
             src0_end   = (src0_end   % NB_COLS) ? src0_end   + NB_COLS - (src0_end   % NB_COLS) : src0_end;
+            if (src0_end > ne01) {
+                src0_end = ne01;
+            }
+
             if (src0_start >= src0_end) {
                 break;
             }
@@ -1808,8 +1829,12 @@ template <typename BLOC_TYPE, int64_t INTER_SIZE, int64_t NB_COLS, ggml_type PAR
             int64_t src0_cur_start = (ith * ne01) / nth;
             int64_t src0_cur_end   = ((ith + 1) * ne01) / nth;
 
+            // Align boundaries to NB_COLS - round up to ensure all data is included
             src0_cur_start = (src0_cur_start % NB_COLS) ? src0_cur_start + NB_COLS - (src0_cur_start % NB_COLS) : src0_cur_start;
             src0_cur_end   = (src0_cur_end   % NB_COLS) ? src0_cur_end   + NB_COLS - (src0_cur_end   % NB_COLS) : src0_cur_end;
+            if (src0_cur_end > ne01) {
+                src0_cur_end = ne01;
+            }
 
             if (src0_cur_start >= src0_cur_end) {
                 return;

ggml/src/ggml-cpu/simd-mappings.h

@@ -956,7 +956,7 @@ do {                                                              \
 
 #define GGML_F32Cx8          __m256
 #define GGML_F32Cx8_ZERO    (__m256)__lasx_xvldi(0)
-#define GGML_F32Cx8_SET1(x) (__m256)__lasx_xvreplgr2vr_w((x))
+#define GGML_F32Cx8_SET1(x) (__m256)__lasx_xvreplfr2vr_s((x))
 
 static inline __m256 __lasx_f32cx8_load(const ggml_fp16_t * x) {
     __m256i a;
@@ -999,34 +999,34 @@ static inline void __lasx_f32cx8_store(ggml_fp16_t * x, __m256 y) {
 
 #define GGML_F32x4         __m128
 #define GGML_F32x4_ZERO    (__m128)__lsx_vldi(0)
-#define GGML_F32x4_SET1(x) (__m128)__lsx_vinsgr2vr_w(__lsx_vldi(0),(x), 0)
+#define GGML_F32x4_SET1(x) (__m128)__lsx_vreplfr2vr_s((x))
 #define GGML_F32x4_LOAD(x) (__m128)__lsx_vld((x), 0)
 #define GGML_F32x4_STORE(x, y)   __lsx_vst(y, x, 0)
 #define GGML_F32x4_FMA(a, b, c) __lsx_vfmadd_s(b, c, a)
 #define GGML_F32x4_ADD     __lsx_vfadd_s
 #define GGML_F32x4_MUL     __lsx_vfmul_s
-#define GGML_F32x4_REDUCE(res, x)                                                     \
-{                                                                                     \
-    int offset = GGML_F32_ARR >> 1;                                                   \
-    for (int i = 0; i < offset; ++i) {                                                \
-        x[i] = __lsx_vfadd_s(x[i], x[offset + i]);                                    \
-    }                                                                                 \
-    offset >>= 1;                                                                     \
-    for (int i = 0; i < offset; ++i) {                                                \
-        x[i] = __lsx_vfadd_s(x[i], x[offset + i]);                                    \
-    }                                                                                 \
-    offset >>= 1;                                                                     \
-    for (int i = 0; i < offset; ++i) {                                                \
-        x[i] = __lsx_vfadd_s(x[i], x[offset + i]);                                    \
-    }                                                                                 \
-    __m128i tmp     = __lsx_vsrli_d((__m128i) x[0], 32);                              \
-    tmp             = (__m128i) __lsx_vfadd_s((__m128) tmp, x[0]);                    \
-    tmp             = __lsx_vpickev_w(__lsx_vldi(0), tmp);                            \
-    const __m128 t0 = (__m128)__lsx_vshuf4i_w(tmp, 0x88);                                     \
-    tmp             = __lsx_vsrli_d((__m128i) t0, 32);                                \
-    tmp             = (__m128i) __lsx_vfadd_s((__m128) tmp, t0);                      \
-    tmp             = __lsx_vpickev_w(__lsx_vldi(0), tmp);                            \
-    res             = (ggml_float) __lsx_vpickve2gr_w(__lsx_vshuf4i_w(tmp, 0x88), 0); \
+
+#define GGML_F32x4_REDUCE(res, x)                               \
+{                                                               \
+    int offset = GGML_F32_ARR >> 1;                             \
+    for (int i = 0; i < offset; ++i) {                          \
+        x[i] = __lsx_vfadd_s(x[i], x[offset+i]);                \
+    }                                                           \
+    offset >>= 1;                                               \
+    for (int i = 0; i < offset; ++i) {                          \
+        x[i] = __lsx_vfadd_s(x[i], x[offset+i]);                \
+    }                                                           \
+    offset >>= 1;                                               \
+    for (int i = 0; i < offset; ++i) {                          \
+        x[i] = __lsx_vfadd_s(x[i], x[offset+i]);                \
+    }                                                           \
+    __m128i t0 = __lsx_vpickev_w((__m128i)x[0], (__m128i)x[0]); \
+    __m128i t1 = __lsx_vpickod_w((__m128i)x[0], (__m128i)x[0]); \
+    __m128 t2 = __lsx_vfadd_s((__m128)t0, (__m128)t1);          \
+    __m128i t3 = __lsx_vpickev_w((__m128i)t2, (__m128i)t2);     \
+    __m128i t4 = __lsx_vpickod_w((__m128i)t2, (__m128i)t2);     \
+    __m128 t5 = __lsx_vfadd_s((__m128)t3, (__m128)t4);          \
+    res = (ggml_float) ((v4f32)t5)[0];                          \
 }
 
 #define GGML_F32_VEC        GGML_F32x4
@@ -1068,7 +1068,7 @@ static inline void __lsx_f16x4_store(ggml_fp16_t * x, __m128 y) {
 
 #define GGML_F32Cx4             __m128
 #define GGML_F32Cx4_ZERO        (__m128)__lsx_vldi(0)
-#define GGML_F32Cx4_SET1(x)     (__m128)__lsx_vinsgr2vr_w(__lsx_vldi(0),(x), 0)
+#define GGML_F32Cx4_SET1(x)     (__m128)__lsx_vreplfr2vr_s((x))
 #define GGML_F32Cx4_LOAD(x)     (__m128)__lsx_f16x4_load(x)
 #define GGML_F32Cx4_STORE(x, y) __lsx_f16x4_store(x, y)
 #define GGML_F32Cx4_FMA         GGML_F32x4_FMA

ggml/src/ggml-cuda/fattn-tile.cu

@@ -14,6 +14,10 @@ void ggml_cuda_flash_attn_ext_tile(ggml_backend_cuda_context & ctx, ggml_tensor
             GGML_ASSERT(V->ne[0] == K->ne[0]);
             ggml_cuda_flash_attn_ext_tile_case< 64,  64>(ctx, dst);
         } break;
+        case  72: {
+            GGML_ASSERT(V->ne[0] == K->ne[0]);
+            ggml_cuda_flash_attn_ext_tile_case< 72,  72>(ctx, dst);
+        } break;
         case  80: {
             GGML_ASSERT(V->ne[0] == K->ne[0]);
             ggml_cuda_flash_attn_ext_tile_case< 80,  80>(ctx, dst);

ggml/src/ggml-cuda/fattn-tile.cuh

@@ -6,7 +6,7 @@
 // nbatch_K == number of K columns to load in parallel for KQ calculation
 
 // TODO optimize kernel parameters for FP16 NVIDIA (P100)
-// TODO optimize kernel parameters for head sizes 40, 80, 96, 112
+// TODO optimize kernel parameters for head sizes 40, 72, 80, 96, 112
 
 // The ROCm compiler cannot handle templating in __launch_bounds__.
 // As a workaround, define a macro to package the kernel parameters as uint32_t:
@@ -32,6 +32,12 @@ static constexpr __host__ __device__ uint32_t ggml_cuda_fattn_tile_get_config_nv
     GGML_CUDA_FATTN_TILE_CONFIG_CASE( 64,  64, 16, 256, 2,  64,  64)
     GGML_CUDA_FATTN_TILE_CONFIG_CASE( 64,  64, 32, 256, 2,  64,  64)
 
+    GGML_CUDA_FATTN_TILE_CONFIG_CASE( 72,  72,  2,  64, 2,  64,  72)
+    GGML_CUDA_FATTN_TILE_CONFIG_CASE( 72,  72,  4, 128, 2,  64,  72)
+    GGML_CUDA_FATTN_TILE_CONFIG_CASE( 72,  72,  8, 256, 2,  64,  72)
+    GGML_CUDA_FATTN_TILE_CONFIG_CASE( 72,  72, 16, 256, 2,  64,  72)
+    GGML_CUDA_FATTN_TILE_CONFIG_CASE( 72,  72, 32, 256, 2,  64,  72)
+
     GGML_CUDA_FATTN_TILE_CONFIG_CASE( 80,  80,  2,  64, 2,  64,  40)
     GGML_CUDA_FATTN_TILE_CONFIG_CASE( 80,  80,  4, 128, 2,  64,  40)
     GGML_CUDA_FATTN_TILE_CONFIG_CASE( 80,  80,  8, 256, 2,  64,  40)
@@ -80,6 +86,12 @@ static constexpr __host__ __device__ uint32_t ggml_cuda_fattn_tile_get_config_nv
     GGML_CUDA_FATTN_TILE_CONFIG_CASE( 64,  64, 16, 128, 3,  64,  64)
     GGML_CUDA_FATTN_TILE_CONFIG_CASE( 64,  64, 32, 256, 2,  64,  64)
 
+    GGML_CUDA_FATTN_TILE_CONFIG_CASE( 72,  72,  2,  64, 2,  32,  72)
+    GGML_CUDA_FATTN_TILE_CONFIG_CASE( 72,  72,  4, 128, 2,  32,  72)
+    GGML_CUDA_FATTN_TILE_CONFIG_CASE( 72,  72,  8, 256, 2,  32,  72)
+    GGML_CUDA_FATTN_TILE_CONFIG_CASE( 72,  72, 16, 256, 2,  32,  72)
+    GGML_CUDA_FATTN_TILE_CONFIG_CASE( 72,  72, 32, 256, 2,  32,  72)
+
     GGML_CUDA_FATTN_TILE_CONFIG_CASE( 80,  80,  2,  64, 2,  32,  40)
     GGML_CUDA_FATTN_TILE_CONFIG_CASE( 80,  80,  4, 128, 2,  32,  40)
     GGML_CUDA_FATTN_TILE_CONFIG_CASE( 80,  80,  8, 256, 2,  32,  40)
@@ -130,6 +142,13 @@ static constexpr __host__ __device__ uint32_t ggml_cuda_fattn_tile_get_config_am
     GGML_CUDA_FATTN_TILE_CONFIG_CASE( 64,  64, 32, 256, 2,  64,  64)
     GGML_CUDA_FATTN_TILE_CONFIG_CASE( 64,  64, 64, 256, 2,  64,  64)
 
+    GGML_CUDA_FATTN_TILE_CONFIG_CASE( 72,  72,  2,  64, 2,  32,  72)
+    GGML_CUDA_FATTN_TILE_CONFIG_CASE( 72,  72,  4, 128, 2,  32,  72)
+    GGML_CUDA_FATTN_TILE_CONFIG_CASE( 72,  72,  8, 256, 2,  32,  72)
+    GGML_CUDA_FATTN_TILE_CONFIG_CASE( 72,  72, 16, 256, 2,  32,  72)
+    GGML_CUDA_FATTN_TILE_CONFIG_CASE( 72,  72, 32, 256, 2,  32,  72)
+    GGML_CUDA_FATTN_TILE_CONFIG_CASE( 72,  72, 64, 256, 2,  32,  72)
+
     GGML_CUDA_FATTN_TILE_CONFIG_CASE( 80,  80,  2,  64, 2,  32,  40)
     GGML_CUDA_FATTN_TILE_CONFIG_CASE( 80,  80,  4, 128, 2,  32,  40)
     GGML_CUDA_FATTN_TILE_CONFIG_CASE( 80,  80,  8, 256, 2,  32,  40)
@@ -185,6 +204,13 @@ static constexpr __host__ __device__ uint32_t ggml_cuda_fattn_tile_get_config_am
     GGML_CUDA_FATTN_TILE_CONFIG_CASE( 64,  64, 32, 128, 4,  64,  64)
     GGML_CUDA_FATTN_TILE_CONFIG_CASE( 64,  64, 64, 128, 5,  64,  64)
 
+    GGML_CUDA_FATTN_TILE_CONFIG_CASE( 72,  72,  2,  64, 2,  32,  72)
+    GGML_CUDA_FATTN_TILE_CONFIG_CASE( 72,  72,  4, 128, 2,  32,  72)
+    GGML_CUDA_FATTN_TILE_CONFIG_CASE( 72,  72,  8, 256, 2,  32,  72)
+    GGML_CUDA_FATTN_TILE_CONFIG_CASE( 72,  72, 16, 256, 2,  32,  72)
+    GGML_CUDA_FATTN_TILE_CONFIG_CASE( 72,  72, 32, 256, 2,  32,  72)
+    GGML_CUDA_FATTN_TILE_CONFIG_CASE( 72,  72, 64, 256, 2,  32,  72)
+
     GGML_CUDA_FATTN_TILE_CONFIG_CASE( 80,  80,  2,  64, 2,  32,  40)
     GGML_CUDA_FATTN_TILE_CONFIG_CASE( 80,  80,  4, 128, 2,  32,  40)
     GGML_CUDA_FATTN_TILE_CONFIG_CASE( 80,  80,  8, 256, 2,  32,  40)
@@ -723,7 +749,7 @@ static __global__ void flash_attn_tile(
 
     if (
 #ifdef GGML_USE_WMMA_FATTN
-            (ncols2 != 1 && DV != 40 && DV != 512) ||
+            (ncols2 != 1 && DV != 40 && DV != 72 && DV != 512) ||
 #endif // GGML_USE_WMMA_FATTN
             (use_logit_softcap && !(DV == 128 || DV == 256))
     ) {
@@ -1198,6 +1224,7 @@ void ggml_cuda_flash_attn_ext_tile(ggml_backend_cuda_context & ctx, ggml_tensor
 
 extern DECL_FATTN_TILE_CASE( 40,  40);
 extern DECL_FATTN_TILE_CASE( 64,  64);
+extern DECL_FATTN_TILE_CASE( 72,  72);
 extern DECL_FATTN_TILE_CASE( 80,  80);
 extern DECL_FATTN_TILE_CASE( 96,  96);
 extern DECL_FATTN_TILE_CASE(112, 112);

ggml/src/ggml-cuda/fattn.cu

@@ -223,6 +223,7 @@ static best_fattn_kernel ggml_cuda_get_best_fattn_kernel(const int device, const
     switch (K->ne[0]) {
         case  40:
         case  64:
+        case  72:
         case  80:
         case  96:
         case 128:
@@ -275,7 +276,7 @@ static best_fattn_kernel ggml_cuda_get_best_fattn_kernel(const int device, const
     const bool can_use_vector_kernel = Q->ne[0] <= 256 && Q->ne[0] % 64 == 0 && K->ne[1] % FATTN_KQ_STRIDE == 0;
 
     // If Turing tensor cores available, use them:
-    if (turing_mma_available(cc) && K->ne[1] % FATTN_KQ_STRIDE == 0 && Q->ne[0] != 40) {
+    if (turing_mma_available(cc) && K->ne[1] % FATTN_KQ_STRIDE == 0 && Q->ne[0] != 40 && Q->ne[0] != 72) {
         if (can_use_vector_kernel) {
             if (!ggml_is_quantized(K->type) && !ggml_is_quantized(V->type)) {
                 if (cc >= GGML_CUDA_CC_ADA_LOVELACE && Q->ne[1] == 1 && Q->ne[3] == 1 && !(gqa_ratio > 4 && K->ne[1] >= 8192)) {
@@ -301,7 +302,7 @@ static best_fattn_kernel ggml_cuda_get_best_fattn_kernel(const int device, const
     }
 
     // Use the WMMA kernel if possible:
-    if (ggml_cuda_should_use_wmma_fattn(cc) && K->ne[1] % FATTN_KQ_STRIDE == 0 && Q->ne[0] != 40 && Q->ne[0] != 576) {
+    if (ggml_cuda_should_use_wmma_fattn(cc) && K->ne[1] % FATTN_KQ_STRIDE == 0 && Q->ne[0] != 40 && Q->ne[0] != 72 && Q->ne[0] != 576) {
         if (can_use_vector_kernel && Q->ne[1] <= 2) {
             return BEST_FATTN_KERNEL_VEC;
         }

ggml/src/ggml-cuda/ggml-cuda.cu

@@ -2115,6 +2115,14 @@ static bool ggml_cuda_should_fuse_mul_mat_vec_f(const ggml_tensor * tensor) {
     const int cc      = ggml_cuda_info().devices[ggml_cuda_get_device()].cc;
     use_mul_mat_vec_f = use_mul_mat_vec_f && ggml_cuda_should_use_mmvf(src0->type, cc, src0->ne, is_mul_mat_id ? src1->ne[2] : src1->ne[1]);
 
+    const bool split = ggml_backend_buft_is_cuda_split(src0->buffer->buft) ||
+                       ggml_backend_buft_is_cuda_split(src1->buffer->buft);
+
+    //TODO: add support for fusion for split buffers
+    if (split) {
+        return false;
+    }
+
     //we only support fusion for ncols_dst = 1
     if (tensor->op == GGML_OP_MUL_MAT && dst->ne[1] != 1) {
         return false;
@@ -2154,6 +2162,15 @@ static bool ggml_cuda_should_fuse_mul_mat_vec_q(const ggml_tensor * tensor) {
         return false;
     }
 
+
+    const bool split = ggml_backend_buft_is_cuda_split(src0->buffer->buft) ||
+                       ggml_backend_buft_is_cuda_split(src1->buffer->buft);
+
+    //TODO: add support for fusion for split buffers
+    if (split) {
+        return false;
+    }
+
     return use_mul_mat_vec_q;
 }
 
@@ -2499,6 +2516,18 @@ static bool ggml_cuda_compute_forward(ggml_backend_cuda_context & ctx, struct gg
                 case GGML_UNARY_OP_XIELU:
                     ggml_cuda_op_xielu(ctx, dst);
                     break;
+                case GGML_UNARY_OP_FLOOR:
+                    ggml_cuda_op_floor(ctx, dst);
+                    break;
+                case GGML_UNARY_OP_CEIL:
+                    ggml_cuda_op_ceil(ctx, dst);
+                    break;
+                case GGML_UNARY_OP_ROUND:
+                    ggml_cuda_op_round(ctx, dst);
+                    break;
+                case GGML_UNARY_OP_TRUNC:
+                    ggml_cuda_op_trunc(ctx, dst);
+                    break;
                 default:
                     return false;
             }
@@ -3769,6 +3798,10 @@ static bool ggml_backend_cuda_device_supports_op(ggml_backend_dev_t dev, const g
                 case GGML_UNARY_OP_TANH:
                 case GGML_UNARY_OP_EXP:
                 case GGML_UNARY_OP_ELU:
+                case GGML_UNARY_OP_FLOOR:
+                case GGML_UNARY_OP_CEIL:
+                case GGML_UNARY_OP_ROUND:
+                case GGML_UNARY_OP_TRUNC:
                     return ggml_is_contiguous(op->src[0]);
                 default:
                     return false;

ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq72-dv72.cu

@@ -0,0 +1,5 @@
+// This file has been autogenerated by generate_cu_files.py, do not edit manually.
+
+#include "../fattn-tile.cuh"
+
+DECL_FATTN_TILE_CASE(72, 72);

ggml/src/ggml-cuda/template-instances/generate_cu_files.py

@@ -3,7 +3,7 @@
 from glob import glob
 import os
 
-HEAD_SIZES_KQ = [40, 64, 80, 96, 112, 128, 256, 576]
+HEAD_SIZES_KQ = [40, 64, 72, 80, 96, 112, 128, 256, 576]
 
 TYPES_KV = ["GGML_TYPE_F16", "GGML_TYPE_Q4_0", "GGML_TYPE_Q4_1", "GGML_TYPE_Q5_0", "GGML_TYPE_Q5_1", "GGML_TYPE_Q8_0"]
 
@@ -81,6 +81,8 @@ for ncols in [8, 16, 32, 64]:
             for head_size_kq in HEAD_SIZES_KQ:
                 if head_size_kq == 40:
                     continue
+                if head_size_kq == 72:
+                    continue
                 if head_size_kq != 576 and ncols2 == 16:
                     continue
                 if head_size_kq == 576 and ncols2 != 16:

ggml/src/ggml-cuda/unary.cu

@@ -85,6 +85,22 @@ static __device__ __forceinline__ float op_elu(float x) {
     return (x > 0.f) ? x : expm1f(x);
 }
 
+static __device__ __forceinline__ float op_floor(float x) {
+    return floorf(x);
+}
+
+static __device__ __forceinline__ float op_ceil(float x) {
+    return ceilf(x);
+}
+
+static __device__ __forceinline__ float op_round(float x) {
+    return round(x);
+}
+
+static __device__ __forceinline__ float op_trunc(float x) {
+    return trunc(x);
+}
+
 template <float (*op)(float), typename T>
 static __global__ void unary_op_kernel(const T * x, T * dst, const int k) {
     const int i = blockDim.x*blockIdx.x + threadIdx.x;
@@ -201,6 +217,22 @@ void ggml_cuda_op_log(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
 void ggml_cuda_op_elu(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
     ggml_cuda_op_unary<op_elu>(ctx, dst);
 }
+
+void ggml_cuda_op_floor(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
+    ggml_cuda_op_unary<op_floor>(ctx, dst);
+}
+
+void ggml_cuda_op_ceil(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
+    ggml_cuda_op_unary<op_ceil>(ctx, dst);
+}
+
+void ggml_cuda_op_round(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
+    ggml_cuda_op_unary<op_round>(ctx, dst);
+}
+
+void ggml_cuda_op_trunc(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
+    ggml_cuda_op_unary<op_trunc>(ctx, dst);
+}
 /* gated ops */
 
 template <float (*op)(float), typename T>

ggml/src/ggml-cuda/unary.cuh

@@ -63,6 +63,14 @@ void ggml_cuda_op_log(ggml_backend_cuda_context & ctx, ggml_tensor * dst);
 
 void ggml_cuda_op_elu(ggml_backend_cuda_context & ctx, ggml_tensor * dst);
 
+void ggml_cuda_op_floor(ggml_backend_cuda_context & ctx, ggml_tensor * dst);
+
+void ggml_cuda_op_ceil(ggml_backend_cuda_context & ctx, ggml_tensor * dst);
+
+void ggml_cuda_op_round(ggml_backend_cuda_context & ctx, ggml_tensor * dst);
+
+void ggml_cuda_op_trunc(ggml_backend_cuda_context & ctx, ggml_tensor * dst);
+
 void ggml_cuda_op_reglu(ggml_backend_cuda_context & ctx, ggml_tensor * dst);
 
 void ggml_cuda_op_geglu(ggml_backend_cuda_context & ctx, ggml_tensor * dst);

ggml/src/ggml-hexagon/htp/ops-utils.h

@@ -43,46 +43,46 @@ static inline int32_t htp_is_one_chunk(void * addr, uint32_t n, uint32_t chunk_s
 }
 
 static inline void htp_dump_int8_line(char * pref, const int8_t * x, int n) {
-    char str[1024], *p = str;
-    p += sprintf(p, "%s: ", pref);
-    for (int i = 0; i < 16; i++) {
-        p += sprintf(p, "%d, ", x[i]);
+    char str[1024], *p = str, *p_end = str + sizeof(str);
+    p += snprintf(p, p_end - p, "%s: ", pref);
+    for (int i = 0; i < n && p < p_end; i++) {
+        p += snprintf(p, p_end - p, "%d, ", x[i]);
     }
     FARF(HIGH, "%s\n", str);
 }
 
 static inline void htp_dump_uint8_line(char * pref, const uint8_t * x, uint32_t n) {
-    char str[1024], *p = str;
-    p += sprintf(p, "%s: ", pref);
-    for (int i = 0; i < n; i++) {
-        p += sprintf(p, "%d, ", x[i]);
+    char str[1024], *p = str, *p_end = str + sizeof(str);
+    p += snprintf(p, p_end - p, "%s: ", pref);
+    for (int i = 0; i < n && p < p_end; i++) {
+        p += snprintf(p, p_end - p, "%d, ", x[i]);
     }
     FARF(HIGH, "%s\n", str);
 }
 
 static inline void htp_dump_int32_line(char * pref, const int32_t * x, uint32_t n) {
-    char str[1024], *p = str;
-    p += sprintf(p, "%s: ", pref);
+    char str[1024], *p = str, *p_end = str + sizeof(str);
+    p += snprintf(p, p_end - p, "%s: ", pref);
     for (int i = 0; i < n; i++) {
-        p += sprintf(p, "%d, ", (int) x[i]);
+        p += snprintf(p, p_end - p, "%d, ", (int) x[i]);
     }
     FARF(HIGH, "%s\n", str);
 }
 
 static inline void htp_dump_fp16_line(char * pref, const __fp16 * x, uint32_t n) {
-    char str[1024], *p = str;
-    p += sprintf(p, "%s: ", pref);
+    char str[1024], *p = str, *p_end = str + sizeof(str);
+    p += snprintf(p, p_end - p, "%s: ", pref);
     for (int i = 0; i < n; i++) {
-        p += sprintf(p, "%.6f, ", (float) x[i]);
+        p += snprintf(p, p_end - p, "%.6f, ", (float) x[i]);
     }
     FARF(HIGH, "%s\n", str);
 }
 
 static inline void htp_dump_fp32_line(char * pref, const float * x, uint32_t n) {
-    char str[1024], *p = str;
-    p += sprintf(p, "%s: ", pref);
+    char str[1024], *p = str, *p_end = str + sizeof(str);
+    p += snprintf(p, p_end - p, "%s: ", pref);
     for (int i = 0; i < n; i++) {
-        p += sprintf(p, "%.6f, ", x[i]);
+        p += snprintf(p, p_end - p, "%.6f, ", x[i]);
     }
     FARF(HIGH, "%s\n", str);
 }

ggml/src/ggml-metal/ggml-metal-device.m

@@ -707,6 +707,7 @@ bool ggml_metal_device_supports_op(ggml_metal_device_t dev, const struct ggml_te
             if (op->src[0]->ne[0] != 32 &&
                 op->src[0]->ne[0] != 40 &&
                 op->src[0]->ne[0] != 64 &&
+                op->src[0]->ne[0] != 72 &&
                 op->src[0]->ne[0] != 80 &&
                 op->src[0]->ne[0] != 96 &&
                 op->src[0]->ne[0] != 112 &&

ggml/src/ggml-metal/ggml-metal.metal

@@ -5362,6 +5362,7 @@ typedef decltype(kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, hal
 template [[host_name("kernel_flash_attn_ext_f32_dk32_dv32"  )]]  kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4,   1, dequantize_f32,  float4x4,   1, dequantize_f32,  32,  32>;
 template [[host_name("kernel_flash_attn_ext_f32_dk40_dv40"  )]]  kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4,   1, dequantize_f32,  float4x4,   1, dequantize_f32,  40,  40>;
 template [[host_name("kernel_flash_attn_ext_f32_dk64_dv64"  )]]  kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4,   1, dequantize_f32,  float4x4,   1, dequantize_f32,  64,  64>;
+template [[host_name("kernel_flash_attn_ext_f32_dk72_dv72"  )]]  kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4,   1, dequantize_f32,  float4x4,   1, dequantize_f32,  72,  72>;
 template [[host_name("kernel_flash_attn_ext_f32_dk80_dv80"  )]]  kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4,   1, dequantize_f32,  float4x4,   1, dequantize_f32,  80,  80>;
 template [[host_name("kernel_flash_attn_ext_f32_dk96_dv96"  )]]  kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4,   1, dequantize_f32,  float4x4,   1, dequantize_f32,  96,  96>;
 template [[host_name("kernel_flash_attn_ext_f32_dk112_dv112")]]  kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4,   1, dequantize_f32,  float4x4,   1, dequantize_f32,  112, 112>;
@@ -5374,6 +5375,7 @@ template [[host_name("kernel_flash_attn_ext_f32_dk576_dv512")]]  kernel flash_at
 template [[host_name("kernel_flash_attn_ext_f16_dk32_dv32"  )]]  kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    half4x4,    1, dequantize_f16,  half4x4,    1, dequantize_f16,  32,  32>;
 template [[host_name("kernel_flash_attn_ext_f16_dk40_dv40"  )]]  kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    half4x4,    1, dequantize_f16,  half4x4,    1, dequantize_f16,  40,  40>;
 template [[host_name("kernel_flash_attn_ext_f16_dk64_dv64"  )]]  kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    half4x4,    1, dequantize_f16,  half4x4,    1, dequantize_f16,  64,  64>;
+template [[host_name("kernel_flash_attn_ext_f16_dk72_dv72"  )]]  kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    half4x4,    1, dequantize_f16,  half4x4,    1, dequantize_f16,  72,  72>;
 template [[host_name("kernel_flash_attn_ext_f16_dk80_dv80"  )]]  kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    half4x4,    1, dequantize_f16,  half4x4,    1, dequantize_f16,  80,  80>;
 template [[host_name("kernel_flash_attn_ext_f16_dk96_dv96"  )]]  kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    half4x4,    1, dequantize_f16,  half4x4,    1, dequantize_f16,  96,  96>;
 template [[host_name("kernel_flash_attn_ext_f16_dk112_dv112")]]  kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    half4x4,    1, dequantize_f16,  half4x4,    1, dequantize_f16,  112, 112>;
@@ -5387,6 +5389,7 @@ template [[host_name("kernel_flash_attn_ext_f16_dk576_dv512")]]  kernel flash_at
 template [[host_name("kernel_flash_attn_ext_bf16_dk32_dv32"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4,  1, dequantize_bf16, bfloat4x4,  1, dequantize_bf16, 32,  32>;
 template [[host_name("kernel_flash_attn_ext_bf16_dk40_dv40"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4,  1, dequantize_bf16, bfloat4x4,  1, dequantize_bf16, 40,  40>;
 template [[host_name("kernel_flash_attn_ext_bf16_dk64_dv64"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4,  1, dequantize_bf16, bfloat4x4,  1, dequantize_bf16, 64,  64>;
+template [[host_name("kernel_flash_attn_ext_bf16_dk72_dv72"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4,  1, dequantize_bf16, bfloat4x4,  1, dequantize_bf16, 72,  72>;
 template [[host_name("kernel_flash_attn_ext_bf16_dk80_dv80"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4,  1, dequantize_bf16, bfloat4x4,  1, dequantize_bf16, 80,  80>;
 template [[host_name("kernel_flash_attn_ext_bf16_dk96_dv96"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4,  1, dequantize_bf16, bfloat4x4,  1, dequantize_bf16, 96,  96>;
 template [[host_name("kernel_flash_attn_ext_bf16_dk112_dv112")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4,  1, dequantize_bf16, bfloat4x4,  1, dequantize_bf16, 112, 112>;
@@ -5400,6 +5403,7 @@ template [[host_name("kernel_flash_attn_ext_bf16_dk576_dv512")]] kernel flash_at
 template [[host_name("kernel_flash_attn_ext_q4_0_dk32_dv32"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 32,  32>;
 template [[host_name("kernel_flash_attn_ext_q4_0_dk40_dv40"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 40,  40>;
 template [[host_name("kernel_flash_attn_ext_q4_0_dk64_dv64"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 64,  64>;
+template [[host_name("kernel_flash_attn_ext_q4_0_dk72_dv72"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 72,  72>;
 template [[host_name("kernel_flash_attn_ext_q4_0_dk80_dv80"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 80,  80>;
 template [[host_name("kernel_flash_attn_ext_q4_0_dk96_dv96"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 96,  96>;
 template [[host_name("kernel_flash_attn_ext_q4_0_dk112_dv112")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 112, 112>;
@@ -5412,6 +5416,7 @@ template [[host_name("kernel_flash_attn_ext_q4_0_dk576_dv512")]] kernel flash_at
 template [[host_name("kernel_flash_attn_ext_q4_1_dk32_dv32"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 32,  32>;
 template [[host_name("kernel_flash_attn_ext_q4_1_dk40_dv40"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 40,  40>;
 template [[host_name("kernel_flash_attn_ext_q4_1_dk64_dv64"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 64,  64>;
+template [[host_name("kernel_flash_attn_ext_q4_1_dk72_dv72"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 72,  72>;
 template [[host_name("kernel_flash_attn_ext_q4_1_dk80_dv80"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 80,  80>;
 template [[host_name("kernel_flash_attn_ext_q4_1_dk96_dv96"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 96,  96>;
 template [[host_name("kernel_flash_attn_ext_q4_1_dk112_dv112")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 112, 112>;
@@ -5424,6 +5429,7 @@ template [[host_name("kernel_flash_attn_ext_q4_1_dk576_dv512")]] kernel flash_at
 template [[host_name("kernel_flash_attn_ext_q5_0_dk32_dv32"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 32,  32>;
 template [[host_name("kernel_flash_attn_ext_q5_0_dk40_dv40"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 40,  40>;
 template [[host_name("kernel_flash_attn_ext_q5_0_dk64_dv64"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 64,  64>;
+template [[host_name("kernel_flash_attn_ext_q5_0_dk72_dv72"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 72,  72>;
 template [[host_name("kernel_flash_attn_ext_q5_0_dk80_dv80"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 80,  80>;
 template [[host_name("kernel_flash_attn_ext_q5_0_dk96_dv96"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 96,  96>;
 template [[host_name("kernel_flash_attn_ext_q5_0_dk112_dv112")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 112, 112>;
@@ -5436,6 +5442,7 @@ template [[host_name("kernel_flash_attn_ext_q5_0_dk576_dv512")]] kernel flash_at
 template [[host_name("kernel_flash_attn_ext_q5_1_dk32_dv32"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 32,  32>;
 template [[host_name("kernel_flash_attn_ext_q5_1_dk40_dv40"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 40,  40>;
 template [[host_name("kernel_flash_attn_ext_q5_1_dk64_dv64"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 64,  64>;
+template [[host_name("kernel_flash_attn_ext_q5_1_dk72_dv72"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 72,  72>;
 template [[host_name("kernel_flash_attn_ext_q5_1_dk80_dv80"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 80,  80>;
 template [[host_name("kernel_flash_attn_ext_q5_1_dk96_dv96"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 96,  96>;
 template [[host_name("kernel_flash_attn_ext_q5_1_dk112_dv112")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 112, 112>;
@@ -5448,6 +5455,7 @@ template [[host_name("kernel_flash_attn_ext_q5_1_dk576_dv512")]] kernel flash_at
 template [[host_name("kernel_flash_attn_ext_q8_0_dk32_dv32"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 32,  32>;
 template [[host_name("kernel_flash_attn_ext_q8_0_dk40_dv40"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 40,  40>;
 template [[host_name("kernel_flash_attn_ext_q8_0_dk64_dv64"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 64,  64>;
+template [[host_name("kernel_flash_attn_ext_q8_0_dk72_dv72"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 72,  72>;
 template [[host_name("kernel_flash_attn_ext_q8_0_dk80_dv80"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 80,  80>;
 template [[host_name("kernel_flash_attn_ext_q8_0_dk96_dv96"  )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 96,  96>;
 template [[host_name("kernel_flash_attn_ext_q8_0_dk112_dv112")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES,    block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 112, 112>;

ggml/src/ggml-opencl/ggml-opencl.cpp

@@ -8399,6 +8399,7 @@ static void ggml_cl_rope(ggml_backend_t backend, const ggml_tensor * src0, const
     const bool is_neox = mode & 2;
     const bool is_mrope = mode & GGML_ROPE_TYPE_MROPE;
     const bool is_vision = mode == GGML_ROPE_TYPE_VISION;
+    const int  is_imrope = mode == GGML_ROPE_TYPE_IMROPE;
 
     if (is_mrope) {
         GGML_ASSERT(sections[0] > 0 || sections[1] > 0 || sections[2] > 0);
@@ -8489,9 +8490,14 @@ static void ggml_cl_rope(ggml_backend_t backend, const ggml_tensor * src0, const
     CL_CHECK(clSetKernelArg(kernel, 30, sizeof(float),    &attn_factor));
     CL_CHECK(clSetKernelArg(kernel, 31, sizeof(float),    &beta_fast));
     CL_CHECK(clSetKernelArg(kernel, 32, sizeof(float),    &beta_slow));
+    // both mrope and vision kernels have sections
     if (is_mrope || is_vision) {
         CL_CHECK(clSetKernelArg(kernel, 33, sizeof(int32_t)*4, &sections));
     }
+    // only mrope has is_imrope
+    if (is_mrope && !is_vision) {
+        CL_CHECK(clSetKernelArg(kernel, 34, sizeof(int), &is_imrope));
+    }
 
     size_t global_work_size[] = {(size_t)ne01*nth, (size_t)ne02, (size_t)ne03};
     size_t local_work_size[] = {(size_t)nth, 1, 1};

ggml/src/ggml-opencl/kernels/rope.cl

@@ -392,7 +392,8 @@ kernel void kernel_rope_multi_f32(
         float attn_factor,
         float beta_fast,
         float beta_slow,
-        int4 sections
+        int4 sections,
+        int  is_imrope
 ) {
     src0 = (global void*)((global char*)src0 + offset0);
     src1 = (global int*)((global char*)src1 + offset1);
@@ -419,17 +420,29 @@ kernel void kernel_rope_multi_f32(
             const int sector = (i0 / 2) % sect_dims;
             float theta_base = 0.0f;
 
-            if (sector < sections.s0) {
-                theta_base = pos[i2];
-            }
-            else if (sector >= sections.s0 && sector < sec_w) {
-                theta_base = pos[i2 + ne2 * 1];
-            }
-            else if (sector >= sec_w && sector < sec_w + sections.s2) {
-                theta_base = pos[i2 + ne2 * 2];
-            }
-            else if (sector >= sec_w + sections.s2) {
-                theta_base = pos[i2 + ne2 * 3];
+            if (is_imrope) {
+                if (sector % 3 == 1 && sector < 3 * sections.s1) { // h
+                    theta_base = (float) pos[i2 + ne02 * 1];
+                } else if (sector % 3 == 2 && sector < 3 * sections.s2) { // w
+                    theta_base = (float) pos[i2 + ne02 * 2];
+                } else if (sector % 3 == 0 && sector < 3 * sections.s0) { // t
+                    theta_base = (float) pos[i2 + ne02 * 0];
+                } else { // e
+                    theta_base = (float) pos[i2 + ne02 * 3];
+                }
+            } else {
+                if (sector < sections.s0) {
+                    theta_base = pos[i2];
+                }
+                else if (sector >= sections.s0 && sector < sec_w) {
+                    theta_base = pos[i2 + ne2 * 1];
+                }
+                else if (sector >= sec_w && sector < sec_w + sections.s2) {
+                    theta_base = pos[i2 + ne2 * 2];
+                }
+                else if (sector >= sec_w + sections.s2) {
+                    theta_base = pos[i2 + ne2 * 3];
+                }
             }
 
             const float theta = theta_base * pow(freq_base, inv_ndims*i0);
@@ -490,7 +503,8 @@ kernel void kernel_rope_multi_f16(
         float attn_factor,
         float beta_fast,
         float beta_slow,
-        int4 sections
+        int4 sections,
+        int  is_imrope
 ) {
     src0 = (global void*)((global char*)src0 + offset0);
     src1 = (global int*)((global char*)src1 + offset1);
@@ -517,17 +531,29 @@ kernel void kernel_rope_multi_f16(
             const int sector = (i0 / 2) % sect_dims;
             float theta_base = 0.0f;
 
-            if (sector < sections.s0) {
-                theta_base = pos[i2];
-            }
-            else if (sector >= sections.s0 && sector < sec_w) {
-                theta_base = pos[i2 + ne2 * 1];
-            }
-            else if (sector >= sec_w && sector < sec_w + sections.s2) {
-                theta_base = pos[i2 + ne2 * 2];
-            }
-            else if (sector >= sec_w + sections.s2) {
-                theta_base = pos[i2 + ne2 * 3];
+            if (is_imrope) {
+                if (sector % 3 == 1 && sector < 3 * sections.s1) { // h
+                    theta_base = (float) pos[i2 + ne02 * 1];
+                } else if (sector % 3 == 2 && sector < 3 * sections.s2) { // w
+                    theta_base = (float) pos[i2 + ne02 * 2];
+                } else if (sector % 3 == 0 && sector < 3 * sections.s0) { // t
+                    theta_base = (float) pos[i2 + ne02 * 0];
+                } else { // e
+                    theta_base = (float) pos[i2 + ne02 * 3];
+                }
+            } else {
+                if (sector < sections.s0) {
+                    theta_base = pos[i2];
+                }
+                else if (sector >= sections.s0 && sector < sec_w) {
+                    theta_base = pos[i2 + ne2 * 1];
+                }
+                else if (sector >= sec_w && sector < sec_w + sections.s2) {
+                    theta_base = pos[i2 + ne2 * 2];
+                }
+                else if (sector >= sec_w + sections.s2) {
+                    theta_base = pos[i2 + ne2 * 3];
+                }
             }
 
             const float theta = theta_base * pow(freq_base, inv_ndims*i0);

ggml/src/ggml-sycl/repeat_back.cpp

@@ -2,26 +2,43 @@
 
 #include "common.hpp"
 
-void ggml_sycl_op_repeat_back(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
+#define GGML_ASSERT_TENSOR_FITS_INT(t) \
+    GGML_ASSERT((t)->ne[0] < INT_MAX && (t)->ne[1] < INT_MAX && (t)->ne[2] < INT_MAX && (t)->ne[3] < INT_MAX)
 
+void ggml_sycl_op_repeat_back(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
     GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32);
     GGML_ASSERT(dst->type == GGML_TYPE_F32);
 
     const float * src0_dd = (const float *) dst->src[0]->data;
     float *       dst_dd  = (float *) dst->data;
 
-    const int64_t ne0 = dst->ne[0], ne1 = dst->ne[1], ne2 = dst->ne[2], ne3 = dst->ne[3];
-    const int64_t ne00 = dst->src[0]->ne[0], ne01 = dst->src[0]->ne[1], ne02 = dst->src[0]->ne[2],
-                  ne03 = dst->src[0]->ne[3];
+    GGML_ASSERT_TENSOR_FITS_INT(dst);
+    GGML_ASSERT_TENSOR_FITS_INT(dst->src[0]);
 
-    const int nr0 = (int) (ne00 / ne0);
-    const int nr1 = (int) (ne01 / ne1);
-    const int nr2 = (int) (ne02 / ne2);
-    const int nr3 = (int) (ne03 / ne3);
+    const int ne0 = dst->ne[0], ne1 = dst->ne[1], ne2 = dst->ne[2], ne3 = dst->ne[3];
+    const int ne00 = dst->src[0]->ne[0], ne01 = dst->src[0]->ne[1], ne02 = dst->src[0]->ne[2],
+              ne03 = dst->src[0]->ne[3];
 
-    const size_t total      = ne0 * ne1 * ne2 * ne3;
-    const int    BLOCK_SIZE = 256;
-    const int    num_blocks = (total + BLOCK_SIZE - 1) / BLOCK_SIZE;
+    const int nr0 = ne00 / ne0;
+    const int nr1 = ne01 / ne1;
+    const int nr2 = ne02 / ne2;
+    const int nr3 = ne03 / ne3;
+
+    const int nb0 = dst->src[0]->nb[0];
+    const int nb1 = dst->src[0]->nb[1];
+    const int nb2 = dst->src[0]->nb[2];
+    const int nb3 = dst->src[0]->nb[3];
+
+    const char * base = (const char *) src0_dd;
+
+    const size_t  total      = (size_t) ne0 * ne1 * ne2 * ne3;
+    constexpr int BLOCK_SIZE = 256;
+    const int     num_blocks = (total + BLOCK_SIZE - 1) / BLOCK_SIZE;
+
+    const float inv_ne0      = 1.0f / ne0;
+    const float inv_ne_01    = 1.0f / (ne0 * ne1);
+    const float inv_ne_012   = 1.0f / (ne0 * ne1 * ne2);
+    const int   repeat_count = nr0 * nr1 * nr2 * nr3;
 
     queue_ptr stream = ctx.stream();
 
@@ -33,24 +50,27 @@ void ggml_sycl_op_repeat_back(ggml_backend_sycl_context & ctx, ggml_tensor * dst
                 return;
             }
 
-            const int i0 = i % ne0;
-            const int i1 = (i / ne0) % ne1;
-            const int i2 = (i / (ne0 * ne1)) % ne2;
-            const int i3 = i / (ne0 * ne1 * ne2);
+            const int i3 = (int) (i * inv_ne_012);
+            const int i2 = (int) (i * inv_ne_01) - i3 * ne2;
+            const int i1 = (int) (i * inv_ne0) - (int) (i * inv_ne_01) * ne1;
+            const int i0 = i - (int) (i * inv_ne0) * ne0;
 
+            int   j0 = 0, j1 = 0, j2 = 0, j3 = 0;
             float acc = 0.0f;
 
-            for (int j3 = 0; j3 < nr3; ++j3) {
-                for (int j2 = 0; j2 < nr2; ++j2) {
-                    for (int j1 = 0; j1 < nr1; ++j1) {
-                        for (int j0 = 0; j0 < nr0; ++j0) {
-                            acc += src0_dd[(i0 + j0 * ne0) + (i1 + j1 * ne1) * ne00 + (i2 + j2 * ne2) * ne00 * ne01 +
-                                           (i3 + j3 * ne3) * ne00 * ne01 * ne02];
-                        }
-                    }
-                }
-            }
+            for (int j = 0; j < repeat_count; ++j) {
+                const float * ptr = (const float *) (base + (i0 + j0 * ne0) * nb0 + (i1 + j1 * ne1) * nb1 +
+                    (i2 + j2 * ne2) * nb2 + (i3 + j3 * ne3) * nb3);
+                acc += *ptr;
 
+                int carry = (++j0 >= nr0);
+                j0 -= carry * nr0;
+                carry = (carry && (++j1 >= nr1));
+                j1 -= carry * nr1;
+                carry = (carry && (++j2 >= nr2));
+                j2 -= carry * nr2;
+                j3 += carry;
+            }
             dst_dd[i] = acc;
         });
 }

ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_base.glsl

@@ -28,8 +28,11 @@ layout (binding = 1) readonly buffer BV4 {B_TYPE_VEC4 data_b_v4[];};
 #endif
 
 layout (binding = 2) writeonly buffer D {D_TYPE data_d[];};
+
+layout (binding = 3) readonly buffer Bias {D_TYPE data_bias[];};
+
 #ifdef MUL_MAT_ID
-layout (binding = 3) readonly buffer IDS {int data_ids[];};
+layout (binding = 4) readonly buffer IDS {int data_ids[];};
 #endif
 
 #include "dequant_funcs.glsl"
@@ -45,6 +48,8 @@ layout (push_constant) uniform parameter
     uint batch_stride_b;
     uint batch_stride_d;
 
+    uint enable_bias;
+
 #ifdef MUL_MAT_ID
     uint nei0;
     uint ne11;
@@ -56,6 +61,10 @@ layout (push_constant) uniform parameter
 #endif
 } p;
 
+#ifdef MUL_MAT_ID
+uint expert_id;
+#endif
+
 void get_offsets(out uint a_offset, out uint b_offset, out uint d_offset) {
 #ifdef MUL_MAT_ID
     const uint expert_idx = gl_GlobalInvocationID.y;
@@ -75,7 +84,7 @@ void get_offsets(out uint a_offset, out uint b_offset, out uint d_offset) {
         batch_idx_a = i03 * p.ne02 + i02;
     }
 #else
-    const uint expert_id = data_ids[expert_idx];
+    expert_id = data_ids[expert_idx];
 #endif
 
     a_offset =
@@ -113,6 +122,13 @@ void reduce_result(inout FLOAT_TYPE temp[NUM_COLS][NUM_ROWS], const in uint32_t
     if (tid == 0) {
         [[unroll]] for (uint j = 0; j < NUM_COLS; ++j) {
             [[unroll]] for (uint n = 0; n < num_rows; ++n) {
+                if (p.enable_bias != 0) {
+#ifdef MUL_MAT_ID
+                    temp[j][n] += FLOAT_TYPE(data_bias[expert_id*p.stride_d + first_row + n]);
+#else
+                    temp[j][n] += FLOAT_TYPE(data_bias[j*p.batch_stride_d + d_offset + first_row + n]);
+#endif
+                }
                 data_d[j*p.batch_stride_d + d_offset + first_row + n] = D_TYPE(temp[j][n]);
             }
         }
@@ -148,6 +164,13 @@ void reduce_result(FLOAT_TYPE temp[NUM_COLS][NUM_ROWS], const in uint32_t d_offs
                 [[unroll]] for (uint s = 0; s < gl_NumSubgroups; ++s) {
                     temp[j][n] += tmpsh[j][n][s];
                 }
+                if (p.enable_bias != 0) {
+#ifdef MUL_MAT_ID
+                    temp[j][n] += FLOAT_TYPE(data_bias[expert_id*p.stride_d + first_row + n]);
+#else
+                    temp[j][n] += FLOAT_TYPE(data_bias[j*p.batch_stride_d + d_offset + first_row + n]);
+#endif
+                }
                 data_d[j*p.batch_stride_d + d_offset + first_row + n] = D_TYPE(temp[j][n]);
             }
         }
@@ -173,6 +196,13 @@ void reduce_result(FLOAT_TYPE temp[NUM_COLS][NUM_ROWS], const in uint32_t d_offs
     if (tid == 0) {
         [[unroll]] for (uint j = 0; j < NUM_COLS; ++j) {
             [[unroll]] for (uint n = 0; n < num_rows; ++n) {
+                if (p.enable_bias != 0) {
+#ifdef MUL_MAT_ID
+                    tmpsh[j][n][0] += FLOAT_TYPE(data_bias[expert_id*p.stride_d + first_row + n]);
+#else
+                    tmpsh[j][n][0] += FLOAT_TYPE(data_bias[j*p.batch_stride_d + d_offset + first_row + n]);
+#endif
+                }
                 data_d[j*p.batch_stride_d + d_offset + first_row + n] = D_TYPE(tmpsh[j][n][0]);
             }
         }

ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_nc.comp

@@ -15,6 +15,8 @@ layout (binding = 2) writeonly buffer D {D_TYPE dst[];};
 layout (binding = 0) readonly buffer AV4 {A_TYPE_VEC4 data_a_v4[];};
 layout (binding = 1) readonly buffer BV4 {B_TYPE_VEC4 data_b_v4[];};
 
+layout (binding = 3) readonly buffer Bias {D_TYPE data_bias[];};
+
 layout (push_constant) uniform parameter
 {
     uint ncols_x;
@@ -29,6 +31,7 @@ layout (push_constant) uniform parameter
     uint nb03;
     uint nb13;
     uint nb23;
+    uint enable_bias;
 } p;
 
 shared FLOAT_TYPE tmp[BLOCK_SIZE];
@@ -117,6 +120,9 @@ void main() {
     }
 
     if (tid == 0) {
+        if (p.enable_bias != 0) {
+            tmp[0] += FLOAT_TYPE(data_bias[idst]);
+        }
         dst[idst] = tmp[0];
     }
 }

ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_p021.comp

@@ -17,6 +17,8 @@ layout (binding = 2) writeonly buffer D {D_TYPE dst[];};
 layout (binding = 0) readonly buffer AV4 {A_TYPE_VEC4 data_a_v4[];};
 layout (binding = 1) readonly buffer BV4 {B_TYPE_VEC4 data_b_v4[];};
 
+layout (binding = 3) readonly buffer Bias {D_TYPE data_bias[];};
+
 layout(constant_id = 0) const int BLOCK_SIZE = 32;
 // gqa_ratio is in the range [1,8]
 layout(constant_id = 1) const uint gqa_ratio = 1;
@@ -29,6 +31,7 @@ layout (push_constant) uniform parameter
     uint nchannels_y;
     uint b_offset;
     uint d_offset;
+    uint enable_bias;
 } p;
 
 #if !USE_SUBGROUP_ADD
@@ -148,6 +151,9 @@ void main() {
         [[unroll]] for (uint c = 0; c < gqa_ratio; ++c) {
             // dst is not transposed and not permuted
             const uint idst = (channel + c)*nrows_dst + row_dst;
+            if (p.enable_bias != 0) {
+                temp[c] += FLOAT_TYPE(data_bias[idst]);
+            }
             dst[idst] = temp[c];
         }
     }

ggml/src/ggml-vulkan/vulkan-shaders/multi_add.comp

@@ -23,16 +23,100 @@ layout (push_constant) uniform parameter2
     uint rms_partials;
 } p;
 
-// Workaround for MoltenVK Bug, see https://github.com/ggml-org/llama.cpp/issues/15498
-// layout (binding = 0) readonly buffer A {A_TYPE data_a[];} a[];
-// layout (binding = 0) writeonly buffer D {D_TYPE data_d[];} d[];
-layout (binding = 0) buffer A {A_TYPE data_a[];} a[];
-layout (binding = 0) buffer D {D_TYPE data_d[];} d[];
-
-layout (binding = 0, std430) buffer PartialBuf {float partial_sums[];} partials[];
+// No readonly/writeonly decorations. Workaround for MoltenVK Bug, see https://github.com/ggml-org/llama.cpp/issues/15498
+layout (binding = 0)  buffer A0 {A_TYPE data_a[];} a0;
+layout (binding = 1)  buffer A1 {A_TYPE data_a[];} a1;
+layout (binding = 2)  buffer A2 {A_TYPE data_a[];} a2;
+layout (binding = 3)  buffer A3 {A_TYPE data_a[];} a3;
+layout (binding = 4)  buffer A4 {A_TYPE data_a[];} a4;
+layout (binding = 5)  buffer A5 {A_TYPE data_a[];} a5;
+layout (binding = 6)  buffer A6 {A_TYPE data_a[];} a6;
+layout (binding = 7)  buffer A7 {A_TYPE data_a[];} a7;
+layout (binding = 8)  buffer A8 {A_TYPE data_a[];} a8;
+layout (binding = 9)  buffer A9 {A_TYPE data_a[];} a9;
+layout (binding = 10) buffer A10 {A_TYPE data_a[];} a10;
+layout (binding = 11) buffer A11 {A_TYPE data_a[];} a11;
+layout (binding = 0)  buffer D0 {D_TYPE data_d[];} d0;
+layout (binding = 1)  buffer D1 {D_TYPE data_d[];} d1;
+layout (binding = 2)  buffer D2 {D_TYPE data_d[];} d2;
+layout (binding = 3)  buffer D3 {D_TYPE data_d[];} d3;
+layout (binding = 4)  buffer D4 {D_TYPE data_d[];} d4;
+layout (binding = 5)  buffer D5 {D_TYPE data_d[];} d5;
+layout (binding = 6)  buffer D6 {D_TYPE data_d[];} d6;
+layout (binding = 7)  buffer D7 {D_TYPE data_d[];} d7;
+layout (binding = 8)  buffer D8 {D_TYPE data_d[];} d8;
+layout (binding = 9)  buffer D9 {D_TYPE data_d[];} d9;
+layout (binding = 10) buffer D10 {D_TYPE data_d[];} d10;
+layout (binding = 11) buffer D11 {D_TYPE data_d[];} d11;
+layout (binding = 0, std430)  buffer PartialBuf0 {float partial_sums[];} partials0;
+layout (binding = 1, std430)  buffer PartialBuf1 {float partial_sums[];} partials1;
+layout (binding = 2, std430)  buffer PartialBuf2 {float partial_sums[];} partials2;
+layout (binding = 3, std430)  buffer PartialBuf3 {float partial_sums[];} partials3;
+layout (binding = 4, std430)  buffer PartialBuf4 {float partial_sums[];} partials4;
+layout (binding = 5, std430)  buffer PartialBuf5 {float partial_sums[];} partials5;
+layout (binding = 6, std430)  buffer PartialBuf6 {float partial_sums[];} partials6;
+layout (binding = 7, std430)  buffer PartialBuf7 {float partial_sums[];} partials7;
+layout (binding = 8, std430)  buffer PartialBuf8 {float partial_sums[];} partials8;
+layout (binding = 9, std430)  buffer PartialBuf9 {float partial_sums[];} partials9;
+layout (binding = 10, std430) buffer PartialBuf10 {float partial_sums[];} partials10;
+layout (binding = 11, std430) buffer PartialBuf11 {float partial_sums[];} partials11;
 
 layout(constant_id = 0) const uint num_srcs = 2;
 
+FLOAT_TYPE load_a(uint b, uint i) {
+    switch (b) {
+    case 0:  return FLOAT_TYPE(a0.data_a[i]);
+    case 1:  return FLOAT_TYPE(a1.data_a[i]);
+    case 2:  return FLOAT_TYPE(a2.data_a[i]);
+    case 3:  return FLOAT_TYPE(a3.data_a[i]);
+    case 4:  return FLOAT_TYPE(a4.data_a[i]);
+    case 5:  return FLOAT_TYPE(a5.data_a[i]);
+    case 6:  return FLOAT_TYPE(a6.data_a[i]);
+    case 7:  return FLOAT_TYPE(a7.data_a[i]);
+    case 8:  return FLOAT_TYPE(a8.data_a[i]);
+    case 9:  return FLOAT_TYPE(a9.data_a[i]);
+    case 10: return FLOAT_TYPE(a10.data_a[i]);
+    case 11: return FLOAT_TYPE(a11.data_a[i]);
+    default: return FLOAT_TYPE(0);
+    }
+}
+
+void store_d(uint b, uint i, FLOAT_TYPE v) {
+    switch (b) {
+    case 0:  d0.data_d[i] = D_TYPE(v); break;
+    case 1:  d1.data_d[i] = D_TYPE(v); break;
+    case 2:  d2.data_d[i] = D_TYPE(v); break;
+    case 3:  d3.data_d[i] = D_TYPE(v); break;
+    case 4:  d4.data_d[i] = D_TYPE(v); break;
+    case 5:  d5.data_d[i] = D_TYPE(v); break;
+    case 6:  d6.data_d[i] = D_TYPE(v); break;
+    case 7:  d7.data_d[i] = D_TYPE(v); break;
+    case 8:  d8.data_d[i] = D_TYPE(v); break;
+    case 9:  d9.data_d[i] = D_TYPE(v); break;
+    case 10: d10.data_d[i] = D_TYPE(v); break;
+    case 11: d11.data_d[i] = D_TYPE(v); break;
+    default: break;
+    }
+}
+
+void store_partial(uint b, uint i, float v) {
+    switch (b) {
+    case 0:  partials0.partial_sums[i] = v; break;
+    case 1:  partials1.partial_sums[i] = v; break;
+    case 2:  partials2.partial_sums[i] = v; break;
+    case 3:  partials3.partial_sums[i] = v; break;
+    case 4:  partials4.partial_sums[i] = v; break;
+    case 5:  partials5.partial_sums[i] = v; break;
+    case 6:  partials6.partial_sums[i] = v; break;
+    case 7:  partials7.partial_sums[i] = v; break;
+    case 8:  partials8.partial_sums[i] = v; break;
+    case 9:  partials9.partial_sums[i] = v; break;
+    case 10: partials10.partial_sums[i] = v; break;
+    case 11: partials11.partial_sums[i] = v; break;
+    default: break;
+    }
+}
+
 uint src_idx(uint s, uint i00, uint i01, uint i02, uint i03) {
     return i03*p.nb[s][3] + i02*p.nb[s][2] + i01*p.nb[s][1] + i00*p.nb[s][0];
 }
@@ -78,10 +162,10 @@ void main() {
 
         FLOAT_TYPE sum = FLOAT_TYPE(0);
         [[unroll]] for (uint s = 0; s < num_srcs; ++s) {
-            sum += FLOAT_TYPE(a[s].data_a[src_idx(s, i00, i01, i02, i03)]);
+            sum += load_a(s, src_idx(s, i00, i01, i02, i03));
         }
         sum_sq += sum*sum;
-        d[num_srcs].data_d[dst_idx(i00, i01, i02, i03)] = D_TYPE(sum);
+        store_d(num_srcs, dst_idx(i00, i01, i02, i03), sum);
 
         idx += num_threads;
     }
@@ -104,7 +188,7 @@ void main() {
         }
 
         if (gl_SubgroupID == 0 && gl_SubgroupInvocationID == 0) {
-            partials[num_srcs + 1].partial_sums[orig_idx / (num_iter * num_threads)] = sum_sq;
+            store_partial(num_srcs + 1, orig_idx / (num_iter * num_threads), sum_sq);
         }
     }
 #endif

ggml/src/ggml-webgpu/ggml-webgpu.cpp

@@ -248,7 +248,7 @@ struct webgpu_context_struct {
 
     webgpu_pipeline memset_pipeline;
     webgpu_pipeline mul_mat_pipeline[30][2];
-    webgpu_pipeline set_rows_pipeline;
+    webgpu_pipeline set_rows_pipeline[1][2];  // dst->type, vectorized
     webgpu_pipeline get_rows_pipeline[30];
     webgpu_pipeline get_rows_f32_no_vec_pipeline;
     webgpu_pipeline cpy_pipeline[2][2];          // src type, dst type
@@ -309,10 +309,12 @@ struct ggml_backend_webgpu_context {
 struct ggml_backend_webgpu_buffer_context {
     webgpu_context webgpu_ctx;
     wgpu::Buffer   buffer;
+    std::string    label;
 
-    ggml_backend_webgpu_buffer_context(webgpu_context ctx, wgpu::Buffer buf) :
+    ggml_backend_webgpu_buffer_context(webgpu_context ctx, wgpu::Buffer buf, std::string lbl) :
         webgpu_ctx(std::move(ctx)),
-        buffer(std::move(buf)) {}
+        buffer(std::move(buf)),
+        label(std::move(lbl)) {}
 };
 
 /* End struct definitions */
@@ -764,10 +766,20 @@ static std::optional<webgpu_command> ggml_webgpu_set_rows(webgpu_context & ctx,
         { .binding = 3, .buffer = error_bufs.dev_buf, .offset = 0, .size = error_bufs.dev_buf.GetSize() }
     };
 
-    size_t   max_wg_size = ctx->max_wg_size_x;
-    uint32_t wg_x        = (src->ne[1] * src->ne[2] * src->ne[3] + max_wg_size - 1) / max_wg_size;
+    size_t max_wg_size = ctx->max_wg_size_x;
 
-    return ggml_backend_webgpu_build(ctx, ctx->set_rows_pipeline, params, entries, wg_x, error_bufs);
+    int             vectorized = src->ne[0] % 4 == 0;
+    webgpu_pipeline pipeline   = ctx->set_rows_pipeline[0][vectorized];
+    uint32_t        threads;
+    if (vectorized) {
+        threads = (src->ne[1] * src->ne[2] * src->ne[3]) * (src->ne[0] / 4);
+    } else {
+        threads = src->ne[0] * src->ne[1] * src->ne[2] * src->ne[3];
+    }
+
+    uint32_t wg_x = (threads + max_wg_size - 1) / max_wg_size;
+
+    return ggml_backend_webgpu_build(ctx, pipeline, params, entries, wg_x, error_bufs);
 }
 
 static webgpu_command ggml_webgpu_get_rows(webgpu_context & ctx,
@@ -1336,11 +1348,11 @@ static void ggml_backend_webgpu_buffer_memset_tensor(ggml_backend_buffer_t buffe
 
     WEBGPU_CPU_PROFILE_TOTAL_START(memset_tensor);
 
-    WEBGPU_LOG_DEBUG("ggml_backend_webgpu_buffer_memset_tensor(" << buffer << ", " << tensor << ", " << value << ", "
-                                                                 << offset << ", " << size << ")");
-
     ggml_backend_webgpu_buffer_context * buf_ctx = (ggml_backend_webgpu_buffer_context *) buffer->context;
 
+    WEBGPU_LOG_DEBUG("ggml_backend_webgpu_buffer_memset_tensor(" << buf_ctx->label << ", " << tensor << ", " << value
+                                                                 << ", " << offset << ", " << size << ")");
+
     size_t total_offset = webgpu_tensor_offset(tensor) + tensor->view_offs + offset;
 
     // This is a trick to set all bytes of a u32 to the same 1 byte value.
@@ -1354,12 +1366,13 @@ static void ggml_backend_webgpu_buffer_set_tensor(ggml_backend_buffer_t buffer,
                                                   const void *          data,
                                                   size_t                offset,
                                                   size_t                size) {
-    WEBGPU_LOG_DEBUG("ggml_backend_webgpu_buffer_set_tensor(" << buffer << ", " << tensor << ", " << data << ", "
-                                                              << offset << ", " << size << ")");
     WEBGPU_CPU_PROFILE_TOTAL_START(set_tensor);
     ggml_backend_webgpu_buffer_context * buf_ctx    = (ggml_backend_webgpu_buffer_context *) buffer->context;
     webgpu_context                       webgpu_ctx = buf_ctx->webgpu_ctx;
 
+    WEBGPU_LOG_DEBUG("ggml_backend_webgpu_buffer_set_tensor(" << buf_ctx->label << ", " << tensor << ", " << data
+                                                              << ", " << offset << ", " << size << ")");
+
     size_t total_offset = webgpu_tensor_offset(tensor) + tensor->view_offs + offset;
 
     webgpu_ctx->queue.WriteBuffer(buf_ctx->buffer, total_offset, data, (size / 4) * 4);
@@ -1397,12 +1410,12 @@ static void ggml_backend_webgpu_buffer_get_tensor(ggml_backend_buffer_t buffer,
                                                   void *                data,
                                                   size_t                offset,
                                                   size_t                size) {
-    WEBGPU_LOG_DEBUG("ggml_backend_webgpu_buffer_get_tensor(" << buffer << ", " << tensor << ", " << data << ", "
-                                                              << offset << ", " << size << ")");
     WEBGPU_CPU_PROFILE_TOTAL_START(get_tensor);
-    ggml_backend_webgpu_buffer_context * buf_ctx    = (ggml_backend_webgpu_buffer_context *) buffer->context;
-    webgpu_context                       webgpu_ctx = buf_ctx->webgpu_ctx;
-    wgpu::Device                         device     = webgpu_ctx->device;
+    ggml_backend_webgpu_buffer_context * buf_ctx = (ggml_backend_webgpu_buffer_context *) buffer->context;
+    WEBGPU_LOG_DEBUG("ggml_backend_webgpu_buffer_get_tensor(" << buf_ctx->label << ", " << tensor << ", " << data
+                                                              << ", " << offset << ", " << size << ")");
+    webgpu_context webgpu_ctx = buf_ctx->webgpu_ctx;
+    wgpu::Device   device     = webgpu_ctx->device;
 
     size_t total_offset = webgpu_tensor_offset(tensor) + tensor->view_offs + offset;
 
@@ -1473,16 +1486,20 @@ static const char * ggml_backend_webgpu_buffer_type_get_name(ggml_backend_buffer
 
 static ggml_backend_buffer_t ggml_backend_webgpu_buffer_type_alloc_buffer(ggml_backend_buffer_type_t buft,
                                                                           size_t                     size) {
-    WEBGPU_LOG_DEBUG("ggml_backend_webgpu_buffer_type_alloc_buffer(" << size << ")");
+    static std::atomic<int> buffer_count;
+    int                     buffer_id = buffer_count++;
+    std::string             buf_name  = "tensor_buf" + std::to_string(buffer_id);
+    WEBGPU_LOG_DEBUG("ggml_backend_webgpu_buffer_type_alloc_buffer_" << buffer_id << ": " << size << " bytes");
     ggml_backend_webgpu_device_context * ctx = static_cast<ggml_backend_webgpu_device_context *>(buft->device->context);
 
     wgpu::Buffer buf;
     ggml_webgpu_create_buffer(ctx->webgpu_ctx->device, buf,
                               (size + WEBGPU_STORAGE_BUF_BINDING_MULT - 1) & ~(WEBGPU_STORAGE_BUF_BINDING_MULT - 1),
                               wgpu::BufferUsage::Storage | wgpu::BufferUsage::CopySrc | wgpu::BufferUsage::CopyDst,
-                              "allocated_buffer");
+                              buf_name.c_str());
 
-    ggml_backend_webgpu_buffer_context * buf_ctx = new ggml_backend_webgpu_buffer_context(ctx->webgpu_ctx, buf);
+    ggml_backend_webgpu_buffer_context * buf_ctx =
+        new ggml_backend_webgpu_buffer_context(ctx->webgpu_ctx, buf, buf_name);
 
     return ggml_backend_buffer_init(buft, ggml_backend_webgpu_buffer_interface, buf_ctx, size);
 }
@@ -1613,8 +1630,10 @@ static void ggml_webgpu_init_mul_mat_pipeline(webgpu_context & webgpu_ctx) {
 }
 
 static void ggml_webgpu_init_set_rows_pipeline(webgpu_context & webgpu_ctx) {
-    ggml_webgpu_create_pipeline(webgpu_ctx->device, webgpu_ctx->set_rows_pipeline, wgsl_set_rows, "set_rows",
-                                ggml_webgpu_wg_size_entry(webgpu_ctx->max_wg_size_x));
+    ggml_webgpu_create_pipeline(webgpu_ctx->device, webgpu_ctx->set_rows_pipeline[0][0], wgsl_set_rows_f16,
+                                "set_rows_f16", ggml_webgpu_wg_size_entry(webgpu_ctx->max_wg_size_x));
+    ggml_webgpu_create_pipeline(webgpu_ctx->device, webgpu_ctx->set_rows_pipeline[0][1], wgsl_set_rows_f16_vec,
+                                "set_rows_f16_vec", ggml_webgpu_wg_size_entry(webgpu_ctx->max_wg_size_x));
 }
 
 static void ggml_webgpu_init_get_rows_pipeline(webgpu_context & webgpu_ctx) {
@@ -1950,8 +1969,10 @@ static bool ggml_backend_webgpu_device_supports_op(ggml_backend_dev_t dev, const
         case GGML_OP_SUB:
         case GGML_OP_MUL:
         case GGML_OP_DIV:
+            // TODO: support non-contiguous tensors, e.g. for MOE_EXPERT_REDUCE
+            // see https://github.com/ggml-org/llama.cpp/pull/16857
             supports_op = (op->type == GGML_TYPE_F32 || op->type == GGML_TYPE_F16) && (src0->type == op->type) &&
-                          (src1->type == op->type);
+                          (src1->type == op->type) && ggml_is_contiguous(src0) && ggml_is_contiguous(src1);
             break;
         case GGML_OP_CPY:
         case GGML_OP_CONT:
@@ -2129,6 +2150,19 @@ static ggml_backend_dev_t ggml_backend_webgpu_reg_get_device(ggml_backend_reg_t
     required_features.push_back(wgpu::FeatureName::TimestampQuery);
 #endif
 
+    // Enable Dawn-specific toggles to increase native performance
+    // TODO: Don't enable for WASM builds, they won't have an effect anyways
+    // TODO: Maybe WebGPU needs a "fast" mode where you can request compilers skip adding checks like these,
+    //       only for native performance?
+    const char * const deviceEnabledToggles[]  = { "skip_validation", "disable_robustness", "disable_workgroup_init",
+                                                   "disable_polyfills_on_integer_div_and_mod" };
+    const char * const deviceDisabledToggles[] = { "timestamp_quantization" };
+    wgpu::DawnTogglesDescriptor deviceTogglesDesc;
+    deviceTogglesDesc.enabledToggles      = deviceEnabledToggles;
+    deviceTogglesDesc.enabledToggleCount  = 4;
+    deviceTogglesDesc.disabledToggles     = deviceDisabledToggles;
+    deviceTogglesDesc.disabledToggleCount = 1;
+
     wgpu::DeviceDescriptor dev_desc;
     dev_desc.requiredLimits       = &ctx->limits;
     dev_desc.requiredFeatures     = required_features.data();
@@ -2146,6 +2180,7 @@ static ggml_backend_dev_t ggml_backend_webgpu_reg_get_device(ggml_backend_reg_t
             GGML_ABORT("ggml_webgpu: Device error! Reason: %d, Message: %s\n", static_cast<int>(reason),
                        std::string(message).c_str());
         });
+    dev_desc.nextInChain = &deviceTogglesDesc;
     ctx->instance.WaitAny(ctx->adapter.RequestDevice(
                               &dev_desc, wgpu::CallbackMode::AllowSpontaneous,
                               [ctx](wgpu::RequestDeviceStatus status, wgpu::Device device, wgpu::StringView message) {
@@ -2243,11 +2278,18 @@ ggml_backend_reg_t ggml_backend_webgpu_reg() {
     ctx.name         = GGML_WEBGPU_NAME;
     ctx.device_count = 1;
 
+    const char * const instanceEnabledToggles[] = { "allow_unsafe_apis" };
+
+    wgpu::DawnTogglesDescriptor instanceTogglesDesc;
+    instanceTogglesDesc.enabledToggles     = instanceEnabledToggles;
+    instanceTogglesDesc.enabledToggleCount = 1;
     wgpu::InstanceDescriptor               instance_descriptor{};
     std::vector<wgpu::InstanceFeatureName> instance_features = { wgpu::InstanceFeatureName::TimedWaitAny };
     instance_descriptor.requiredFeatures                     = instance_features.data();
     instance_descriptor.requiredFeatureCount                 = instance_features.size();
-    webgpu_ctx->instance                                     = wgpu::CreateInstance(&instance_descriptor);
+    instance_descriptor.nextInChain                          = &instanceTogglesDesc;
+
+    webgpu_ctx->instance = wgpu::CreateInstance(&instance_descriptor);
     GGML_ASSERT(webgpu_ctx->instance != nullptr);
 
     static ggml_backend_reg reg = {

ggml/src/ggml-webgpu/wgsl-shaders/set_rows.tmpl.wgsl

@@ -1,13 +1,38 @@
+#define(VARIANTS)
+
+[
+  {
+    "SHADER_SUFFIX": "f16_vec",
+    "REPLS": {
+      "TYPE" : "vec4<f32>",
+      "DST_TYPE": "vec4<f16>",
+      "VEC_SIZE": 4
+    }
+  },
+  {
+    "SHADER_SUFFIX": "f16",
+    "REPLS": {
+      "TYPE" : "f32",
+      "DST_TYPE": "f16",
+      "VEC_SIZE": 1
+    }
+  }
+]
+
+#end(VARIANTS)
+
+#define(SHADER)
+
 enable f16;
 
 @group(0) @binding(0)
-var<storage, read_write> src: array<f32>;
+var<storage, read_write> src: array<{{TYPE}}>;
 
 @group(0) @binding(1)
 var<storage, read_write> idx: array<u32>;
 
 @group(0) @binding(2)
-var<storage, read_write> dst: array<f16>;
+var<storage, read_write> dst: array<{{DST_TYPE}}>;
 
 @group(0) @binding(3)
 var<storage, read_write> error: atomic<u32>;
@@ -47,10 +72,14 @@ var<uniform> params: Params;
 override wg_size: u32;
 @compute @workgroup_size(wg_size)
 fn main(@builtin(global_invocation_id) gid: vec3<u32>) {
-    if (gid.x >= params.n_rows * params.ne2 * params.ne3) {
+    if (gid.x >= (params.ne3 * params.ne2 * params.n_rows * params.ne0) / {{VEC_SIZE}}) {
         return;
     }
-    var i = gid.x;
+
+    // getting the row from gid
+    let elems_per_row = params.ne0 / {{VEC_SIZE}};
+    var i = gid.x / elems_per_row;
+
     let i_src3 = i / (params.ne2 * params.n_rows);
 
     i = i % (params.ne2 * params.n_rows);
@@ -75,7 +104,9 @@ fn main(@builtin(global_invocation_id) gid: vec3<u32>) {
     let i_dst_row = params.offset_dst + idx_high_val * params.stride_dst1 + i_src2 * params.stride_dst2 + i_src3 * params.stride_dst3;
     let i_src_row = params.offset_src + i_src1 * params.stride_src1 + i_src2 * params.stride_src2 + i_src3 * params.stride_src3;
 
-    for (var i: u32 = 0; i < params.ne0; i++) {
-      dst[i_dst_row + i] = f16(src[i_src_row + i]);
-    }
+    let col_idx = (gid.x % elems_per_row);
+    dst[i_dst_row/{{VEC_SIZE}} + col_idx] = {{DST_TYPE}}(src[i_src_row/{{VEC_SIZE}} + col_idx]);
 }
+
+#end(SHADER)
+

gguf-py/gguf/constants.py

@@ -426,6 +426,7 @@ class MODEL_ARCH(IntEnum):
     APERTUS          = auto()
     COGVLM           = auto()
     MINIMAXM2        = auto()
+    PANGU_EMBED      = auto()
 
 
 class VISION_PROJECTOR_TYPE(IntEnum):
@@ -793,6 +794,7 @@ MODEL_ARCH_NAMES: dict[MODEL_ARCH, str] = {
     MODEL_ARCH.APERTUS:          "apertus",
     MODEL_ARCH.MINIMAXM2:        "minimax-m2",
     MODEL_ARCH.COGVLM:           "cogvlm",
+    MODEL_ARCH.PANGU_EMBED:      "pangu-embedded",
 }
 
 VISION_PROJECTOR_TYPE_NAMES: dict[VISION_PROJECTOR_TYPE, str] = {
@@ -2958,6 +2960,20 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
         MODEL_TENSOR.VISEXP_UP,
         MODEL_TENSOR.VISEXP_DOWN,
     ],
+    MODEL_ARCH.PANGU_EMBED: [
+        MODEL_TENSOR.TOKEN_EMBD,
+        MODEL_TENSOR.OUTPUT_NORM,
+        MODEL_TENSOR.OUTPUT,
+        MODEL_TENSOR.ATTN_NORM,
+        MODEL_TENSOR.ATTN_Q,
+        MODEL_TENSOR.ATTN_K,
+        MODEL_TENSOR.ATTN_V,
+        MODEL_TENSOR.ATTN_OUT,
+        MODEL_TENSOR.FFN_NORM,
+        MODEL_TENSOR.FFN_GATE,
+        MODEL_TENSOR.FFN_DOWN,
+        MODEL_TENSOR.FFN_UP,
+    ],
     # TODO
 }
 
@@ -3013,6 +3029,10 @@ MODEL_TENSOR_SKIP: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
     MODEL_ARCH.BAILINGMOE: [
         MODEL_TENSOR.ROPE_FREQS,
     ],
+    MODEL_ARCH.PANGU_EMBED: [
+        MODEL_TENSOR.ROPE_FREQS,
+        MODEL_TENSOR.ATTN_ROT_EMBD,
+    ],
 }
 
 #
@@ -3186,6 +3206,7 @@ class VisionProjectorType:
     KIMIVL = "kimivl"
     LIGHTONOCR = "lightonocr"
     COGVLM = "cogvlm"
+    JANUS_PRO = "janus_pro"
 
 
 # Items here are (block size, type size)

gguf-py/gguf/tensor_mapping.py

@@ -1183,6 +1183,7 @@ class TensorNameMap:
             "model.mm_projector.mlp.mlp.{bid}",
             "vision_model.vision_adapter.mlp.fc{bid}", # llama 4
             "mlp1.{bid}", # InternVL
+            "model.aligner.fc1.hidden_layers.{bid}", # Janus Pro
         ),
 
         MODEL_TENSOR.V_MMPROJ_PEG: (
@@ -1291,6 +1292,7 @@ class TensorNameMap:
             "model.vision_tower.encoder.layer.{bid}.attention.projection_layer", # Intern-S1
             "vpm.encoder.layers.{bid}.self_attn.out_proj",
             "model.vision_model.encoder.layers.{bid}.self_attn.out_proj", # SmolVLM
+            "model.vision_model.encoder.layers.{bid}.self_attn.projection_layer", # Janus Pro
             "vision_model.model.layers.{bid}.self_attn.o_proj", # llama4
             "vision_tower.transformer.layers.{bid}.attention.o_proj", # pixtral-hf
             "vision_encoder.transformer.layers.{bid}.attention.wo", # pixtral

include/llama.h

@@ -461,7 +461,10 @@ extern "C" {
     LLAMA_API bool llama_supports_gpu_offload(void);
     LLAMA_API bool llama_supports_rpc        (void);
 
+    // NOTE: After creating a llama_context, it is recommended to query the actual values using these functions
+    //       In some cases the requested values via llama_context_params may differ from the actual values used by the context
     LLAMA_API uint32_t llama_n_ctx      (const struct llama_context * ctx);
+    LLAMA_API uint32_t llama_n_ctx_seq  (const struct llama_context * ctx);
     LLAMA_API uint32_t llama_n_batch    (const struct llama_context * ctx);
     LLAMA_API uint32_t llama_n_ubatch   (const struct llama_context * ctx);
     LLAMA_API uint32_t llama_n_seq_max  (const struct llama_context * ctx);
@@ -585,7 +588,7 @@ extern "C" {
     LLAMA_API int32_t llama_adapter_meta_val_str_by_index(const struct llama_adapter_lora * adapter, int32_t i, char * buf, size_t buf_size);
 
     // Manually free a LoRA adapter
-    // Note: loaded adapters will be free when the associated model is deleted
+    // NOTE: loaded adapters will be free when the associated model is deleted
     LLAMA_API void llama_adapter_lora_free(struct llama_adapter_lora * adapter);
 
     // Get the invocation tokens if the current lora is an alora
@@ -1111,8 +1114,6 @@ extern "C" {
     //        // sample from the logits of the last token in the batch
     //        const llama_token id = llama_sampler_sample(smpl, ctx, -1);
     //
-    //        // accepting the token updates the internal state of certain samplers (e.g. grammar, repetition, etc.)
-    //        llama_sampler_accept(smpl, id);
     //        ...
     //    }
     //

scripts/bench-models.sh

@@ -0,0 +1,74 @@
+#!/usr/bin/env bash
+
+RESULTS="bench-models-results.txt"
+: > "$RESULTS"
+
+ARGS_BB="-c 270336 -npp 512,4096,8192 -npl 1,2,4,8,16,32 -ntg 32"
+ARGS_B="-d 0,4096,8192,16384,32768 -p 2048 -n 32"
+
+QUICK=0
+while (( "$#" )); do
+  case "$1" in
+    --quick) QUICK=1; shift ;;
+    *) shift ;;
+  esac
+done
+
+if (( QUICK )); then
+  ARGS_BB="-c 20480 -npp 512,4096 -npl 1,2,4 -ntg 32"
+  ARGS_B="-d 0 -p 2048 -n 32"
+fi
+
+run_model() {
+  local HFR=$1
+  local HFF=$2
+
+  printf "## ${HFR}\n" | tee -a "$RESULTS"
+  printf "\n" | tee -a "$RESULTS"
+  printf "Model: https://huggingface.co/${HFR}\n" | tee -a "$RESULTS"
+  printf "\n" | tee -a "$RESULTS"
+
+  printf -- "- \`llama-batched-bench\`\n" | tee -a "$RESULTS"
+  printf "\n" | tee -a "$RESULTS"
+
+  ./bin/llama-batched-bench \
+    -hfr "${HFR}" -hff "${HFF}" \
+    -m "${HFF}" -fa 1 -ub 2048 --no-mmap \
+    ${ARGS_BB} | tee -a "$RESULTS"
+
+  printf "\n" | tee -a "$RESULTS"
+
+  printf -- "- \`llama-bench\`\n" | tee -a "$RESULTS"
+  printf "\n" | tee -a "$RESULTS"
+
+  ./bin/llama-bench \
+    -m "${HFF}" -fa 1 -ub 2048 -mmp 0 \
+    ${ARGS_B} | tee -a "$RESULTS"
+
+  printf "\n" | tee -a "$RESULTS"
+
+  printf "\n"
+}
+
+run_model "ggml-org/gpt-oss-20b-GGUF"                       "gpt-oss-20b-mxfp4.gguf"
+run_model "ggml-org/gpt-oss-120b-GGUF"                      "gpt-oss-120b-mxfp4-00001-of-00003.gguf"
+run_model "ggml-org/Qwen3-Coder-30B-A3B-Instruct-Q8_0-GGUF" "qwen3-coder-30b-a3b-instruct-q8_0.gguf"
+run_model "ggml-org/Qwen2.5-Coder-7B-Q8_0-GGUF"             "qwen2.5-coder-7b-q8_0.gguf"
+run_model "ggml-org/gemma-3-4b-it-qat-GGUF"                 "gemma-3-4b-it-qat-Q4_0.gguf"
+
+if [[ -f models-extra.txt ]]; then
+    while read -r HFR HFF; do
+        [[ -z "$HFR" ]] && continue
+        run_model "$HFR" "$HFF"
+    done < models-extra.txt
+fi
+
+printf "\n=====================================\n"
+printf "\n"
+
+cat "$RESULTS"
+
+printf "\n"
+printf "Done! Results are written to $RESULTS\n"
+printf "\n"
+

scripts/sync-ggml.last

@@ -1 +1 @@
-e02fb860ccbba8967905bceff23b677e88105280
+7b6abb2b92fcef35cb01c6ce6ada9bd85306522d

src/CMakeLists.txt

@@ -99,6 +99,7 @@ add_library(llama
             models/openai-moe-iswa.cpp
             models/openelm.cpp
             models/orion.cpp
+            models/pangu-embedded.cpp
             models/phi2.cpp
             models/phi3.cpp
             models/plamo.cpp

src/llama-arch.cpp

@@ -107,6 +107,7 @@ static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
     { LLM_ARCH_APERTUS,          "apertus"          },
     { LLM_ARCH_MINIMAX_M2,       "minimax-m2"       },
     { LLM_ARCH_COGVLM,           "cogvlm"           },
+    { LLM_ARCH_PANGU_EMBED,      "pangu-embedded"   },
     { LLM_ARCH_UNKNOWN,          "(unknown)"        },
 };
 
@@ -2377,6 +2378,23 @@ static const std::map<llm_arch, std::map<llm_tensor, const char *>> LLM_TENSOR_N
             { LLM_TENSOR_FFN_EXP_PROBS_B,    "blk.%d.exp_probs_b" },
         },
     },
+    {
+        LLM_ARCH_PANGU_EMBED,
+        {
+            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
+            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
+            { LLM_TENSOR_OUTPUT,          "output" },
+            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
+            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
+            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
+            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
+            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
+            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
+            { LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },
+            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
+            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
+        },
+    },
     {
         LLM_ARCH_COGVLM,
         {

src/llama-arch.h

@@ -111,6 +111,7 @@ enum llm_arch {
     LLM_ARCH_APERTUS,
     LLM_ARCH_MINIMAX_M2,
     LLM_ARCH_COGVLM,
+    LLM_ARCH_PANGU_EMBED,
     LLM_ARCH_UNKNOWN,
 };
 

src/llama-chat.cpp

@@ -73,6 +73,7 @@ static const std::map<std::string, llm_chat_template> LLM_CHAT_TEMPLATES = {
     { "kimi-k2",           LLM_CHAT_TEMPLATE_KIMI_K2           },
     { "seed_oss",          LLM_CHAT_TEMPLATE_SEED_OSS          },
     { "grok-2",            LLM_CHAT_TEMPLATE_GROK_2            },
+    { "pangu-embedded",    LLM_CHAT_TEMPLATE_PANGU_EMBED       },
 };
 
 llm_chat_template llm_chat_template_from_str(const std::string & name) {
@@ -213,6 +214,8 @@ llm_chat_template llm_chat_detect_template(const std::string & tmpl) {
         return LLM_CHAT_TEMPLATE_SEED_OSS;
     } else if (tmpl_contains("'Assistant: '  + message['content'] + '<|separator|>")) {
         return LLM_CHAT_TEMPLATE_GROK_2;
+    } else if (tmpl_contains(LU8("[unused9]系统：[unused10]"))) {
+        return LLM_CHAT_TEMPLATE_PANGU_EMBED;
     }
     return LLM_CHAT_TEMPLATE_UNKNOWN;
 }
@@ -813,6 +816,35 @@ int32_t llm_chat_apply_template(
         if (add_ass) {
             ss << "Assistant:";
         }
+    }else if (tmpl == LLM_CHAT_TEMPLATE_PANGU_EMBED) {
+        // [unused9]系统：xxx[unused10]
+        // [unused9]用户：xxx[unused10]
+        // [unused9]助手：xxx[unused10]
+        // ...
+        for (size_t i = 0; i < chat.size(); ++i) {
+            const auto & msg = chat[i];
+            const std::string & role = msg->role;
+            const std::string & content = msg->content;
+
+            if (i == 0 && role != "system") {
+                ss << "[unused9]系统：[unused10]";
+            }
+
+            if (role == "system") {
+                ss << "[unused9]系统：" << content << "[unused10]";
+            } else if (role == "user") {
+                ss << "[unused9]用户：" << content << "[unused10]";
+            } else if (role == "assistant") {
+                ss << "[unused9]助手：" << content << "[unused10]";
+            } else if (role == "tool") {
+                ss << "[unused9]工具：" << content << "[unused10]";
+            } else if (role == "function") {
+                ss << "[unused9]方法：" << content << "[unused10]";
+            }
+        }
+        if (add_ass) {
+            ss << "[unused9]助手：";
+        }
     } else {
         // template not supported
         return -1;

src/llama-chat.h

@@ -53,6 +53,7 @@ enum llm_chat_template {
     LLM_CHAT_TEMPLATE_KIMI_K2,
     LLM_CHAT_TEMPLATE_SEED_OSS,
     LLM_CHAT_TEMPLATE_GROK_2,
+    LLM_CHAT_TEMPLATE_PANGU_EMBED,
     LLM_CHAT_TEMPLATE_UNKNOWN,
 };
 

src/llama-context.cpp

@@ -112,11 +112,24 @@ llama_context::llama_context(
         }
     }
 
-    const uint32_t n_ctx_per_seq = cparams.n_ctx / cparams.n_seq_max;
+    if (cparams.kv_unified) {
+        cparams.n_ctx_seq = cparams.n_ctx;
+    } else {
+        cparams.n_ctx_seq = cparams.n_ctx / cparams.n_seq_max;
+
+        if (cparams.n_ctx_seq == 0) {
+            throw std::runtime_error("n_ctx_seq == 0");
+        }
+
+        if (cparams.n_ctx != cparams.n_ctx_seq * cparams.n_seq_max) {
+            cparams.n_ctx =  cparams.n_ctx_seq * cparams.n_seq_max;
+            LLAMA_LOG_WARN("%s: n_ctx is not divisible by n_seq_max - rounding down to %u\n", __func__, cparams.n_ctx);
+        }
+    }
 
     LLAMA_LOG_INFO("%s: n_seq_max     = %u\n",   __func__, cparams.n_seq_max);
     LLAMA_LOG_INFO("%s: n_ctx         = %u\n",   __func__, cparams.n_ctx);
-    LLAMA_LOG_INFO("%s: n_ctx_per_seq = %u\n",   __func__, n_ctx_per_seq);
+    LLAMA_LOG_INFO("%s: n_ctx_seq     = %u\n",   __func__, cparams.n_ctx_seq);
     LLAMA_LOG_INFO("%s: n_batch       = %u\n",   __func__, cparams.n_batch);
     LLAMA_LOG_INFO("%s: n_ubatch      = %u\n",   __func__, cparams.n_ubatch);
     LLAMA_LOG_INFO("%s: causal_attn   = %d\n",   __func__, cparams.causal_attn);
@@ -125,14 +138,14 @@ llama_context::llama_context(
     LLAMA_LOG_INFO("%s: freq_base     = %.1f\n", __func__, cparams.rope_freq_base);
     LLAMA_LOG_INFO("%s: freq_scale    = %g\n",   __func__, cparams.rope_freq_scale);
 
-    if (n_ctx_per_seq < hparams.n_ctx_train) {
-        LLAMA_LOG_WARN("%s: n_ctx_per_seq (%u) < n_ctx_train (%u) -- the full capacity of the model will not be utilized\n",
-                __func__, n_ctx_per_seq, hparams.n_ctx_train);
+    if (cparams.n_ctx_seq < hparams.n_ctx_train) {
+        LLAMA_LOG_WARN("%s: n_ctx_seq (%u) < n_ctx_train (%u) -- the full capacity of the model will not be utilized\n",
+                __func__, cparams.n_ctx_seq, hparams.n_ctx_train);
     }
 
-    if (n_ctx_per_seq > hparams.n_ctx_train) {
-        LLAMA_LOG_WARN("%s: n_ctx_per_seq (%u) > n_ctx_train (%u) -- possible training context overflow\n",
-                __func__, n_ctx_per_seq, hparams.n_ctx_train);
+    if (cparams.n_ctx_seq > hparams.n_ctx_train) {
+        LLAMA_LOG_WARN("%s: n_ctx_seq (%u) > n_ctx_train (%u) -- possible training context overflow\n",
+                __func__, cparams.n_ctx_seq, hparams.n_ctx_train);
     }
 
     if (!hparams.vocab_only) {
@@ -453,8 +466,8 @@ uint32_t llama_context::n_ctx() const {
     return cparams.n_ctx;
 }
 
-uint32_t llama_context::n_ctx_per_seq() const {
-    return cparams.n_ctx / cparams.n_seq_max;
+uint32_t llama_context::n_ctx_seq() const {
+    return cparams.n_ctx_seq;
 }
 
 uint32_t llama_context::n_batch() const {
@@ -2383,6 +2396,10 @@ uint32_t llama_n_ctx(const llama_context * ctx) {
     return ctx->n_ctx();
 }
 
+uint32_t llama_n_ctx_seq(const llama_context * ctx) {
+    return ctx->n_ctx_seq();
+}
+
 uint32_t llama_n_batch(const llama_context * ctx) {
     return ctx->n_batch();
 }

src/llama-context.h

@@ -43,11 +43,11 @@ struct llama_context {
 
     ggml_backend_sched_t get_sched() const;
 
-    uint32_t n_ctx()         const;
-    uint32_t n_ctx_per_seq() const;
-    uint32_t n_batch()       const;
-    uint32_t n_ubatch()      const;
-    uint32_t n_seq_max()     const;
+    uint32_t n_ctx()     const;
+    uint32_t n_ctx_seq() const;
+    uint32_t n_batch()   const;
+    uint32_t n_ubatch()  const;
+    uint32_t n_seq_max() const;
 
     uint32_t n_threads()       const;
     uint32_t n_threads_batch() const;

src/llama-cparams.h

@@ -8,6 +8,7 @@
 
 struct llama_cparams {
     uint32_t n_ctx;           // context size used during inference
+    uint32_t n_ctx_seq;       // context for a single sequence
     uint32_t n_batch;
     uint32_t n_ubatch;
     uint32_t n_seq_max;

src/llama-model.cpp

@@ -2177,6 +2177,15 @@ void llama_model::load_hparams(llama_model_loader & ml) {
                     default: type = LLM_TYPE_UNKNOWN;
                 }
             } break;
+        case LLM_ARCH_PANGU_EMBED:
+            {
+                ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
+                switch (hparams.n_layer) {
+                    case 26: type = LLM_TYPE_1B; break; // openPangu-Embedded-1B-V1.1
+                    case 34: type = LLM_TYPE_7B; break; // openPangu-Embedded-7B-V1.1
+                    default: type = LLM_TYPE_UNKNOWN;
+                }
+            } break;
         default: throw std::runtime_error("unsupported model architecture");
     }
 
@@ -6263,6 +6272,50 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
                         layer.visexp_ffn_up   = create_tensor(tn(LLM_TENSOR_VISEXP_FFN_UP,   "weight", i), {n_embd,   n_ff}, 0);
                     }
                 } break;
+            case LLM_ARCH_PANGU_EMBED:
+                {
+                    tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+
+                    // output
+                    output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
+                    output      = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, TENSOR_NOT_REQUIRED);
+
+                    // if output is NULL, init from the input tok embed
+                    if (output == NULL) {
+                        output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, TENSOR_DUPLICATED);
+                    }
+
+                    for (int i = 0; i < n_layer; ++i) {
+                        auto & layer = layers[i];
+
+                        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
+
+                        // weight tensors
+                        layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "weight", i), {n_embd, n_embd_head_k * n_head}, 0);
+                        layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "weight", i), {n_embd, n_embd_k_gqa}, 0);
+                        layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "weight", i), {n_embd, n_embd_v_gqa}, 0);
+                        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd_head_k * n_head, n_embd}, 0);
+
+                        // bias tensors
+                        layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "bias", i), {n_embd_head_k * n_head}, 0);
+                        layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "bias", i), {n_embd_gqa}, 0);
+                        layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "bias", i), {n_embd_gqa}, 0);
+                        layer.bo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i), {n_embd}, 0);
+
+                        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
+
+                        if (hparams.rope_scaling_type_train == LLAMA_ROPE_SCALING_TYPE_LONGROPE) {
+                            layer.rope_long  = create_tensor(tn(LLM_TENSOR_ROPE_FACTORS_LONG,  "weight", i), {n_rot/2}, TENSOR_NOT_REQUIRED | (i != 0 ? TENSOR_DUPLICATED : 0));
+                            layer.rope_short = create_tensor(tn(LLM_TENSOR_ROPE_FACTORS_SHORT, "weight", i), {n_rot/2}, TENSOR_NOT_REQUIRED | (i != 0 ? TENSOR_DUPLICATED : 0));
+                        } else {
+                            layer.rope_freqs = create_tensor(tn(LLM_TENSOR_ROPE_FREQS, "weight", i), {n_rot/2}, TENSOR_NOT_REQUIRED | (i != 0 ? TENSOR_DUPLICATED : 0));
+                        }
+
+                        layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd,   n_ff}, 0);
+                        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {  n_ff, n_embd}, 0);
+                        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd,   n_ff}, 0);
+                    }
+                } break;
             default:
                 throw std::runtime_error("unknown architecture");
         }
@@ -6712,14 +6765,14 @@ float llama_model::get_rope_freq_scale(const llama_cparams & cparams, int il) co
 }
 
 ggml_tensor * llama_model::get_rope_factors(const llama_cparams & cparams, int il) const {
-    const uint32_t n_ctx_per_seq = cparams.n_ctx / cparams.n_seq_max;
+    const uint32_t n_ctx_seq = cparams.n_ctx_seq;
 
     // choose long/short freq factors based on the context size
     if (layers[il].rope_freqs != nullptr) {
         return layers[il].rope_freqs;
     }
 
-    if (n_ctx_per_seq > hparams.n_ctx_orig_yarn) {
+    if (n_ctx_seq > hparams.n_ctx_orig_yarn) {
         return layers[il].rope_long;
     }
 
@@ -6795,12 +6848,6 @@ llama_memory_i * llama_model::create_memory(const llama_memory_params & params,
                         /* filter_attn       */ std::move(filter_attn),
                         /* filter_recr       */ std::move(filter_recr));
                 } else {
-                    uint32_t n_ctx_per_stream = cparams.n_ctx;
-
-                    if (!cparams.kv_unified) {
-                        n_ctx_per_stream = (cparams.n_ctx + cparams.n_seq_max - 1)/cparams.n_seq_max;
-                    }
-
                     llama_memory_i::layer_reuse_cb reuse = nullptr;
 
                     if (arch == LLM_ARCH_GEMMA3N) {
@@ -6824,7 +6871,7 @@ llama_memory_i * llama_model::create_memory(const llama_memory_params & params,
                                 cparams.offload_kqv,
                                 params.swa_full,
                                 cparams.kv_unified,
-                                n_ctx_per_stream,
+                                cparams.n_ctx_seq,
                                 cparams.n_seq_max,
                                 cparams.n_ubatch,
                                 1,
@@ -6840,7 +6887,7 @@ llama_memory_i * llama_model::create_memory(const llama_memory_params & params,
                                 !cparams.flash_attn,
                                 cparams.offload_kqv,
                                 cparams.kv_unified,
-                                n_ctx_per_stream,
+                                cparams.n_ctx_seq,
                                 cparams.n_seq_max,
                                 1,
                                 hparams.n_swa,
@@ -7266,6 +7313,10 @@ ggml_cgraph * llama_model::build_graph(const llm_graph_params & params) const {
             {
                 llm = std::make_unique<llm_build_cogvlm>(*this, params);
             } break;
+        case LLM_ARCH_PANGU_EMBED:
+            {
+                llm = std::make_unique<llm_build_pangu_embedded>(*this, params);
+            }break;
         default:
             GGML_ABORT("fatal error");
     }
@@ -7485,6 +7536,7 @@ llama_rope_type llama_model_rope_type(const llama_model * model) {
         case LLM_ARCH_APERTUS:
         case LLM_ARCH_MINIMAX_M2:
         case LLM_ARCH_COGVLM:
+        case LLM_ARCH_PANGU_EMBED:
             return LLAMA_ROPE_TYPE_NEOX;
 
         case LLM_ARCH_QWEN2VL:

src/models/gemma2-iswa.cpp

@@ -1,125 +1,125 @@
 #include "models.h"
 
 llm_build_gemma2_iswa::llm_build_gemma2_iswa(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_k;
+    const int64_t n_embd_head = hparams.n_embd_head_k;
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        inpL = ggml_scale(ctx0, inpL, sqrtf(n_embd));
-        cb(inpL, "inp_scaled", -1);
+    inpL = ggml_scale(ctx0, inpL, sqrtf(n_embd));
+    cb(inpL, "inp_scaled", -1);
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        auto * inp_attn = build_attn_inp_kv_iswa();
+    auto * inp_attn = build_attn_inp_kv_iswa();
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            // norm
-            cur = build_norm(inpL,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "attn_norm", il);
+    for (int il = 0; il < n_layer; ++il) {
+        // norm
+        cur = build_norm(inpL,
+                model.layers[il].attn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_norm", il);
 
-            // self-attention
-            {
-                // compute Q and K and RoPE them
-                ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
-                cb(Qcur, "Qcur", il);
+        // self-attention
+        {
+            // compute Q and K and RoPE them
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            cb(Qcur, "Qcur", il);
 
-                ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            cb(Kcur, "Kcur", il);
 
-                ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            cb(Vcur, "Vcur", il);
 
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
 
-                Qcur = ggml_rope_ext(
-                        ctx0, Qcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow);
+            Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow);
 
-                Kcur = ggml_rope_ext(
-                        ctx0, Kcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow);
+            Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow);
 
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
 
-                Qcur = ggml_scale(ctx0, Qcur, hparams.f_attention_scale);
+            Qcur = ggml_scale(ctx0, Qcur, hparams.f_attention_scale);
 
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, NULL,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f, il);
-            }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
-                inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
-            }
-            cur = build_norm(cur,
-                    model.layers[il].attn_post_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "attn_post_norm", il);
-
-            ggml_tensor * sa_out = ggml_add(ctx0, cur, inpL);
-            cb(sa_out, "sa_out", il);
-
-            cur = build_norm(sa_out,
-                    model.layers[il].ffn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "ffn_norm", il);
-
-            // feed-forward network
-            {
-                cur = build_ffn(cur,
-                        model.layers[il].ffn_up,   NULL, NULL,
-                        model.layers[il].ffn_gate, NULL, NULL,
-                        model.layers[il].ffn_down, NULL, NULL,
-                        NULL,
-                        LLM_FFN_GELU, LLM_FFN_PAR, il);
-                cb(cur, "ffn_out", il);
-            }
-            cur = build_norm(cur,
-                    model.layers[il].ffn_post_norm, NULL,
-                    LLM_NORM_RMS, -1);
-            cb(cur, "ffn_post_norm", -1);
-
-            cur = ggml_add(ctx0, cur, sa_out);
-
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, NULL,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f, il);
+        }
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
+            inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
         }
-        cur = inpL;
-
         cur = build_norm(cur,
-                model.output_norm, NULL,
+                model.layers[il].attn_post_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_post_norm", il);
+
+        ggml_tensor * sa_out = ggml_add(ctx0, cur, inpL);
+        cb(sa_out, "sa_out", il);
+
+        cur = build_norm(sa_out,
+                model.layers[il].ffn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "ffn_norm", il);
+
+        // feed-forward network
+        {
+            cur = build_ffn(cur,
+                    model.layers[il].ffn_up,   NULL, NULL,
+                    model.layers[il].ffn_gate, NULL, NULL,
+                    model.layers[il].ffn_down, NULL, NULL,
+                    NULL,
+                    LLM_FFN_GELU, LLM_FFN_PAR, il);
+            cb(cur, "ffn_out", il);
+        }
+        cur = build_norm(cur,
+                model.layers[il].ffn_post_norm, NULL,
                 LLM_NORM_RMS, -1);
+        cb(cur, "ffn_post_norm", -1);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+        cur = ggml_add(ctx0, cur, sa_out);
 
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
 
-        // final logit soft-capping
-        cur = ggml_scale(ctx0, cur, 1.0f / hparams.f_final_logit_softcapping);
-        cur = ggml_tanh(ctx0, cur);
-        cur = ggml_scale(ctx0, cur, hparams.f_final_logit_softcapping);
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
 
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+    cur = build_norm(cur,
+            model.output_norm, NULL,
+            LLM_NORM_RMS, -1);
 
-        ggml_build_forward_expand(gf, cur);
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+
+    // final logit soft-capping
+    cur = ggml_scale(ctx0, cur, 1.0f / hparams.f_final_logit_softcapping);
+    cur = ggml_tanh(ctx0, cur);
+    cur = ggml_scale(ctx0, cur, hparams.f_final_logit_softcapping);
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/gemma3-iswa.cpp

@@ -1,131 +1,131 @@
 #include "models.h"
 
 llm_build_gemma3_iswa::llm_build_gemma3_iswa(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_k;
+    const int64_t n_embd_head = hparams.n_embd_head_k;
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        // important: do not normalize weights for raw embeddings input (i.e. encoded image emdeddings)
-        if (ubatch.token) {
-            inpL = ggml_scale(ctx0, inpL, sqrtf(n_embd));
-            cb(inpL, "inp_scaled", -1);
+    // important: do not normalize weights for raw embeddings input (i.e. encoded image emdeddings)
+    if (ubatch.token) {
+        inpL = ggml_scale(ctx0, inpL, sqrtf(n_embd));
+        cb(inpL, "inp_scaled", -1);
+    }
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
+
+    // TODO: is causal == true correct? might need some changes
+    auto * inp_attn = build_attn_inp_kv_iswa();
+
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
+
+    for (int il = 0; il < n_layer; ++il) {
+        const float freq_base_l  = model.get_rope_freq_base (cparams, il);
+        const float freq_scale_l = model.get_rope_freq_scale(cparams, il);
+
+        // norm
+        cur = build_norm(inpL, model.layers[il].attn_norm, NULL, LLM_NORM_RMS, il);
+        cb(cur, "attn_norm", il);
+
+        // self-attention
+        {
+            // compute Q and K and RoPE them
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            cb(Qcur, "Qcur", il);
+
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            cb(Kcur, "Kcur", il);
+
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            cb(Vcur, "Vcur", il);
+
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
+
+            Qcur = build_norm(Qcur, model.layers[il].attn_q_norm, NULL, LLM_NORM_RMS, il);
+            cb(Qcur, "Qcur_normed", il);
+
+            Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base_l, freq_scale_l,
+                    ext_factor, attn_factor, beta_fast, beta_slow);
+
+            Kcur = build_norm(Kcur, model.layers[il].attn_k_norm, NULL, LLM_NORM_RMS, il);
+            cb(Kcur, "Kcur_normed", il);
+
+            Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base_l, freq_scale_l,
+                    ext_factor, attn_factor, beta_fast, beta_slow);
+
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
+
+            // ref: https://github.com/google/gemma_pytorch/blob/014acb7ac4563a5f77c76d7ff98f31b568c16508/gemma/model.py#L315
+            Qcur = ggml_scale(ctx0, Qcur, hparams.f_attention_scale);
+
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, NULL,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f, il);
         }
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
-
-        // TODO: is causal == true correct? might need some changes
-        auto * inp_attn = build_attn_inp_kv_iswa();
-
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
-
-        for (int il = 0; il < n_layer; ++il) {
-            const float freq_base_l  = model.get_rope_freq_base (cparams, il);
-            const float freq_scale_l = model.get_rope_freq_scale(cparams, il);
-
-            // norm
-            cur = build_norm(inpL, model.layers[il].attn_norm, NULL, LLM_NORM_RMS, il);
-            cb(cur, "attn_norm", il);
-
-            // self-attention
-            {
-                // compute Q and K and RoPE them
-                ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
-                cb(Qcur, "Qcur", il);
-
-                ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
-
-                ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
-
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
-
-                Qcur = build_norm(Qcur, model.layers[il].attn_q_norm, NULL, LLM_NORM_RMS, il);
-                cb(Qcur, "Qcur_normed", il);
-
-                Qcur = ggml_rope_ext(
-                        ctx0, Qcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base_l, freq_scale_l,
-                        ext_factor, attn_factor, beta_fast, beta_slow);
-
-                Kcur = build_norm(Kcur, model.layers[il].attn_k_norm, NULL, LLM_NORM_RMS, il);
-                cb(Kcur, "Kcur_normed", il);
-
-                Kcur = ggml_rope_ext(
-                        ctx0, Kcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base_l, freq_scale_l,
-                        ext_factor, attn_factor, beta_fast, beta_slow);
-
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
-
-                // ref: https://github.com/google/gemma_pytorch/blob/014acb7ac4563a5f77c76d7ff98f31b568c16508/gemma/model.py#L315
-                Qcur = ggml_scale(ctx0, Qcur, hparams.f_attention_scale);
-
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, NULL,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f, il);
-            }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
-                inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
-            }
-            cur = build_norm(cur,
-                    model.layers[il].attn_post_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "attn_post_norm", il);
-
-            ggml_tensor * sa_out = ggml_add(ctx0, cur, inpL);
-            cb(sa_out, "sa_out", il);
-
-            cur = build_norm(sa_out,
-                    model.layers[il].ffn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "ffn_norm", il);
-
-            // feed-forward network
-            {
-                cur = build_ffn(cur,
-                        model.layers[il].ffn_up,   NULL, NULL,
-                        model.layers[il].ffn_gate, NULL, NULL,
-                        model.layers[il].ffn_down, NULL, NULL,
-                        NULL,
-                        LLM_FFN_GELU, LLM_FFN_PAR, il);
-                cb(cur, "ffn_out", il);
-            }
-            cur = build_norm(cur,
-                    model.layers[il].ffn_post_norm, NULL,
-                    LLM_NORM_RMS, -1);
-            cb(cur, "ffn_post_norm", -1);
-
-            cur = ggml_add(ctx0, cur, sa_out);
-
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
+            inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
         }
-        cur = inpL;
-
         cur = build_norm(cur,
-                model.output_norm, NULL,
+                model.layers[il].attn_post_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_post_norm", il);
+
+        ggml_tensor * sa_out = ggml_add(ctx0, cur, inpL);
+        cb(sa_out, "sa_out", il);
+
+        cur = build_norm(sa_out,
+                model.layers[il].ffn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "ffn_norm", il);
+
+        // feed-forward network
+        {
+            cur = build_ffn(cur,
+                    model.layers[il].ffn_up,   NULL, NULL,
+                    model.layers[il].ffn_gate, NULL, NULL,
+                    model.layers[il].ffn_down, NULL, NULL,
+                    NULL,
+                    LLM_FFN_GELU, LLM_FFN_PAR, il);
+            cb(cur, "ffn_out", il);
+        }
+        cur = build_norm(cur,
+                model.layers[il].ffn_post_norm, NULL,
                 LLM_NORM_RMS, -1);
+        cb(cur, "ffn_post_norm", -1);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+        cur = ggml_add(ctx0, cur, sa_out);
 
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
 
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
 
-        ggml_build_forward_expand(gf, cur);
+    cur = build_norm(cur,
+            model.output_norm, NULL,
+            LLM_NORM_RMS, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/glm4-moe.cpp

@@ -1,153 +1,153 @@
 #include "models.h"
 
 llm_build_glm4_moe::llm_build_glm4_moe(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_head = hparams.n_embd_head_v;
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        auto * inp_attn = build_attn_inp_kv();
+    auto * inp_attn = build_attn_inp_kv();
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        // Only process up to last layer (skip final NextN layer)
-        // Final layer tensors are loaded but not processed in forward pass
-        const int n_transformer_layers = n_layer - hparams.nextn_predict_layers;
-        for (int il = 0; il < n_transformer_layers; ++il) {
-            ggml_tensor * inpSA = inpL;
+    // Only process up to last layer (skip final NextN layer)
+    // Final layer tensors are loaded but not processed in forward pass
+    const int n_transformer_layers = n_layer - hparams.nextn_predict_layers;
+    for (int il = 0; il < n_transformer_layers; ++il) {
+        ggml_tensor * inpSA = inpL;
 
-            // Pre-attention norm
-            cur = build_norm(inpL, model.layers[il].attn_norm, NULL, LLM_NORM_RMS, il);
-            cb(cur, "attn_norm", il);
+        // Pre-attention norm
+        cur = build_norm(inpL, model.layers[il].attn_norm, NULL, LLM_NORM_RMS, il);
+        cb(cur, "attn_norm", il);
 
-            // self-attention
-            {
-                ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
-                if (model.layers[il].bq) {
-                    Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
-                }
-                cb(Qcur, "Qcur", il);
-
-                ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
-                if (model.layers[il].bk) {
-                    Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
-                }
-                cb(Kcur, "Kcur", il);
-
-                ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
-                if (model.layers[il].bv) {
-                    Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
-                }
-                cb(Vcur, "Vcur", il);
-
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
-
-                // Apply Q/K norm if available (GLM-4.5 355B variant)
-                if (model.layers[il].attn_q_norm) {
-                    Qcur = build_norm(Qcur, model.layers[il].attn_q_norm, NULL, LLM_NORM_RMS, il);
-                    cb(Qcur, "Qcur_normed", il);
-                }
-                if (model.layers[il].attn_k_norm) {
-                    Kcur = build_norm(Kcur, model.layers[il].attn_k_norm, NULL, LLM_NORM_RMS, il);
-                    cb(Kcur, "Kcur_normed", il);
-                }
-                Qcur = ggml_rope_ext(
-                        ctx0, Qcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-
-                Kcur = ggml_rope_ext(
-                        ctx0, Kcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
-
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, NULL,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
+        // self-attention
+        {
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            if (model.layers[il].bq) {
+                Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
             }
-            if (il == n_transformer_layers - 1 && inp_out_ids) {
-                cur   = ggml_get_rows(ctx0, cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+            cb(Qcur, "Qcur", il);
+
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            if (model.layers[il].bk) {
+                Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
             }
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
+            cb(Kcur, "Kcur", il);
 
-            // Post-attention norm
-            cur = build_norm(ffn_inp, model.layers[il].attn_post_norm, NULL, LLM_NORM_RMS, il);
-            cb(cur, "post_attn_norm", il);
-
-            // Check if this is a dense layer (n_layer_dense_lead=1, so layer 0 is dense)
-            if (static_cast<uint32_t>(il) < hparams.n_layer_dense_lead) {
-                // Dense FFN layer
-                cur = build_ffn(cur,
-                        model.layers[il].ffn_up,   NULL, NULL,
-                        model.layers[il].ffn_gate, NULL, NULL,
-                        model.layers[il].ffn_down, NULL, NULL,
-                        NULL,
-                        LLM_FFN_SILU, LLM_FFN_PAR, il);
-                cb(cur, "ffn_out", il);
-            } else {
-                // Process routed experts using existing MoE infrastructure
-                ggml_tensor * routed_out = build_moe_ffn(cur,
-                        model.layers[il].ffn_gate_inp,
-                        model.layers[il].ffn_up_exps,
-                        model.layers[il].ffn_gate_exps,
-                        model.layers[il].ffn_down_exps,
-                        model.layers[il].ffn_exp_probs_b,
-                        n_expert, n_expert_used,
-                        LLM_FFN_SILU, hparams.expert_weights_norm,
-                        true, hparams.expert_weights_scale,
-                        (llama_expert_gating_func_type) hparams.expert_gating_func,
-                        il);
-                cb(routed_out, "ffn_moe_out", il);
-
-                // Process shared expert on original input
-                ggml_tensor * shared_out = build_ffn(cur,
-                        model.layers[il].ffn_up_shexp,   NULL, NULL,
-                        model.layers[il].ffn_gate_shexp, NULL, NULL,
-                        model.layers[il].ffn_down_shexp, NULL, NULL,
-                        NULL,
-                        LLM_FFN_SILU, LLM_FFN_PAR, il);
-                cb(shared_out, "ffn_shexp_out", il);
-
-                // Final output: routed_output + shared_output
-                cur = ggml_add(ctx0, routed_out, shared_out);
-                cb(cur, "ffn_out", il);
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            if (model.layers[il].bv) {
+                Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
             }
-            cur = ggml_add(ctx0, cur, ffn_inp);
+            cb(Vcur, "Vcur", il);
 
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
 
-            // input for next layer
-            inpL = cur;
+            // Apply Q/K norm if available (GLM-4.5 355B variant)
+            if (model.layers[il].attn_q_norm) {
+                Qcur = build_norm(Qcur, model.layers[il].attn_q_norm, NULL, LLM_NORM_RMS, il);
+                cb(Qcur, "Qcur_normed", il);
+            }
+            if (model.layers[il].attn_k_norm) {
+                Kcur = build_norm(Kcur, model.layers[il].attn_k_norm, NULL, LLM_NORM_RMS, il);
+                cb(Kcur, "Kcur_normed", il);
+            }
+            Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
+
+            Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
+
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
+
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, NULL,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
         }
-        cur = inpL;
-        cur = build_norm(cur, model.output_norm, NULL, LLM_NORM_RMS, -1);
+        if (il == n_transformer_layers - 1 && inp_out_ids) {
+            cur   = ggml_get_rows(ctx0, cur, inp_out_ids);
+            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+        }
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+        cb(ffn_inp, "ffn_inp", il);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+        // Post-attention norm
+        cur = build_norm(ffn_inp, model.layers[il].attn_post_norm, NULL, LLM_NORM_RMS, il);
+        cb(cur, "post_attn_norm", il);
 
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
+        // Check if this is a dense layer (n_layer_dense_lead=1, so layer 0 is dense)
+        if (static_cast<uint32_t>(il) < hparams.n_layer_dense_lead) {
+            // Dense FFN layer
+            cur = build_ffn(cur,
+                    model.layers[il].ffn_up,   NULL, NULL,
+                    model.layers[il].ffn_gate, NULL, NULL,
+                    model.layers[il].ffn_down, NULL, NULL,
+                    NULL,
+                    LLM_FFN_SILU, LLM_FFN_PAR, il);
+            cb(cur, "ffn_out", il);
+        } else {
+            // Process routed experts using existing MoE infrastructure
+            ggml_tensor * routed_out = build_moe_ffn(cur,
+                    model.layers[il].ffn_gate_inp,
+                    model.layers[il].ffn_up_exps,
+                    model.layers[il].ffn_gate_exps,
+                    model.layers[il].ffn_down_exps,
+                    model.layers[il].ffn_exp_probs_b,
+                    n_expert, n_expert_used,
+                    LLM_FFN_SILU, hparams.expert_weights_norm,
+                    true, hparams.expert_weights_scale,
+                    (llama_expert_gating_func_type) hparams.expert_gating_func,
+                    il);
+            cb(routed_out, "ffn_moe_out", il);
 
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+            // Process shared expert on original input
+            ggml_tensor * shared_out = build_ffn(cur,
+                    model.layers[il].ffn_up_shexp,   NULL, NULL,
+                    model.layers[il].ffn_gate_shexp, NULL, NULL,
+                    model.layers[il].ffn_down_shexp, NULL, NULL,
+                    NULL,
+                    LLM_FFN_SILU, LLM_FFN_PAR, il);
+            cb(shared_out, "ffn_shexp_out", il);
 
-        ggml_build_forward_expand(gf, cur);
+            // Final output: routed_output + shared_output
+            cur = ggml_add(ctx0, routed_out, shared_out);
+            cb(cur, "ffn_out", il);
+        }
+        cur = ggml_add(ctx0, cur, ffn_inp);
+
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
+
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
+    cur = build_norm(cur, model.output_norm, NULL, LLM_NORM_RMS, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/grok.cpp

@@ -1,160 +1,159 @@
 #include "models.h"
 
 llm_build_grok::llm_build_grok(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_head = hparams.n_embd_head_v;
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        auto * inp_attn = build_attn_inp_kv();
+    auto * inp_attn = build_attn_inp_kv();
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            ggml_tensor * inpSA = inpL;
+    for (int il = 0; il < n_layer; ++il) {
+        ggml_tensor * inpSA = inpL;
 
-            // norm
-            cur = build_norm(inpL,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "attn_norm", il);
+        // norm
+        cur = build_norm(inpL,
+                model.layers[il].attn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_norm", il);
 
-
-            // self-attention
-            {
-                // compute Q and K and RoPE them
-                ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+        // self-attention
+        {
+            // compute Q and K and RoPE them
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            cb(Qcur, "Qcur", il);
+            if (model.layers[il].bq) {
+                Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
                 cb(Qcur, "Qcur", il);
-                if (model.layers[il].bq) {
-                    Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
-                    cb(Qcur, "Qcur", il);
-                }
-                ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            }
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            cb(Kcur, "Kcur", il);
+            if (model.layers[il].bk) {
+                Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
                 cb(Kcur, "Kcur", il);
-                if (model.layers[il].bk) {
-                    Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
-                    cb(Kcur, "Kcur", il);
-                }
-                ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            }
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            cb(Vcur, "Vcur", il);
+            if (model.layers[il].bv) {
+                Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
                 cb(Vcur, "Vcur", il);
-                if (model.layers[il].bv) {
-                    Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
-                    cb(Vcur, "Vcur", il);
-                }
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
-
-                Qcur = ggml_rope_ext(
-                        ctx0, Qcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-
-                Kcur = ggml_rope_ext(
-                        ctx0, Kcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
-
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f, il);
             }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
-            cur = build_norm(cur,
-                    model.layers[il].attn_out_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "attn_out_norm", il);
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
 
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
+            Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-            // feed-forward network
-            cur = build_norm(ffn_inp,
-                    model.layers[il].ffn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "ffn_norm", il);
+            Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-            // MoE branch
-            ggml_tensor * moe_out = build_moe_ffn(cur,
-                    model.layers[il].ffn_gate_inp,
-                    model.layers[il].ffn_up_exps,
-                    model.layers[il].ffn_gate_exps,
-                    model.layers[il].ffn_down_exps,
-                    nullptr,
-                    n_expert, n_expert_used,
-                    LLM_FFN_GELU, true,
-                    false, 0.0,
-                    LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
-                    il);
-            cb(moe_out, "ffn_moe_out", il);
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
 
-            if (model.layers[il].ffn_up) {
-                ggml_tensor * ffn_out = build_ffn(cur,
-                        model.layers[il].ffn_up,   NULL, NULL,
-                        model.layers[il].ffn_gate, NULL, NULL,
-                        model.layers[il].ffn_down, NULL, NULL,
-                        NULL,
-                        LLM_FFN_GELU, LLM_FFN_PAR, il);
-                cb(ffn_out, "ffn_out", il);
-
-                cur = ggml_scale(ctx0, ggml_add(ctx0, ffn_out, moe_out), std::sqrt(2) / 2);
-                cb(cur, "ffn_out", il);
-            } else {
-                cur = moe_out;
-            }
-            cur = build_norm(cur,
-                    model.layers[il].ffn_post_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "ffn_post_norm", il);
-
-            cur = ggml_add(ctx0, cur, ffn_inp);
-            cb(cur, "ffn_out", il);
-
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, model.layers[il].bo,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f, il);
+        }
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
         }
-        cur = inpL;
-
         cur = build_norm(cur,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, -1);
+                model.layers[il].attn_out_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_out_norm", il);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+        cb(ffn_inp, "ffn_inp", il);
 
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
+        // feed-forward network
+        cur = build_norm(ffn_inp,
+                model.layers[il].ffn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "ffn_norm", il);
 
-        cur = ggml_scale(ctx0, cur, hparams.f_logit_scale);
+        // MoE branch
+        ggml_tensor * moe_out = build_moe_ffn(cur,
+                model.layers[il].ffn_gate_inp,
+                model.layers[il].ffn_up_exps,
+                model.layers[il].ffn_gate_exps,
+                model.layers[il].ffn_down_exps,
+                nullptr,
+                n_expert, n_expert_used,
+                LLM_FFN_GELU, true,
+                false, 0.0,
+                LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
+                il);
+        cb(moe_out, "ffn_moe_out", il);
 
-        // final logit soft-capping
-        if (hparams.f_final_logit_softcapping) {
-            cur = ggml_scale(ctx0, cur, 1.0f / hparams.f_final_logit_softcapping);
-            cur = ggml_tanh(ctx0, cur);
-            cur = ggml_scale(ctx0, cur, hparams.f_final_logit_softcapping);
+        if (model.layers[il].ffn_up) {
+            ggml_tensor * ffn_out = build_ffn(cur,
+                    model.layers[il].ffn_up,   NULL, NULL,
+                    model.layers[il].ffn_gate, NULL, NULL,
+                    model.layers[il].ffn_down, NULL, NULL,
+                    NULL,
+                    LLM_FFN_GELU, LLM_FFN_PAR, il);
+            cb(ffn_out, "ffn_out", il);
+
+            cur = ggml_scale(ctx0, ggml_add(ctx0, ffn_out, moe_out), std::sqrt(2) / 2);
+            cb(cur, "ffn_out", il);
+        } else {
+            cur = moe_out;
         }
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+        cur = build_norm(cur,
+                model.layers[il].ffn_post_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "ffn_post_norm", il);
 
-        ggml_build_forward_expand(gf, cur);
+        cur = ggml_add(ctx0, cur, ffn_inp);
+        cb(cur, "ffn_out", il);
+
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
+
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
+
+    cur = build_norm(cur,
+            model.output_norm, NULL,
+            LLM_NORM_RMS, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+
+    cur = ggml_scale(ctx0, cur, hparams.f_logit_scale);
+
+    // final logit soft-capping
+    if (hparams.f_final_logit_softcapping) {
+        cur = ggml_scale(ctx0, cur, 1.0f / hparams.f_final_logit_softcapping);
+        cur = ggml_tanh(ctx0, cur);
+        cur = ggml_scale(ctx0, cur, hparams.f_final_logit_softcapping);
+    }
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/hunyuan-dense.cpp

@@ -1,132 +1,132 @@
 #include "models.h"
 
 llm_build_hunyuan_dense::llm_build_hunyuan_dense(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_head = hparams.n_embd_head_v;
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        auto * inp_attn = build_attn_inp_kv();
+    auto * inp_attn = build_attn_inp_kv();
 
-        const float kq_scale = 1.0f / sqrtf(float(n_embd_head));
+    const float kq_scale = 1.0f / sqrtf(float(n_embd_head));
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            ggml_tensor * inpSA = inpL;
+    for (int il = 0; il < n_layer; ++il) {
+        ggml_tensor * inpSA = inpL;
 
-            // norm
-            cur = build_norm(inpL,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "attn_norm", il);
-            // self-attention
-            {
-                // rope freq factors for llama3; may return nullptr for llama2 and other models
-                ggml_tensor * rope_factors = model.get_rope_factors(cparams, il);
+        // norm
+        cur = build_norm(inpL,
+                model.layers[il].attn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_norm", il);
+        // self-attention
+        {
+            // rope freq factors for llama3; may return nullptr for llama2 and other models
+            ggml_tensor * rope_factors = model.get_rope_factors(cparams, il);
 
-                // compute Q and K and RoPE them
-                ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            // compute Q and K and RoPE them
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            cb(Qcur, "Qcur", il);
+            if (model.layers[il].bq) {
+                Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
                 cb(Qcur, "Qcur", il);
-                if (model.layers[il].bq) {
-                    Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
-                    cb(Qcur, "Qcur", il);
-                }
-                ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
-                if (model.layers[il].bk) {
-                    Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
-                    cb(Kcur, "Kcur", il);
-                }
-                ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
-                if (model.layers[il].bv) {
-                    Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
-                    cb(Vcur, "Vcur", il);
-                }
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
-
-                Qcur = ggml_rope_ext(
-                         ctx0, Qcur, inp_pos, rope_factors,
-                         n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                         ext_factor, attn_factor, beta_fast, beta_slow
-                         );
-
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
-
-                Kcur = ggml_rope_ext(
-                         ctx0, Kcur, inp_pos, rope_factors,
-                         n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                         ext_factor, attn_factor, beta_fast, beta_slow
-                         );
-
-                Kcur = build_norm(Kcur,
-                         model.layers[il].attn_k_norm, nullptr,
-                         LLM_NORM_RMS, il);
-                cb(Kcur, "Kcur_norm", il);
-
-                Qcur = build_norm(Qcur,
-                         model.layers[il].attn_q_norm, nullptr,
-                         LLM_NORM_RMS, il);
-                cb(Qcur, "Qcur_norm", il);
-
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, kq_scale, il);
-                cb(cur, "attn_out", il);
             }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            cb(Kcur, "Kcur", il);
+            if (model.layers[il].bk) {
+                Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
+                cb(Kcur, "Kcur", il);
             }
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            cb(Vcur, "Vcur", il);
+            if (model.layers[il].bv) {
+                Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
+                cb(Vcur, "Vcur", il);
+            }
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
 
-            cur = build_norm(ffn_inp,
-                    model.layers[il].ffn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "ffn_norm", il);
-            // feed-forward network (non-MoE)
-            ggml_tensor * cur_mlp = build_ffn(cur,
-                        model.layers[il].ffn_up,   NULL, NULL,
-                        model.layers[il].ffn_gate, NULL, NULL,
-                        model.layers[il].ffn_down, NULL, NULL,
-                        NULL,
-                        LLM_FFN_SILU, LLM_FFN_PAR, il);
-            cb(cur_mlp, "ffn_out", il);
+            Qcur = ggml_rope_ext(
+                        ctx0, Qcur, inp_pos, rope_factors,
+                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                        ext_factor, attn_factor, beta_fast, beta_slow
+                        );
 
-            cur = ggml_add(ctx0, cur_mlp, ffn_inp);
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
 
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
+            Kcur = ggml_rope_ext(
+                        ctx0, Kcur, inp_pos, rope_factors,
+                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                        ext_factor, attn_factor, beta_fast, beta_slow
+                        );
 
-            // input for next layer
-            inpL = cur;
+            Kcur = build_norm(Kcur,
+                        model.layers[il].attn_k_norm, nullptr,
+                        LLM_NORM_RMS, il);
+            cb(Kcur, "Kcur_norm", il);
+
+            Qcur = build_norm(Qcur,
+                        model.layers[il].attn_q_norm, nullptr,
+                        LLM_NORM_RMS, il);
+            cb(Qcur, "Qcur_norm", il);
+
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, model.layers[il].bo,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, kq_scale, il);
+            cb(cur, "attn_out", il);
         }
-        cur = inpL;
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+        }
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+        cb(ffn_inp, "ffn_inp", il);
 
-        cur = build_norm(cur,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, -1);
+        cur = build_norm(ffn_inp,
+                model.layers[il].ffn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "ffn_norm", il);
+        // feed-forward network (non-MoE)
+        ggml_tensor * cur_mlp = build_ffn(cur,
+                    model.layers[il].ffn_up,   NULL, NULL,
+                    model.layers[il].ffn_gate, NULL, NULL,
+                    model.layers[il].ffn_down, NULL, NULL,
+                    NULL,
+                    LLM_FFN_SILU, LLM_FFN_PAR, il);
+        cb(cur_mlp, "ffn_out", il);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+        cur = ggml_add(ctx0, cur_mlp, ffn_inp);
 
-        ggml_build_forward_expand(gf, cur);
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
+
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
+
+    cur = build_norm(cur,
+            model.output_norm, NULL,
+            LLM_NORM_RMS, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/hunyuan-moe.cpp

@@ -1,154 +1,154 @@
 #include "models.h"
 
 llm_build_hunyuan_moe::llm_build_hunyuan_moe(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_head = hparams.n_embd_head_v;
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        auto * inp_attn = build_attn_inp_kv();
+    auto * inp_attn = build_attn_inp_kv();
 
-        const float kq_scale = 1.0f / sqrtf(float(n_embd_head));
+    const float kq_scale = 1.0f / sqrtf(float(n_embd_head));
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            ggml_tensor * inpSA = inpL;
+    for (int il = 0; il < n_layer; ++il) {
+        ggml_tensor * inpSA = inpL;
 
-            // norm
-            cur = build_norm(inpL,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "attn_norm", il);
-
-            // self-attention
-            {
-                // rope freq factors for llama3; may return nullptr for llama2 and other models
-                ggml_tensor * rope_factors = model.get_rope_factors(cparams, il);
-
-                // compute Q and K and RoPE them
-                ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
-                cb(Qcur, "Qcur", il);
-                if (model.layers[il].bq) {
-                    Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
-                    cb(Qcur, "Qcur", il);
-                }
-                ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
-                if (model.layers[il].bk) {
-                    Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
-                    cb(Kcur, "Kcur", il);
-                }
-                ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
-                if (model.layers[il].bv) {
-                    Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
-                    cb(Vcur, "Vcur", il);
-                }
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
-
-                Qcur = ggml_rope_ext(
-                        ctx0, Qcur, inp_pos, rope_factors,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
-
-                Kcur = ggml_rope_ext(
-                        ctx0, Kcur, inp_pos, rope_factors,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-
-                Kcur = build_norm(Kcur,
-                        model.layers[il].attn_k_norm, nullptr,
-                        LLM_NORM_RMS, il);
-                cb(Kcur, "Kcur_norm", il);
-
-                Qcur = build_norm(Qcur,
-                        model.layers[il].attn_q_norm, nullptr,
-                        LLM_NORM_RMS, il);
-                cb(Qcur, "Qcur_norm", il);
-
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, kq_scale, il);
-                cb(cur, "attn_out", il);
-            }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
-
-            cur = build_norm(ffn_inp,
-                model.layers[il].ffn_norm, NULL,
+        // norm
+        cur = build_norm(inpL,
+                model.layers[il].attn_norm, NULL,
                 LLM_NORM_RMS, il);
-            cb(cur, "ffn_norm", il);
+        cb(cur, "attn_norm", il);
 
-            // feed-forward network (non-MoE)
-            ggml_tensor * cur_mlp = build_ffn(cur,
-                    model.layers[il].ffn_up_shexp,   NULL, NULL,
-                    model.layers[il].ffn_gate_shexp, NULL, NULL,
-                    model.layers[il].ffn_down_shexp, NULL, NULL,
-                    NULL,
-                    LLM_FFN_SILU, LLM_FFN_PAR, il);
-            cb(cur_mlp, "ffn_mlp", il);
+        // self-attention
+        {
+            // rope freq factors for llama3; may return nullptr for llama2 and other models
+            ggml_tensor * rope_factors = model.get_rope_factors(cparams, il);
 
-            // MoE branch
-            ggml_tensor * cur_moe = build_moe_ffn(cur,
-                    model.layers[il].ffn_gate_inp,
-                    model.layers[il].ffn_up_exps,
-                    model.layers[il].ffn_gate_exps,
-                    model.layers[il].ffn_down_exps,
-                    nullptr,
-                    n_expert, n_expert_used,
-                    LLM_FFN_SILU,
-                    true, // norm_topk_prob
-                    false,
-                    0.0,
-                    LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
-                    il);
-            cb(cur_moe, "ffn_moe_out", il);
+            // compute Q and K and RoPE them
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            cb(Qcur, "Qcur", il);
+            if (model.layers[il].bq) {
+                Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
+                cb(Qcur, "Qcur", il);
+            }
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            cb(Kcur, "Kcur", il);
+            if (model.layers[il].bk) {
+                Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
+                cb(Kcur, "Kcur", il);
+            }
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            cb(Vcur, "Vcur", il);
+            if (model.layers[il].bv) {
+                Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
+                cb(Vcur, "Vcur", il);
+            }
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
 
-            ggml_tensor * ffn_out = ggml_add(ctx0, cur_moe, cur_mlp);
-            cb(ffn_out, "ffn_out", il);
+            Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, rope_factors,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-            cur = ggml_add(ctx0, ffn_out, ffn_inp);
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
 
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
+            Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, rope_factors,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-            // input for next layer
-            inpL = cur;
+            Kcur = build_norm(Kcur,
+                    model.layers[il].attn_k_norm, nullptr,
+                    LLM_NORM_RMS, il);
+            cb(Kcur, "Kcur_norm", il);
+
+            Qcur = build_norm(Qcur,
+                    model.layers[il].attn_q_norm, nullptr,
+                    LLM_NORM_RMS, il);
+            cb(Qcur, "Qcur_norm", il);
+
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, model.layers[il].bo,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, kq_scale, il);
+            cb(cur, "attn_out", il);
         }
-        cur = inpL;
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+        }
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+        cb(ffn_inp, "ffn_inp", il);
 
-        cur = build_norm(cur,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, -1);
+        cur = build_norm(ffn_inp,
+            model.layers[il].ffn_norm, NULL,
+            LLM_NORM_RMS, il);
+        cb(cur, "ffn_norm", il);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+        // feed-forward network (non-MoE)
+        ggml_tensor * cur_mlp = build_ffn(cur,
+                model.layers[il].ffn_up_shexp,   NULL, NULL,
+                model.layers[il].ffn_gate_shexp, NULL, NULL,
+                model.layers[il].ffn_down_shexp, NULL, NULL,
+                NULL,
+                LLM_FFN_SILU, LLM_FFN_PAR, il);
+        cb(cur_mlp, "ffn_mlp", il);
 
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+        // MoE branch
+        ggml_tensor * cur_moe = build_moe_ffn(cur,
+                model.layers[il].ffn_gate_inp,
+                model.layers[il].ffn_up_exps,
+                model.layers[il].ffn_gate_exps,
+                model.layers[il].ffn_down_exps,
+                nullptr,
+                n_expert, n_expert_used,
+                LLM_FFN_SILU,
+                true, // norm_topk_prob
+                false,
+                0.0,
+                LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
+                il);
+        cb(cur_moe, "ffn_moe_out", il);
 
-        ggml_build_forward_expand(gf, cur);
+        ggml_tensor * ffn_out = ggml_add(ctx0, cur_moe, cur_mlp);
+        cb(ffn_out, "ffn_out", il);
+
+        cur = ggml_add(ctx0, ffn_out, ffn_inp);
+
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
+
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
+
+    cur = build_norm(cur,
+            model.output_norm, NULL,
+            LLM_NORM_RMS, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/internlm2.cpp

@@ -1,121 +1,120 @@
 #include "models.h"
 
-
 llm_build_internlm2::llm_build_internlm2(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_head = hparams.n_embd_head_v;
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        auto * inp_attn = build_attn_inp_kv();
+    auto * inp_attn = build_attn_inp_kv();
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            ggml_tensor * inpSA = inpL;
+    for (int il = 0; il < n_layer; ++il) {
+        ggml_tensor * inpSA = inpL;
 
-            // norm
-            cur = build_norm(inpL,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "attn_norm", il);
+        // norm
+        cur = build_norm(inpL,
+                model.layers[il].attn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_norm", il);
 
-            // self-attention
-            {
-                // compute Q and K and RoPE them
-                ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+        // self-attention
+        {
+            // compute Q and K and RoPE them
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            cb(Qcur, "Qcur", il);
+            if (model.layers[il].bq) {
+                Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
                 cb(Qcur, "Qcur", il);
-                if (model.layers[il].bq) {
-                    Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
-                    cb(Qcur, "Qcur", il);
-                }
-                ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
-                if (model.layers[il].bk) {
-                    Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
-                    cb(Kcur, "Kcur", il);
-                }
-                ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
-                if (model.layers[il].bv) {
-                    Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
-                    cb(Vcur, "Vcur", il);
-                }
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
-
-                Qcur = ggml_rope_ext(
-                        ctx0, Qcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-
-                Kcur = ggml_rope_ext(
-                        ctx0, Kcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
-
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
             }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            cb(Kcur, "Kcur", il);
+            if (model.layers[il].bk) {
+                Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
+                cb(Kcur, "Kcur", il);
             }
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            cb(Vcur, "Vcur", il);
+            if (model.layers[il].bv) {
+                Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
+                cb(Vcur, "Vcur", il);
+            }
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
 
-            // feed-forward network
-            cur = build_norm(ffn_inp,
-                    model.layers[il].ffn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "ffn_norm", il);
+            Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-            cur = build_ffn(cur,
-                    model.layers[il].ffn_up,   NULL, NULL,
-                    model.layers[il].ffn_gate, NULL, NULL,
-                    model.layers[il].ffn_down, NULL, NULL,
-                    NULL,
-                    LLM_FFN_SILU, LLM_FFN_PAR, il);
-            cb(cur, "ffn_out", il);
+            Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-            cur = ggml_add(ctx0, cur, ffn_inp);
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
 
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, model.layers[il].bo,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
         }
-        cur = inpL;
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+        }
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+        cb(ffn_inp, "ffn_inp", il);
 
-        cur = build_norm(cur,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, -1);
+        // feed-forward network
+        cur = build_norm(ffn_inp,
+                model.layers[il].ffn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "ffn_norm", il);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+        cur = build_ffn(cur,
+                model.layers[il].ffn_up,   NULL, NULL,
+                model.layers[il].ffn_gate, NULL, NULL,
+                model.layers[il].ffn_down, NULL, NULL,
+                NULL,
+                LLM_FFN_SILU, LLM_FFN_PAR, il);
+        cb(cur, "ffn_out", il);
 
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
+        cur = ggml_add(ctx0, cur, ffn_inp);
 
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
 
-        ggml_build_forward_expand(gf, cur);
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
+
+    cur = build_norm(cur,
+            model.output_norm, NULL,
+            LLM_NORM_RMS, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/jais.cpp

@@ -1,86 +1,86 @@
 #include "models.h"
 
 llm_build_jais::llm_build_jais(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
-        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
+    const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        auto * inp_attn = build_attn_inp_kv();
+    auto * inp_attn = build_attn_inp_kv();
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            cur = build_norm(inpL,
-                    model.layers[il].attn_norm,
-                    model.layers[il].attn_norm_b,
-                    LLM_NORM, il);
-            cb(cur, "attn_norm", il);
-
-            // self-attention
-            {
-                cur = build_lora_mm(model.layers[il].wqkv, cur);
-                cb(cur, "wqkv", il);
-
-                cur = ggml_add(ctx0, cur, model.layers[il].bqkv);
-                cb(cur, "bqkv", il);
-
-                ggml_tensor * Qcur = ggml_view_3d(ctx0, cur, n_embd_head, n_head,    n_tokens, n_embd_head*sizeof(float), cur->nb[1], 0*cur->nb[0]*(n_embd));
-                ggml_tensor * Kcur = ggml_view_3d(ctx0, cur, n_embd_head, n_head_kv, n_tokens, n_embd_head*sizeof(float), cur->nb[1], 1*cur->nb[0]*(n_embd));
-                ggml_tensor * Vcur = ggml_view_3d(ctx0, cur, n_embd_head, n_head_kv, n_tokens, n_embd_head*sizeof(float), cur->nb[1], 1*cur->nb[0]*(n_embd + n_embd_gqa));
-
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
-
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/float(n_embd_head), il);
-            }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
-                inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
-            }
-            // add the input
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpL);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // FF
-            {
-                cur = build_norm(ffn_inp,
-                        model.layers[il].ffn_norm,
-                        model.layers[il].ffn_norm_b,
-                        LLM_NORM, il);
-                cb(cur, "ffn_norm", il);
-
-                cur = build_ffn(cur,
-                        model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
-                        model.layers[il].ffn_gate, model.layers[il].ffn_gate_b, NULL,
-                        model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
-                        NULL,
-                        LLM_FFN_SILU, LLM_FFN_PAR, il);
-                cb(cur, "ffn_out", il);
-            }
-            inpL = ggml_add(ctx0, cur, ffn_inp);
-            cb(inpL, "l_out", il);
-        }
+    for (int il = 0; il < n_layer; ++il) {
         cur = build_norm(inpL,
-                model.output_norm,
-                model.output_norm_b,
-                LLM_NORM, -1);
+                model.layers[il].attn_norm,
+                model.layers[il].attn_norm_b,
+                LLM_NORM, il);
+        cb(cur, "attn_norm", il);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+        // self-attention
+        {
+            cur = build_lora_mm(model.layers[il].wqkv, cur);
+            cb(cur, "wqkv", il);
 
-        cur = build_lora_mm(model.output, cur);
+            cur = ggml_add(ctx0, cur, model.layers[il].bqkv);
+            cb(cur, "bqkv", il);
 
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+            ggml_tensor * Qcur = ggml_view_3d(ctx0, cur, n_embd_head, n_head,    n_tokens, n_embd_head*sizeof(float), cur->nb[1], 0*cur->nb[0]*(n_embd));
+            ggml_tensor * Kcur = ggml_view_3d(ctx0, cur, n_embd_head, n_head_kv, n_tokens, n_embd_head*sizeof(float), cur->nb[1], 1*cur->nb[0]*(n_embd));
+            ggml_tensor * Vcur = ggml_view_3d(ctx0, cur, n_embd_head, n_head_kv, n_tokens, n_embd_head*sizeof(float), cur->nb[1], 1*cur->nb[0]*(n_embd + n_embd_gqa));
 
-        ggml_build_forward_expand(gf, cur);
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
+
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, model.layers[il].bo,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/float(n_embd_head), il);
+        }
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
+            inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
+        }
+        // add the input
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpL);
+        cb(ffn_inp, "ffn_inp", il);
+
+        // FF
+        {
+            cur = build_norm(ffn_inp,
+                    model.layers[il].ffn_norm,
+                    model.layers[il].ffn_norm_b,
+                    LLM_NORM, il);
+            cb(cur, "ffn_norm", il);
+
+            cur = build_ffn(cur,
+                    model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
+                    model.layers[il].ffn_gate, model.layers[il].ffn_gate_b, NULL,
+                    model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
+                    NULL,
+                    LLM_FFN_SILU, LLM_FFN_PAR, il);
+            cb(cur, "ffn_out", il);
+        }
+        inpL = ggml_add(ctx0, cur, ffn_inp);
+        cb(inpL, "l_out", il);
+    }
+    cur = build_norm(inpL,
+            model.output_norm,
+            model.output_norm_b,
+            LLM_NORM, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    cur = build_lora_mm(model.output, cur);
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/jamba.cpp

@@ -1,107 +1,106 @@
 #include "models.h"
 
-
 llm_build_jamba::llm_build_jamba(const llama_model & model, const llm_graph_params & params) : llm_graph_context_mamba(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_head = hparams.n_embd_head_v;
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        // {n_embd, n_tokens}
-        inpL = build_inp_embd(model.tok_embd);
+    // {n_embd, n_tokens}
+    inpL = build_inp_embd(model.tok_embd);
 
-        auto * inp_hybrid = build_inp_mem_hybrid();
+    auto * inp_hybrid = build_inp_mem_hybrid();
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            const int64_t n_head_kv = hparams.n_head_kv(il);
+    for (int il = 0; il < n_layer; ++il) {
+        const int64_t n_head_kv = hparams.n_head_kv(il);
 
-            cur = build_norm(inpL, model.layers[il].attn_norm, NULL, LLM_NORM_RMS, il);
-            cb(cur, "attn_norm", il);
+        cur = build_norm(inpL, model.layers[il].attn_norm, NULL, LLM_NORM_RMS, il);
+        cb(cur, "attn_norm", il);
 
-            if (n_head_kv == 0) {
-                cur = build_mamba_layer(inp_hybrid->get_recr(), cur, model, ubatch, il);
-            } else {
-                // Attention
+        if (n_head_kv == 0) {
+            cur = build_mamba_layer(inp_hybrid->get_recr(), cur, model, ubatch, il);
+        } else {
+            // Attention
 
-                struct ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
-                struct ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
-                struct ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            struct ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            struct ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            struct ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
 
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
 
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
 
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
 
-                // No RoPE :)
-                cur = build_attn(inp_hybrid->get_attn(),
-                        model.layers[il].wo, NULL,
-                        Qcur, Kcur, Vcur, NULL, NULL, NULL, 1.0f/sqrtf(float(n_embd_head)), il);
-            }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
-                inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
-            }
-            // residual
-            struct ggml_tensor * ffn_inp = ggml_add(ctx0, inpL, cur);
-            cb(cur, "ffn_inp", il);
-
-            cur = build_norm(ffn_inp, model.layers[il].ffn_norm, NULL, LLM_NORM_RMS, il);
-            cb(cur, "ffn_norm", il);
-
-            // feed-forward network
-            if (model.layers[il].ffn_gate_inp == nullptr) {
-                // FFN
-                cur = build_ffn(cur,
-                        model.layers[il].ffn_up,   NULL, NULL,
-                        model.layers[il].ffn_gate, NULL, NULL,
-                        model.layers[il].ffn_down, NULL, NULL,
-                        NULL,
-                        LLM_FFN_SILU, LLM_FFN_PAR, il);
-                cb(cur, "ffn_out", il);
-            } else {
-                // MoE branch
-                cur = build_moe_ffn(cur,
-                        model.layers[il].ffn_gate_inp,
-                        model.layers[il].ffn_up_exps,
-                        model.layers[il].ffn_gate_exps,
-                        model.layers[il].ffn_down_exps,
-                        nullptr,
-                        n_expert, n_expert_used,
-                        LLM_FFN_SILU, false,
-                        false, 0.0,
-                        LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
-                        il);
-                cb(cur, "ffn_moe_out", il);
-            }
-            // residual
-            cur = ggml_add(ctx0, ffn_inp, cur);
-
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
+            // No RoPE :)
+            cur = build_attn(inp_hybrid->get_attn(),
+                    model.layers[il].wo, NULL,
+                    Qcur, Kcur, Vcur, NULL, NULL, NULL, 1.0f/sqrtf(float(n_embd_head)), il);
         }
-        // final rmsnorm
-        cur = build_norm(inpL, model.output_norm, NULL, LLM_NORM_RMS, -1);
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
+            inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
+        }
+        // residual
+        struct ggml_tensor * ffn_inp = ggml_add(ctx0, inpL, cur);
+        cb(cur, "ffn_inp", il);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+        cur = build_norm(ffn_inp, model.layers[il].ffn_norm, NULL, LLM_NORM_RMS, il);
+        cb(cur, "ffn_norm", il);
 
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
+        // feed-forward network
+        if (model.layers[il].ffn_gate_inp == nullptr) {
+            // FFN
+            cur = build_ffn(cur,
+                    model.layers[il].ffn_up,   NULL, NULL,
+                    model.layers[il].ffn_gate, NULL, NULL,
+                    model.layers[il].ffn_down, NULL, NULL,
+                    NULL,
+                    LLM_FFN_SILU, LLM_FFN_PAR, il);
+            cb(cur, "ffn_out", il);
+        } else {
+            // MoE branch
+            cur = build_moe_ffn(cur,
+                    model.layers[il].ffn_gate_inp,
+                    model.layers[il].ffn_up_exps,
+                    model.layers[il].ffn_gate_exps,
+                    model.layers[il].ffn_down_exps,
+                    nullptr,
+                    n_expert, n_expert_used,
+                    LLM_FFN_SILU, false,
+                    false, 0.0,
+                    LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
+                    il);
+            cb(cur, "ffn_moe_out", il);
+        }
+        // residual
+        cur = ggml_add(ctx0, ffn_inp, cur);
 
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
 
-        ggml_build_forward_expand(gf, cur);
+        // input for next layer
+        inpL = cur;
+    }
+    // final rmsnorm
+    cur = build_norm(inpL, model.output_norm, NULL, LLM_NORM_RMS, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/llada-moe.cpp

@@ -1,123 +1,122 @@
 #include "models.h"
 
 llm_build_llada_moe::llm_build_llada_moe(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_head = hparams.n_embd_head_v;
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        auto * inp_attn = build_attn_inp_no_cache();
+    auto * inp_attn = build_attn_inp_no_cache();
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            ggml_tensor * inpSA = inpL;
+    for (int il = 0; il < n_layer; ++il) {
+        ggml_tensor * inpSA = inpL;
 
-            // norm
-            cur = build_norm(inpL,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "attn_norm", il);
+        // norm
+        cur = build_norm(inpL,
+                model.layers[il].attn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_norm", il);
 
-            // self_attention
-            {
-                // compute Q and K and RoPE them
-                ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
-                cb(Qcur, "Qcur", il);
+        // self_attention
+        {
+            // compute Q and K and RoPE them
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            cb(Qcur, "Qcur", il);
 
-                ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            cb(Kcur, "Kcur", il);
 
-                ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            cb(Vcur, "Vcur", il);
 
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
 
-                Qcur = build_norm(Qcur, model.layers[il].attn_q_norm, NULL, LLM_NORM_RMS, il);
-                cb(Qcur, "Qcur_normed", il);
+            Qcur = build_norm(Qcur, model.layers[il].attn_q_norm, NULL, LLM_NORM_RMS, il);
+            cb(Qcur, "Qcur_normed", il);
 
-                Kcur = build_norm(Kcur, model.layers[il].attn_k_norm, NULL, LLM_NORM_RMS, il);
-                cb(Kcur, "Kcur_normed", il);
+            Kcur = build_norm(Kcur, model.layers[il].attn_k_norm, NULL, LLM_NORM_RMS, il);
+            cb(Kcur, "Kcur_normed", il);
 
-                Qcur = ggml_rope_ext(
-                        ctx0, Qcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
+            Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-                Kcur = ggml_rope_ext(
-                        ctx0, Kcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
+            Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
 
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, NULL,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
-            }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // MoE branch
-            cur = build_norm(ffn_inp,
-                    model.layers[il].ffn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "ffn_norm", il);
-
-            cur = build_moe_ffn(cur,
-                    model.layers[il].ffn_gate_inp,
-                    model.layers[il].ffn_up_exps,
-                    model.layers[il].ffn_gate_exps,
-                    model.layers[il].ffn_down_exps,
-                    nullptr,
-                    n_expert, n_expert_used,
-                    LLM_FFN_SILU, false,
-                    false, 0.0,
-                    LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
-                    il);
-            cb(cur, "ffn_moe_out", il);
-
-            cur = ggml_add(ctx0, cur, ffn_inp);
-
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, NULL,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
         }
-        cur = inpL;
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+        }
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+        cb(ffn_inp, "ffn_inp", il);
 
-        cur = build_norm(cur,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, -1);
+        // MoE branch
+        cur = build_norm(ffn_inp,
+                model.layers[il].ffn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "ffn_norm", il);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+        cur = build_moe_ffn(cur,
+                model.layers[il].ffn_gate_inp,
+                model.layers[il].ffn_up_exps,
+                model.layers[il].ffn_gate_exps,
+                model.layers[il].ffn_down_exps,
+                nullptr,
+                n_expert, n_expert_used,
+                LLM_FFN_SILU, false,
+                false, 0.0,
+                LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
+                il);
+        cb(cur, "ffn_moe_out", il);
 
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
+        cur = ggml_add(ctx0, cur, ffn_inp);
 
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
 
-        ggml_build_forward_expand(gf, cur);
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
+
+    cur = build_norm(cur,
+            model.output_norm, NULL,
+            LLM_NORM_RMS, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }
-

src/models/llada.cpp

@@ -1,101 +1,99 @@
 #include "models.h"
 
+llm_build_llada::llm_build_llada(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
+    // LLaDA is similar to LLaMA but uses non-causal attention for diffusion
+    const int64_t n_embd_head = hparams.n_embd_head_v;
 
-llm_build_llada::llm_build_llada(const llama_model & model, const llm_graph_params & params) :
-        llm_graph_context(params) {
-        // LLaDA is similar to LLaMA but uses non-causal attention for diffusion
-        const int64_t n_embd_head = hparams.n_embd_head_v;
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    inpL = build_inp_embd(model.tok_embd);
 
-        inpL = build_inp_embd(model.tok_embd);
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // Non-causal attention for diffusion
+    auto * inp_attn = build_attn_inp_no_cache();
 
-        // Non-causal attention for diffusion
-        auto * inp_attn = build_attn_inp_no_cache();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    for (int il = 0; il < n_layer; ++il) {
+        ggml_tensor * inpSA = inpL;
 
-        for (int il = 0; il < n_layer; ++il) {
-            ggml_tensor * inpSA = inpL;
+        // norm
+        cur = build_norm(inpL, model.layers[il].attn_norm, NULL, LLM_NORM_RMS, il);
+        cb(cur, "attn_norm", il);
 
-            // norm
-            cur = build_norm(inpL, model.layers[il].attn_norm, NULL, LLM_NORM_RMS, il);
-            cb(cur, "attn_norm", il);
+        // self-attention
+        {
+            // compute separate Q, K, V projections without bias, matching LLaDALlamaBlock
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
 
-            // self-attention
-            {
-                // compute separate Q, K, V projections without bias, matching LLaDALlamaBlock
-                ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
-                ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
-                ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
 
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
 
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
+            Qcur = ggml_rope_ext(ctx0, Qcur, inp_pos, nullptr, n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                                    ext_factor, attn_factor, beta_fast, beta_slow);
 
-                Qcur = ggml_rope_ext(ctx0, Qcur, inp_pos, nullptr, n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                                     ext_factor, attn_factor, beta_fast, beta_slow);
+            Kcur = ggml_rope_ext(ctx0, Kcur, inp_pos, nullptr, n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                                    ext_factor, attn_factor, beta_fast, beta_slow);
 
-                Kcur = ggml_rope_ext(ctx0, Kcur, inp_pos, nullptr, n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                                     ext_factor, attn_factor, beta_fast, beta_slow);
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
 
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
-
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, NULL,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f / sqrtf(float(n_embd_head)), il);
-            }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur   = ggml_get_rows(ctx0, cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // feed-forward network
-            cur = build_norm(ffn_inp, model.layers[il].ffn_norm, NULL, LLM_NORM_RMS, il);
-            cb(cur, "ffn_norm", il);
-
-            cur = build_ffn(cur,
-                    model.layers[il].ffn_up, NULL, NULL,
-                    model.layers[il].ffn_gate, NULL, NULL,
-                    model.layers[il].ffn_down, NULL, NULL,
-                    NULL, LLM_FFN_SILU, LLM_FFN_PAR, il);
-            cb(cur, "ffn_out", il);
-
-            cur = ggml_add(ctx0, cur, ffn_inp);
-
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, NULL,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f / sqrtf(float(n_embd_head)), il);
         }
-        cur = inpL;
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur   = ggml_get_rows(ctx0, cur, inp_out_ids);
+            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+        }
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+        cb(ffn_inp, "ffn_inp", il);
 
-        cur = build_norm(cur, model.output_norm, NULL, LLM_NORM_RMS, -1);
+        // feed-forward network
+        cur = build_norm(ffn_inp, model.layers[il].ffn_norm, NULL, LLM_NORM_RMS, il);
+        cb(cur, "ffn_norm", il);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+        cur = build_ffn(cur,
+                model.layers[il].ffn_up, NULL, NULL,
+                model.layers[il].ffn_gate, NULL, NULL,
+                model.layers[il].ffn_down, NULL, NULL,
+                NULL, LLM_FFN_SILU, LLM_FFN_PAR, il);
+        cb(cur, "ffn_out", il);
 
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
+        cur = ggml_add(ctx0, cur, ffn_inp);
 
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
 
-        ggml_build_forward_expand(gf, cur);
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
+
+    cur = build_norm(cur, model.output_norm, NULL, LLM_NORM_RMS, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/llama.cpp

@@ -1,156 +1,155 @@
 #include "models.h"
 
-
 llm_build_llama::llm_build_llama(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_head = hparams.n_embd_head_v;
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        auto * inp_attn = build_attn_inp_kv();
+    auto * inp_attn = build_attn_inp_kv();
 
-        const float kq_scale = hparams.f_attention_scale == 0.0f ? 1.0f/sqrtf(float(n_embd_head)) : hparams.f_attention_scale;
+    const float kq_scale = hparams.f_attention_scale == 0.0f ? 1.0f/sqrtf(float(n_embd_head)) : hparams.f_attention_scale;
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            ggml_tensor * inpSA = inpL;
+    for (int il = 0; il < n_layer; ++il) {
+        ggml_tensor * inpSA = inpL;
 
-            // norm
-            cur = build_norm(inpL,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "attn_norm", il);
+        // norm
+        cur = build_norm(inpL,
+                model.layers[il].attn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_norm", il);
 
-            // self-attention
-            {
-                // rope freq factors for llama3; may return nullptr for llama2 and other models
-                ggml_tensor * rope_factors = model.get_rope_factors(cparams, il);
+        // self-attention
+        {
+            // rope freq factors for llama3; may return nullptr for llama2 and other models
+            ggml_tensor * rope_factors = model.get_rope_factors(cparams, il);
 
-                // compute Q and K and RoPE them
-                ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            // compute Q and K and RoPE them
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            cb(Qcur, "Qcur", il);
+            if (model.layers[il].bq) {
+                Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
                 cb(Qcur, "Qcur", il);
-                if (model.layers[il].bq) {
-                    Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
-                    cb(Qcur, "Qcur", il);
-                }
-                ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            }
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            cb(Kcur, "Kcur", il);
+            if (model.layers[il].bk) {
+                Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
                 cb(Kcur, "Kcur", il);
-                if (model.layers[il].bk) {
-                    Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
-                    cb(Kcur, "Kcur", il);
-                }
-                ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            }
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            cb(Vcur, "Vcur", il);
+            if (model.layers[il].bv) {
+                Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
                 cb(Vcur, "Vcur", il);
-                if (model.layers[il].bv) {
-                    Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
-                    cb(Vcur, "Vcur", il);
-                }
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
-
-                Qcur = ggml_rope_ext(
-                        ctx0, Qcur, inp_pos, rope_factors,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-
-                Kcur = ggml_rope_ext(
-                        ctx0, Kcur, inp_pos, rope_factors,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
-
-                if (hparams.use_kq_norm) {
-                    // Llama4TextL2Norm
-                    Qcur = ggml_rms_norm(ctx0, Qcur, hparams.f_norm_rms_eps);
-                    Kcur = ggml_rms_norm(ctx0, Kcur, hparams.f_norm_rms_eps);
-                    cb(Qcur, "Qcur_normed", il);
-                    cb(Kcur, "Kcur_normed", il);
-                }
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, kq_scale, il);
-                cb(cur, "attn_out", il);
             }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
+
+            Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, rope_factors,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
+
+            Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, rope_factors,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
+
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
+
+            if (hparams.use_kq_norm) {
+                // Llama4TextL2Norm
+                Qcur = ggml_rms_norm(ctx0, Qcur, hparams.f_norm_rms_eps);
+                Kcur = ggml_rms_norm(ctx0, Kcur, hparams.f_norm_rms_eps);
+                cb(Qcur, "Qcur_normed", il);
+                cb(Kcur, "Kcur_normed", il);
             }
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // feed-forward network (non-MoE)
-            if (model.layers[il].ffn_gate_inp == nullptr) {
-
-                cur = build_norm(ffn_inp,
-                        model.layers[il].ffn_norm, NULL,
-                        LLM_NORM_RMS, il);
-                cb(cur, "ffn_norm", il);
-
-                cur = build_ffn(cur,
-                        model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
-                        model.layers[il].ffn_gate, model.layers[il].ffn_gate_b, NULL,
-                        model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
-                        NULL,
-                        LLM_FFN_SILU, LLM_FFN_PAR, il);
-                cb(cur, "ffn_out", il);
-            } else {
-                // MoE branch
-                cur = build_norm(ffn_inp,
-                        model.layers[il].ffn_norm, NULL,
-                        LLM_NORM_RMS, il);
-                cb(cur, "ffn_norm", il);
-
-                cur = build_moe_ffn(cur,
-                        model.layers[il].ffn_gate_inp,
-                        model.layers[il].ffn_up_exps,
-                        model.layers[il].ffn_gate_exps,
-                        model.layers[il].ffn_down_exps,
-                        nullptr,
-                        n_expert, n_expert_used,
-                        LLM_FFN_SILU, true,
-                        false, 0.0,
-                        LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
-                        il);
-                cb(cur, "ffn_moe_out", il);
-            }
-            cur = ggml_add(ctx0, cur, ffn_inp);
-            cb(cur, "ffn_out", il);
-
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, model.layers[il].bo,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, kq_scale, il);
+            cb(cur, "attn_out", il);
         }
-        cur = inpL;
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+        }
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+        cb(ffn_inp, "ffn_inp", il);
 
-        cur = build_norm(cur,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, -1);
+        // feed-forward network (non-MoE)
+        if (model.layers[il].ffn_gate_inp == nullptr) {
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+            cur = build_norm(ffn_inp,
+                    model.layers[il].ffn_norm, NULL,
+                    LLM_NORM_RMS, il);
+            cb(cur, "ffn_norm", il);
 
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
+            cur = build_ffn(cur,
+                    model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
+                    model.layers[il].ffn_gate, model.layers[il].ffn_gate_b, NULL,
+                    model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
+                    NULL,
+                    LLM_FFN_SILU, LLM_FFN_PAR, il);
+            cb(cur, "ffn_out", il);
+        } else {
+            // MoE branch
+            cur = build_norm(ffn_inp,
+                    model.layers[il].ffn_norm, NULL,
+                    LLM_NORM_RMS, il);
+            cb(cur, "ffn_norm", il);
 
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+            cur = build_moe_ffn(cur,
+                    model.layers[il].ffn_gate_inp,
+                    model.layers[il].ffn_up_exps,
+                    model.layers[il].ffn_gate_exps,
+                    model.layers[il].ffn_down_exps,
+                    nullptr,
+                    n_expert, n_expert_used,
+                    LLM_FFN_SILU, true,
+                    false, 0.0,
+                    LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
+                    il);
+            cb(cur, "ffn_moe_out", il);
+        }
+        cur = ggml_add(ctx0, cur, ffn_inp);
+        cb(cur, "ffn_out", il);
 
-        ggml_build_forward_expand(gf, cur);
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
+
+        // input for next layer
+        inpL = cur;
     }
+    cur = inpL;
+
+    cur = build_norm(cur,
+            model.output_norm, NULL,
+            LLM_NORM_RMS, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
+}

src/models/minicpm3.cpp

@@ -1,200 +1,199 @@
 #include "models.h"
 
-
 llm_build_minicpm3::llm_build_minicpm3(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        //TODO: if the model varies, these parameters need to be read from the model
-        const int64_t n_embd_base = 256;
-        const float scale_embd  = 12.0f;
-        const float scale_depth = 1.4f;
-        const float kq_scale = 1.0f / sqrtf(float(hparams.n_embd_head_k));
+    //TODO: if the model varies, these parameters need to be read from the model
+    const int64_t n_embd_base = 256;
+    const float scale_embd  = 12.0f;
+    const float scale_depth = 1.4f;
+    const float kq_scale = 1.0f / sqrtf(float(hparams.n_embd_head_k));
 
-        const uint32_t n_embd_head_qk_rope = hparams.n_rot;
-        const uint32_t n_embd_head_qk_nope = hparams.n_embd_head_k - hparams.n_rot;
-        const uint32_t kv_lora_rank = hparams.n_lora_kv;
+    const uint32_t n_embd_head_qk_rope = hparams.n_rot;
+    const uint32_t n_embd_head_qk_nope = hparams.n_embd_head_k - hparams.n_rot;
+    const uint32_t kv_lora_rank = hparams.n_lora_kv;
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        // scale the input embeddings
-        inpL = ggml_scale(ctx0, inpL, scale_embd);
-        cb(inpL, "inp_scaled", -1);
+    // scale the input embeddings
+    inpL = ggml_scale(ctx0, inpL, scale_embd);
+    cb(inpL, "inp_scaled", -1);
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        auto * inp_attn = build_attn_inp_kv();
+    auto * inp_attn = build_attn_inp_kv();
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            ggml_tensor * inpSA = inpL;
+    for (int il = 0; il < n_layer; ++il) {
+        ggml_tensor * inpSA = inpL;
 
-            ggml_tensor * rope_factors = model.get_rope_factors(cparams, il);
+        ggml_tensor * rope_factors = model.get_rope_factors(cparams, il);
 
-            // norm
-            cur = build_norm(inpL,
-                    model.layers[il].attn_norm, NULL,
+        // norm
+        cur = build_norm(inpL,
+                model.layers[il].attn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_norm", il);
+
+        // self_attention
+        {
+            ggml_tensor * q = NULL;
+            // {n_embd, q_lora_rank} * {n_embd, n_tokens} -> {q_lora_rank, n_tokens}
+            q = ggml_mul_mat(ctx0, model.layers[il].wq_a, cur);
+            cb(q, "q", il);
+
+            q = build_norm(q,
+                    model.layers[il].attn_q_a_norm, NULL,
                     LLM_NORM_RMS, il);
-            cb(cur, "attn_norm", il);
+            cb(q, "q", il);
 
-            // self_attention
-            {
-                ggml_tensor * q = NULL;
-                // {n_embd, q_lora_rank} * {n_embd, n_tokens} -> {q_lora_rank, n_tokens}
-                q = ggml_mul_mat(ctx0, model.layers[il].wq_a, cur);
-                cb(q, "q", il);
+            // {q_lora_rank, n_head * hparams.n_embd_head_k} * {q_lora_rank, n_tokens} -> {n_head * hparams.n_embd_head_k, n_tokens}
+            q = ggml_mul_mat(ctx0, model.layers[il].wq_b, q);
+            cb(q, "q", il);
 
-                q = build_norm(q,
-                        model.layers[il].attn_q_a_norm, NULL,
-                        LLM_NORM_RMS, il);
-                cb(q, "q", il);
+            // split into {n_head * n_embd_head_qk_nope, n_tokens}
+            ggml_tensor * q_nope = ggml_view_3d(ctx0, q, n_embd_head_qk_nope, n_head, n_tokens,
+                    ggml_row_size(q->type, hparams.n_embd_head_k),
+                    ggml_row_size(q->type, hparams.n_embd_head_k * n_head),
+                    0);
+            cb(q_nope, "q_nope", il);
 
-                // {q_lora_rank, n_head * hparams.n_embd_head_k} * {q_lora_rank, n_tokens} -> {n_head * hparams.n_embd_head_k, n_tokens}
-                q = ggml_mul_mat(ctx0, model.layers[il].wq_b, q);
-                cb(q, "q", il);
+            // and {n_head * n_embd_head_qk_rope, n_tokens}
+            ggml_tensor * q_pe = ggml_view_3d(ctx0, q, n_embd_head_qk_rope, n_head, n_tokens,
+                    ggml_row_size(q->type, hparams.n_embd_head_k),
+                    ggml_row_size(q->type, hparams.n_embd_head_k * n_head),
+                    ggml_row_size(q->type, n_embd_head_qk_nope));
+            cb(q_pe, "q_pe", il);
 
-                // split into {n_head * n_embd_head_qk_nope, n_tokens}
-                ggml_tensor * q_nope = ggml_view_3d(ctx0, q, n_embd_head_qk_nope, n_head, n_tokens,
-                        ggml_row_size(q->type, hparams.n_embd_head_k),
-                        ggml_row_size(q->type, hparams.n_embd_head_k * n_head),
-                        0);
-                cb(q_nope, "q_nope", il);
+            // {n_embd, kv_lora_rank + n_embd_head_qk_rope} * {n_embd, n_tokens} -> {kv_lora_rank + n_embd_head_qk_rope, n_tokens}
+            ggml_tensor * kv_pe_compresseed = ggml_mul_mat(ctx0, model.layers[il].wkv_a_mqa, cur);
+            cb(kv_pe_compresseed, "kv_pe_compresseed", il);
 
-                // and {n_head * n_embd_head_qk_rope, n_tokens}
-                ggml_tensor * q_pe = ggml_view_3d(ctx0, q, n_embd_head_qk_rope, n_head, n_tokens,
-                        ggml_row_size(q->type, hparams.n_embd_head_k),
-                        ggml_row_size(q->type, hparams.n_embd_head_k * n_head),
-                        ggml_row_size(q->type, n_embd_head_qk_nope));
-                cb(q_pe, "q_pe", il);
+            // split into {kv_lora_rank, n_tokens}
+            ggml_tensor * kv_compressed = ggml_view_2d(ctx0, kv_pe_compresseed, kv_lora_rank, n_tokens,
+                    kv_pe_compresseed->nb[1],
+                    0);
+            cb(kv_compressed, "kv_compressed", il);
 
-                // {n_embd, kv_lora_rank + n_embd_head_qk_rope} * {n_embd, n_tokens} -> {kv_lora_rank + n_embd_head_qk_rope, n_tokens}
-                ggml_tensor * kv_pe_compresseed = ggml_mul_mat(ctx0, model.layers[il].wkv_a_mqa, cur);
-                cb(kv_pe_compresseed, "kv_pe_compresseed", il);
+            // and {n_embd_head_qk_rope, n_tokens}
+            ggml_tensor * k_pe = ggml_view_3d(ctx0, kv_pe_compresseed, n_embd_head_qk_rope, 1, n_tokens,
+                    kv_pe_compresseed->nb[1],
+                    kv_pe_compresseed->nb[1],
+                    ggml_row_size(kv_pe_compresseed->type, kv_lora_rank));
+            cb(k_pe, "k_pe", il);
 
-                // split into {kv_lora_rank, n_tokens}
-                ggml_tensor * kv_compressed = ggml_view_2d(ctx0, kv_pe_compresseed, kv_lora_rank, n_tokens,
-                        kv_pe_compresseed->nb[1],
-                        0);
-                cb(kv_compressed, "kv_compressed", il);
+            kv_compressed = build_norm(kv_compressed,
+                    model.layers[il].attn_kv_a_norm, NULL,
+                    LLM_NORM_RMS, il);
+            cb(kv_compressed, "kv_compressed", il);
 
-                // and {n_embd_head_qk_rope, n_tokens}
-                ggml_tensor * k_pe = ggml_view_3d(ctx0, kv_pe_compresseed, n_embd_head_qk_rope, 1, n_tokens,
-                        kv_pe_compresseed->nb[1],
-                        kv_pe_compresseed->nb[1],
-                        ggml_row_size(kv_pe_compresseed->type, kv_lora_rank));
-                cb(k_pe, "k_pe", il);
+            // {kv_lora_rank, n_head * (n_embd_head_qk_nope + n_embd_head_v)} * {kv_lora_rank, n_tokens} -> {n_head * (n_embd_head_qk_nope + n_embd_head_v), n_tokens}
+            ggml_tensor * kv = ggml_mul_mat(ctx0, model.layers[il].wkv_b, kv_compressed);
+            cb(kv, "kv", il);
 
-                kv_compressed = build_norm(kv_compressed,
-                        model.layers[il].attn_kv_a_norm, NULL,
-                        LLM_NORM_RMS, il);
-                cb(kv_compressed, "kv_compressed", il);
+            // split into {n_head * n_embd_head_qk_nope, n_tokens}
+            ggml_tensor * k_nope = ggml_view_3d(ctx0, kv, n_embd_head_qk_nope, n_head, n_tokens,
+                    ggml_row_size(kv->type, n_embd_head_qk_nope + hparams.n_embd_head_v),
+                    ggml_row_size(kv->type, n_head * (n_embd_head_qk_nope + hparams.n_embd_head_v)),
+                    0);
+            cb(k_nope, "k_nope", il);
 
-                // {kv_lora_rank, n_head * (n_embd_head_qk_nope + n_embd_head_v)} * {kv_lora_rank, n_tokens} -> {n_head * (n_embd_head_qk_nope + n_embd_head_v), n_tokens}
-                ggml_tensor * kv = ggml_mul_mat(ctx0, model.layers[il].wkv_b, kv_compressed);
-                cb(kv, "kv", il);
+            // and {n_head * n_embd_head_v, n_tokens}
+            ggml_tensor * v_states = ggml_view_3d(ctx0, kv, hparams.n_embd_head_v, n_head, n_tokens,
+                    ggml_row_size(kv->type, (n_embd_head_qk_nope + hparams.n_embd_head_v)),
+                    ggml_row_size(kv->type, (n_embd_head_qk_nope + hparams.n_embd_head_v)*n_head),
+                    ggml_row_size(kv->type, (n_embd_head_qk_nope)));
+            cb(v_states, "v_states", il);
 
-                // split into {n_head * n_embd_head_qk_nope, n_tokens}
-                ggml_tensor * k_nope = ggml_view_3d(ctx0, kv, n_embd_head_qk_nope, n_head, n_tokens,
-                        ggml_row_size(kv->type, n_embd_head_qk_nope + hparams.n_embd_head_v),
-                        ggml_row_size(kv->type, n_head * (n_embd_head_qk_nope + hparams.n_embd_head_v)),
-                        0);
-                cb(k_nope, "k_nope", il);
+            v_states = ggml_cont(ctx0, v_states);
+            cb(v_states, "v_states", il);
 
-                // and {n_head * n_embd_head_v, n_tokens}
-                ggml_tensor * v_states = ggml_view_3d(ctx0, kv, hparams.n_embd_head_v, n_head, n_tokens,
-                        ggml_row_size(kv->type, (n_embd_head_qk_nope + hparams.n_embd_head_v)),
-                        ggml_row_size(kv->type, (n_embd_head_qk_nope + hparams.n_embd_head_v)*n_head),
-                        ggml_row_size(kv->type, (n_embd_head_qk_nope)));
-                cb(v_states, "v_states", il);
+            q_pe = ggml_rope_ext(
+                    ctx0, q_pe, inp_pos, rope_factors,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
+            cb(q_pe, "q_pe", il);
 
-                v_states = ggml_cont(ctx0, v_states);
-                cb(v_states, "v_states", il);
+            // shared RoPE key
+            k_pe = ggml_rope_ext(
+                    ctx0, k_pe, inp_pos, rope_factors,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
+            cb(k_pe, "k_pe", il);
 
-                q_pe = ggml_rope_ext(
-                        ctx0, q_pe, inp_pos, rope_factors,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-                cb(q_pe, "q_pe", il);
+            ggml_tensor * q_states = ggml_concat(ctx0, q_nope, q_pe, 0);
+            cb(q_states, "q_states", il);
 
-                // shared RoPE key
-                k_pe = ggml_rope_ext(
-                        ctx0, k_pe, inp_pos, rope_factors,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-                cb(k_pe, "k_pe", il);
+            ggml_tensor * k_states = ggml_concat(ctx0, k_nope, ggml_repeat(ctx0, k_pe, q_pe), 0);
+            cb(k_states, "k_states", il);
 
-                ggml_tensor * q_states = ggml_concat(ctx0, q_nope, q_pe, 0);
-                cb(q_states, "q_states", il);
-
-                ggml_tensor * k_states = ggml_concat(ctx0, k_nope, ggml_repeat(ctx0, k_pe, q_pe), 0);
-                cb(k_states, "k_states", il);
-
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, NULL,
-                        q_states, k_states, v_states, nullptr, nullptr, nullptr, kq_scale, il);
-            }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
-            // scale_res - scale the hidden states for residual connection
-            const float scale_res = scale_depth/sqrtf(float(n_layer)); // TODO: is this correct?
-            cur = ggml_scale(ctx0, cur, scale_res);
-            cb(cur, "hidden_scaled", il);
-
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // feed-forward network
-            {
-                cur = build_norm(ffn_inp,
-                        model.layers[il].ffn_norm, NULL,
-                        LLM_NORM_RMS, il);
-                cb(cur, "ffn_norm", il);
-
-                cur = build_ffn(cur,
-                        model.layers[il].ffn_up,   NULL, NULL,
-                        model.layers[il].ffn_gate, NULL, NULL,
-                        model.layers[il].ffn_down, NULL, NULL,
-                        NULL,
-                        LLM_FFN_SILU, LLM_FFN_PAR, il);
-                cb(cur, "ffn_out", il);
-            }
-            // scale the hidden states for residual connection
-            cur = ggml_scale(ctx0, cur, scale_res);
-            cb(cur, "hidden_scaled_ffn", il);
-
-            cur = ggml_add(ctx0, cur, ffn_inp);
-
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, NULL,
+                    q_states, k_states, v_states, nullptr, nullptr, nullptr, kq_scale, il);
         }
-        cur = inpL;
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+        }
+        // scale_res - scale the hidden states for residual connection
+        const float scale_res = scale_depth/sqrtf(float(n_layer)); // TODO: is this correct?
+        cur = ggml_scale(ctx0, cur, scale_res);
+        cb(cur, "hidden_scaled", il);
 
-        cur = build_norm(cur,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, -1);
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+        cb(ffn_inp, "ffn_inp", il);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+        // feed-forward network
+        {
+            cur = build_norm(ffn_inp,
+                    model.layers[il].ffn_norm, NULL,
+                    LLM_NORM_RMS, il);
+            cb(cur, "ffn_norm", il);
 
-        // lm_head scaling
-        const float scale_lmhead = float(n_embd_base)/float(n_embd);
-        cur = ggml_scale(ctx0, cur, scale_lmhead);
-        cb(cur, "lmhead_scaling", -1);
+            cur = build_ffn(cur,
+                    model.layers[il].ffn_up,   NULL, NULL,
+                    model.layers[il].ffn_gate, NULL, NULL,
+                    model.layers[il].ffn_down, NULL, NULL,
+                    NULL,
+                    LLM_FFN_SILU, LLM_FFN_PAR, il);
+            cb(cur, "ffn_out", il);
+        }
+        // scale the hidden states for residual connection
+        cur = ggml_scale(ctx0, cur, scale_res);
+        cb(cur, "hidden_scaled_ffn", il);
 
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
+        cur = ggml_add(ctx0, cur, ffn_inp);
 
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
 
-        ggml_build_forward_expand(gf, cur);
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
+
+    cur = build_norm(cur,
+            model.output_norm, NULL,
+            LLM_NORM_RMS, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head scaling
+    const float scale_lmhead = float(n_embd_base)/float(n_embd);
+    cur = ggml_scale(ctx0, cur, scale_lmhead);
+    cb(cur, "lmhead_scaling", -1);
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/models.h

@@ -361,6 +361,10 @@ struct llm_build_orion : public llm_graph_context {
     llm_build_orion(const llama_model & model, const llm_graph_params & params);
 };
 
+struct llm_build_pangu_embedded : public llm_graph_context {
+    llm_build_pangu_embedded(const llama_model & model, const llm_graph_params & params);
+};
+
 struct llm_build_phi2 : public llm_graph_context {
     llm_build_phi2(const llama_model & model, const llm_graph_params & params);
 };

src/models/nemotron.cpp

@@ -1,122 +1,122 @@
 #include "models.h"
 
 llm_build_nemotron::llm_build_nemotron(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_head = hparams.n_embd_head_v;
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        //GGML_ASSERT(n_embd_head == hparams.n_rot);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    //GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        auto * inp_attn = build_attn_inp_kv();
+    auto * inp_attn = build_attn_inp_kv();
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            ggml_tensor * inpSA = inpL;
+    for (int il = 0; il < n_layer; ++il) {
+        ggml_tensor * inpSA = inpL;
 
-            // norm
-            cur = build_norm(inpL,
-                    model.layers[il].attn_norm,
-                    model.layers[il].attn_norm_b,
-                    LLM_NORM, il);
-            cb(cur, "attn_norm", il);
+        // norm
+        cur = build_norm(inpL,
+                model.layers[il].attn_norm,
+                model.layers[il].attn_norm_b,
+                LLM_NORM, il);
+        cb(cur, "attn_norm", il);
 
-            // self-attention
-            {
-                // compute Q and K and RoPE them
-                ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+        // self-attention
+        {
+            // compute Q and K and RoPE them
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            cb(Qcur, "Qcur", il);
+            if (model.layers[il].bq) {
+                Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
                 cb(Qcur, "Qcur", il);
-                if (model.layers[il].bq) {
-                    Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
-                    cb(Qcur, "Qcur", il);
-                }
-                ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
-                if (model.layers[il].bk) {
-                    Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
-                    cb(Kcur, "Kcur", il);
-                }
-                ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
-                if (model.layers[il].bv) {
-                    Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
-                    cb(Vcur, "Vcur", il);
-                }
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
-
-                Qcur = ggml_rope_ext(
-                        ctx0, Qcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-
-                Kcur = ggml_rope_ext(
-                        ctx0, Kcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
-
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
             }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            cb(Kcur, "Kcur", il);
+            if (model.layers[il].bk) {
+                Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
+                cb(Kcur, "Kcur", il);
             }
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            cb(Vcur, "Vcur", il);
+            if (model.layers[il].bv) {
+                Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
+                cb(Vcur, "Vcur", il);
+            }
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
 
-            // feed-forward network
-            cur = build_norm(ffn_inp,
-                    model.layers[il].ffn_norm,
-                    model.layers[il].ffn_norm_b,
-                    LLM_NORM, il);
-            cb(cur, "ffn_norm", il);
+            Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-            cur = build_ffn(cur,
-                    model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
-                    NULL,                      NULL,                        NULL,
-                    model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
-                    NULL,
-                    LLM_FFN_RELU_SQR, LLM_FFN_SEQ, il);
+            Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-            cur = ggml_add(ctx0, cur, ffn_inp);
-            cb(cur, "ffn_out", il);
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
 
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, model.layers[il].bo,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
         }
-        cur = inpL;
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+        }
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+        cb(ffn_inp, "ffn_inp", il);
 
-        cur = build_norm(cur,
-                model.output_norm, model.output_norm_b,
-                LLM_NORM, -1);
+        // feed-forward network
+        cur = build_norm(ffn_inp,
+                model.layers[il].ffn_norm,
+                model.layers[il].ffn_norm_b,
+                LLM_NORM, il);
+        cb(cur, "ffn_norm", il);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+        cur = build_ffn(cur,
+                model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
+                NULL,                      NULL,                        NULL,
+                model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
+                NULL,
+                LLM_FFN_RELU_SQR, LLM_FFN_SEQ, il);
 
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
+        cur = ggml_add(ctx0, cur, ffn_inp);
+        cb(cur, "ffn_out", il);
 
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
 
-        ggml_build_forward_expand(gf, cur);
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
+
+    cur = build_norm(cur,
+            model.output_norm, model.output_norm_b,
+            LLM_NORM, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/neo-bert.cpp

@@ -1,104 +1,104 @@
 #include "models.h"
 
 llm_build_neo_bert::llm_build_neo_bert(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
-        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
+    const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
-        ggml_tensor * inp_pos = build_inp_pos();
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        // construct input embeddings (token, type, position)
-        inpL = build_inp_embd(model.tok_embd);
-        cb(inpL, "inp_embd", -1);
+    // construct input embeddings (token, type, position)
+    inpL = build_inp_embd(model.tok_embd);
+    cb(inpL, "inp_embd", -1);
 
-        auto * inp_attn = build_attn_inp_no_cache();
+    auto * inp_attn = build_attn_inp_no_cache();
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            ggml_tensor * cur = inpL;
+    for (int il = 0; il < n_layer; ++il) {
+        ggml_tensor * cur = inpL;
 
-            // pre-norm
-            cur = build_norm(inpL,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, il);
+        // pre-norm
+        cur = build_norm(inpL,
+                model.layers[il].attn_norm, NULL,
+                LLM_NORM_RMS, il);
 
-            {
-                ggml_tensor * Qcur;
-                ggml_tensor * Kcur;
-                ggml_tensor * Vcur;
+        {
+            ggml_tensor * Qcur;
+            ggml_tensor * Kcur;
+            ggml_tensor * Vcur;
 
-                // self-attention
-                cur = build_lora_mm(model.layers[il].wqkv, cur);
-                cb(cur, "wqkv", il);
+            // self-attention
+            cur = build_lora_mm(model.layers[il].wqkv, cur);
+            cb(cur, "wqkv", il);
 
-                Qcur = ggml_view_3d(ctx0, cur, n_embd_head, n_head,    n_tokens, n_embd_head*sizeof(float), cur->nb[1], 0*sizeof(float)*(n_embd));
-                Kcur = ggml_view_3d(ctx0, cur, n_embd_head, n_head_kv, n_tokens, n_embd_head*sizeof(float), cur->nb[1], 1*sizeof(float)*(n_embd));
-                Vcur = ggml_view_3d(ctx0, cur, n_embd_head, n_head_kv, n_tokens, n_embd_head*sizeof(float), cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa));
+            Qcur = ggml_view_3d(ctx0, cur, n_embd_head, n_head,    n_tokens, n_embd_head*sizeof(float), cur->nb[1], 0*sizeof(float)*(n_embd));
+            Kcur = ggml_view_3d(ctx0, cur, n_embd_head, n_head_kv, n_tokens, n_embd_head*sizeof(float), cur->nb[1], 1*sizeof(float)*(n_embd));
+            Vcur = ggml_view_3d(ctx0, cur, n_embd_head, n_head_kv, n_tokens, n_embd_head*sizeof(float), cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa));
 
-                // RoPE
-                Qcur = ggml_rope_ext(
-                        ctx0, Qcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
+            // RoPE
+            Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-                Kcur = ggml_rope_ext(
-                        ctx0, Kcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
+            Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
 
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, nullptr,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
-                cb(cur, "kqv_out", il);
-            }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
-                inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
-            }
-            // re-add the layer input
-            cur = ggml_add(ctx0, cur, inpL);
-
-            ggml_tensor * ffn_inp = cur;
-            cb(ffn_inp, "ffn_inp", il);
-
-            // pre-norm
-            cur = build_norm(ffn_inp,
-                    model.layers[il].ffn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "ffn_norm", il);
-
-            // feed-forward network
-            cur = build_ffn(cur,
-                    model.layers[il].ffn_up,
-                    NULL, NULL, NULL, NULL, NULL,
-                    model.layers[il].ffn_down,
-                    NULL, NULL, NULL,
-                    LLM_FFN_SWIGLU, LLM_FFN_SEQ, il);
-
-            // attentions bypass the intermediate layer
-            cur = ggml_add(ctx0, cur, ffn_inp);
-
-            // input for next layer
-            inpL = cur;
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, nullptr,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
+            cb(cur, "kqv_out", il);
         }
-        cur = inpL;
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
+            inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
+        }
+        // re-add the layer input
+        cur = ggml_add(ctx0, cur, inpL);
 
-        cur = build_norm(cur,
-                model.output_norm_enc, NULL,
-                LLM_NORM_RMS, -1);
+        ggml_tensor * ffn_inp = cur;
+        cb(ffn_inp, "ffn_inp", il);
 
-        cb(cur, "result_embd", -1);
-        res->t_embd = cur;
+        // pre-norm
+        cur = build_norm(ffn_inp,
+                model.layers[il].ffn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "ffn_norm", il);
 
-        ggml_build_forward_expand(gf, cur);
+        // feed-forward network
+        cur = build_ffn(cur,
+                model.layers[il].ffn_up,
+                NULL, NULL, NULL, NULL, NULL,
+                model.layers[il].ffn_down,
+                NULL, NULL, NULL,
+                LLM_FFN_SWIGLU, LLM_FFN_SEQ, il);
+
+        // attentions bypass the intermediate layer
+        cur = ggml_add(ctx0, cur, ffn_inp);
+
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
+
+    cur = build_norm(cur,
+            model.output_norm_enc, NULL,
+            LLM_NORM_RMS, -1);
+
+    cb(cur, "result_embd", -1);
+    res->t_embd = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/olmo.cpp

@@ -1,121 +1,121 @@
 #include "models.h"
 
 llm_build_olmo::llm_build_olmo(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_head = hparams.n_embd_head_v;
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        auto * inp_attn = build_attn_inp_kv();
+    auto * inp_attn = build_attn_inp_kv();
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            ggml_tensor * inpSA = inpL;
+    for (int il = 0; il < n_layer; ++il) {
+        ggml_tensor * inpSA = inpL;
 
-            // norm
-            cur = build_norm(inpL,
-                    NULL, NULL,
-                    LLM_NORM, il);
-            cb(cur, "attn_norm", il);
-
-            // self-attention
-            {
-                // compute Q and K and RoPE them
-                ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
-                cb(Qcur, "Qcur", il);
-                if (hparams.f_clamp_kqv > 0.0f) {
-                    Qcur = ggml_clamp(ctx0, Qcur, -hparams.f_clamp_kqv, hparams.f_clamp_kqv);
-                    cb(Qcur, "Qcur", il);
-                }
-                ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
-                if (hparams.f_clamp_kqv > 0.0f) {
-                    Kcur = ggml_clamp(ctx0, Kcur, -hparams.f_clamp_kqv, hparams.f_clamp_kqv);
-                    cb(Kcur, "Kcur", il);
-                }
-                ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
-                if (hparams.f_clamp_kqv > 0.0f) {
-                    Vcur = ggml_clamp(ctx0, Vcur, -hparams.f_clamp_kqv, hparams.f_clamp_kqv);
-                    cb(Vcur, "Vcur", il);
-                }
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
-
-                Qcur = ggml_rope_ext(
-                        ctx0, Qcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-
-                Kcur = ggml_rope_ext(
-                        ctx0, Kcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
-
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, nullptr,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
-            }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // feed-forward network
-            cur = build_norm(ffn_inp,
-                    NULL, NULL,
-                    LLM_NORM, il);
-            cb(cur, "ffn_norm", il);
-
-            cur = build_ffn(cur,
-                    model.layers[il].ffn_up,   NULL, NULL,
-                    model.layers[il].ffn_gate, NULL, NULL,
-                    model.layers[il].ffn_down, NULL, NULL,
-                    NULL,
-                    LLM_FFN_SILU, LLM_FFN_PAR, il);
-            cb(cur, "ffn_out", il);
-
-            cur = ggml_add(ctx0, cur, ffn_inp);
-            cb(cur, "ffn_out", il);
-
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
-        }
-        cur = inpL;
-
-        cur = build_norm(cur,
+        // norm
+        cur = build_norm(inpL,
                 NULL, NULL,
-                LLM_NORM, -1);
+                LLM_NORM, il);
+        cb(cur, "attn_norm", il);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+        // self-attention
+        {
+            // compute Q and K and RoPE them
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            cb(Qcur, "Qcur", il);
+            if (hparams.f_clamp_kqv > 0.0f) {
+                Qcur = ggml_clamp(ctx0, Qcur, -hparams.f_clamp_kqv, hparams.f_clamp_kqv);
+                cb(Qcur, "Qcur", il);
+            }
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            cb(Kcur, "Kcur", il);
+            if (hparams.f_clamp_kqv > 0.0f) {
+                Kcur = ggml_clamp(ctx0, Kcur, -hparams.f_clamp_kqv, hparams.f_clamp_kqv);
+                cb(Kcur, "Kcur", il);
+            }
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            cb(Vcur, "Vcur", il);
+            if (hparams.f_clamp_kqv > 0.0f) {
+                Vcur = ggml_clamp(ctx0, Vcur, -hparams.f_clamp_kqv, hparams.f_clamp_kqv);
+                cb(Vcur, "Vcur", il);
+            }
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
 
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
+            Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+            Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-        ggml_build_forward_expand(gf, cur);
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
+
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, nullptr,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
+        }
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+        }
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+        cb(ffn_inp, "ffn_inp", il);
+
+        // feed-forward network
+        cur = build_norm(ffn_inp,
+                NULL, NULL,
+                LLM_NORM, il);
+        cb(cur, "ffn_norm", il);
+
+        cur = build_ffn(cur,
+                model.layers[il].ffn_up,   NULL, NULL,
+                model.layers[il].ffn_gate, NULL, NULL,
+                model.layers[il].ffn_down, NULL, NULL,
+                NULL,
+                LLM_FFN_SILU, LLM_FFN_PAR, il);
+        cb(cur, "ffn_out", il);
+
+        cur = ggml_add(ctx0, cur, ffn_inp);
+        cb(cur, "ffn_out", il);
+
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
+
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
+
+    cur = build_norm(cur,
+            NULL, NULL,
+            LLM_NORM, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/olmo2.cpp

@@ -1,150 +1,149 @@
 #include "models.h"
 
-
 template <bool iswa>
 llm_build_olmo2<iswa>::llm_build_olmo2(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_head = hparams.n_embd_head_v;
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        using inp_attn_type = std::conditional_t<iswa, llm_graph_input_attn_kv_iswa, llm_graph_input_attn_kv>;
-        inp_attn_type * inp_attn = nullptr;
+    using inp_attn_type = std::conditional_t<iswa, llm_graph_input_attn_kv_iswa, llm_graph_input_attn_kv>;
+    inp_attn_type * inp_attn = nullptr;
 
-        if constexpr (iswa) {
-            inp_attn = build_attn_inp_kv_iswa();
-        } else {
-            inp_attn = build_attn_inp_kv();
-        }
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    if constexpr (iswa) {
+        inp_attn = build_attn_inp_kv_iswa();
+    } else {
+        inp_attn = build_attn_inp_kv();
+    }
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            ggml_tensor * inpSA = inpL;
+    for (int il = 0; il < n_layer; ++il) {
+        ggml_tensor * inpSA = inpL;
 
-            cur = inpL;
-
-            // self_attention
-            {
-                // compute Q and K and RoPE them
-                ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
-                cb(Qcur, "Qcur", il);
-
-                ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
-
-                ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
-
-                Qcur = build_norm(Qcur, model.layers[il].attn_q_norm, NULL,
-                        LLM_NORM_RMS, il);
-                cb(Qcur, "Qcur_normed", il);
-
-                Kcur = build_norm(Kcur, model.layers[il].attn_k_norm, NULL,
-                        LLM_NORM_RMS, il);
-                cb(Kcur, "Kcur_normed", il);
-
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
-
-                const bool is_swa = hparams.is_swa(il);
-
-                if (is_swa) {
-                    // For sliding window layers, Olmo3 use regular rope with no yarn rope scaling.
-                    // This is achieved here by setting freq_scale and attn_factor to 1.
-                    // We also set ext_factor to 0 to avoid a few unnecessary computations.
-                    Qcur = ggml_rope_ext(
-                        ctx0, Qcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, 1.0,
-                        0.0, 1.0, beta_fast, beta_slow
-                        );
-
-                    Kcur = ggml_rope_ext(
-                        ctx0, Kcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, 1.0,
-                        0.0, 1.0, beta_fast, beta_slow
-                        );
-                } else {
-                    Qcur = ggml_rope_ext(
-                        ctx0, Qcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-
-                    Kcur = ggml_rope_ext(
-                        ctx0, Kcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-                }
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
-
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, NULL,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
-            }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
-            cur = build_norm(cur,
-                    model.layers[il].attn_post_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "attn_post_norm", il);
-
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // feed-forward network
-            cur = build_ffn(ffn_inp,
-                    model.layers[il].ffn_up,   NULL, NULL,
-                    model.layers[il].ffn_gate, NULL, NULL,
-                    model.layers[il].ffn_down, NULL, NULL,
-                    NULL,
-                    LLM_FFN_SILU, LLM_FFN_PAR, il);
-            cb(cur, "ffn_out", il);
-
-            cur = build_norm(cur,
-                    model.layers[il].ffn_post_norm, NULL,
-                    LLM_NORM_RMS, -1);
-            cb(cur, "ffn_post_norm", -1);
-
-            cur = ggml_add(ctx0, cur, ffn_inp);
-            cb(cur, "ffn_out", il);
-
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
-        }
         cur = inpL;
 
+        // self_attention
+        {
+            // compute Q and K and RoPE them
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            cb(Qcur, "Qcur", il);
+
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            cb(Kcur, "Kcur", il);
+
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            cb(Vcur, "Vcur", il);
+
+            Qcur = build_norm(Qcur, model.layers[il].attn_q_norm, NULL,
+                    LLM_NORM_RMS, il);
+            cb(Qcur, "Qcur_normed", il);
+
+            Kcur = build_norm(Kcur, model.layers[il].attn_k_norm, NULL,
+                    LLM_NORM_RMS, il);
+            cb(Kcur, "Kcur_normed", il);
+
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
+
+            const bool is_swa = hparams.is_swa(il);
+
+            if (is_swa) {
+                // For sliding window layers, Olmo3 use regular rope with no yarn rope scaling.
+                // This is achieved here by setting freq_scale and attn_factor to 1.
+                // We also set ext_factor to 0 to avoid a few unnecessary computations.
+                Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, 1.0,
+                    0.0, 1.0, beta_fast, beta_slow
+                    );
+
+                Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, 1.0,
+                    0.0, 1.0, beta_fast, beta_slow
+                    );
+            } else {
+                Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
+
+                Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
+            }
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
+
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, NULL,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
+        }
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+        }
         cur = build_norm(cur,
-                model.output_norm, NULL,
+                model.layers[il].attn_post_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_post_norm", il);
+
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+        cb(ffn_inp, "ffn_inp", il);
+
+        // feed-forward network
+        cur = build_ffn(ffn_inp,
+                model.layers[il].ffn_up,   NULL, NULL,
+                model.layers[il].ffn_gate, NULL, NULL,
+                model.layers[il].ffn_down, NULL, NULL,
+                NULL,
+                LLM_FFN_SILU, LLM_FFN_PAR, il);
+        cb(cur, "ffn_out", il);
+
+        cur = build_norm(cur,
+                model.layers[il].ffn_post_norm, NULL,
                 LLM_NORM_RMS, -1);
+        cb(cur, "ffn_post_norm", -1);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+        cur = ggml_add(ctx0, cur, ffn_inp);
+        cb(cur, "ffn_out", il);
 
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
 
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
-
-        ggml_build_forward_expand(gf, cur);
+        // input for next layer
+        inpL = cur;
     }
+    cur = inpL;
+
+    cur = build_norm(cur,
+            model.output_norm, NULL,
+            LLM_NORM_RMS, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
+}
 
 // Explicit template instantiations
 template struct llm_build_olmo2<false>;

src/models/olmoe.cpp

@@ -1,124 +1,124 @@
 #include "models.h"
 
 llm_build_olmoe::llm_build_olmoe(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_head = hparams.n_embd_head_v;
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        auto * inp_attn = build_attn_inp_kv();
+    auto * inp_attn = build_attn_inp_kv();
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            ggml_tensor * inpSA = inpL;
+    for (int il = 0; il < n_layer; ++il) {
+        ggml_tensor * inpSA = inpL;
 
-            // norm
-            cur = build_norm(inpL,
-                    model.layers[il].attn_norm, NULL,
+        // norm
+        cur = build_norm(inpL,
+                model.layers[il].attn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_norm", il);
+
+        // self_attention
+        {
+            // compute Q and K and RoPE them
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            cb(Qcur, "Qcur", il);
+
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            cb(Kcur, "Kcur", il);
+
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            cb(Vcur, "Vcur", il);
+
+            Qcur = build_norm(Qcur, model.layers[il].attn_q_norm, NULL,
                     LLM_NORM_RMS, il);
-            cb(cur, "attn_norm", il);
+            cb(Qcur, "Qcur_normed", il);
 
-            // self_attention
-            {
-                // compute Q and K and RoPE them
-                ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
-                cb(Qcur, "Qcur", il);
-
-                ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
-
-                ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
-
-                Qcur = build_norm(Qcur, model.layers[il].attn_q_norm, NULL,
-                        LLM_NORM_RMS, il);
-                cb(Qcur, "Qcur_normed", il);
-
-                Kcur = build_norm(Kcur, model.layers[il].attn_k_norm, NULL,
-                        LLM_NORM_RMS, il);
-                cb(Kcur, "Kcur_normed", il);
-
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
-
-                Qcur = ggml_rope_ext(
-                        ctx0, Qcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-
-                Kcur = ggml_rope_ext(
-                        ctx0, Kcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
-
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, NULL,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
-            }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // MoE branch
-            cur = build_norm(ffn_inp,
-                    model.layers[il].ffn_norm, NULL,
+            Kcur = build_norm(Kcur, model.layers[il].attn_k_norm, NULL,
                     LLM_NORM_RMS, il);
-            cb(cur, "ffn_norm", il);
+            cb(Kcur, "Kcur_normed", il);
 
-            cur = build_moe_ffn(cur,
-                    model.layers[il].ffn_gate_inp,
-                    model.layers[il].ffn_up_exps,
-                    model.layers[il].ffn_gate_exps,
-                    model.layers[il].ffn_down_exps,
-                    nullptr,
-                    n_expert, n_expert_used,
-                    LLM_FFN_SILU, false,
-                    false, 0.0,
-                    LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
-                    il);
-            cb(cur, "ffn_moe_out", il);
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
 
-            cur = ggml_add(ctx0, cur, ffn_inp);
+            Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
+            Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-            // input for next layer
-            inpL = cur;
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
+
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, NULL,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
         }
-        cur = inpL;
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+        }
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+        cb(ffn_inp, "ffn_inp", il);
 
-        cur = build_norm(cur,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, -1);
+        // MoE branch
+        cur = build_norm(ffn_inp,
+                model.layers[il].ffn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "ffn_norm", il);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+        cur = build_moe_ffn(cur,
+                model.layers[il].ffn_gate_inp,
+                model.layers[il].ffn_up_exps,
+                model.layers[il].ffn_gate_exps,
+                model.layers[il].ffn_down_exps,
+                nullptr,
+                n_expert, n_expert_used,
+                LLM_FFN_SILU, false,
+                false, 0.0,
+                LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
+                il);
+        cb(cur, "ffn_moe_out", il);
 
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
+        cur = ggml_add(ctx0, cur, ffn_inp);
 
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
 
-        ggml_build_forward_expand(gf, cur);
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
+
+    cur = build_norm(cur,
+            model.output_norm, NULL,
+            LLM_NORM_RMS, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/openai-moe-iswa.cpp

@@ -1,123 +1,123 @@
 #include "models.h"
 
 llm_build_openai_moe_iswa::llm_build_openai_moe_iswa(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        auto * inp_attn = build_attn_inp_kv_iswa();
+    auto * inp_attn = build_attn_inp_kv_iswa();
 
-        for (int il = 0; il < n_layer; ++il) {
-            ggml_tensor * inpSA = inpL;
+    for (int il = 0; il < n_layer; ++il) {
+        ggml_tensor * inpSA = inpL;
 
-            // norm
-            cur = build_norm(inpL,
-                    model.layers[il].attn_norm, nullptr,
-                    LLM_NORM_RMS, il);
-            cb(cur, "attn_norm", il);
+        // norm
+        cur = build_norm(inpL,
+                model.layers[il].attn_norm, nullptr,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_norm", il);
 
-            // self-attention
-            {
-                // compute Q and K and RoPE them
-                ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+        // self-attention
+        {
+            // compute Q and K and RoPE them
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            cb(Qcur, "Qcur", il);
+            if (model.layers[il].bq) {
+                Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
                 cb(Qcur, "Qcur", il);
-                if (model.layers[il].bq) {
-                    Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
-                    cb(Qcur, "Qcur", il);
-                }
-                ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
-                if (model.layers[il].bk) {
-                    Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
-                    cb(Kcur, "Kcur", il);
-                }
-                ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
-                if (model.layers[il].bv) {
-                    Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
-                    cb(Vcur, "Vcur", il);
-                }
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_rot, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_rot, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_rot, n_head_kv, n_tokens);
-
-                Qcur = ggml_rope_ext(
-                        ctx0, Qcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-
-                Kcur = ggml_rope_ext(
-                        ctx0, Kcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
-
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Qcur, Kcur, Vcur, nullptr, model.layers[il].attn_sinks, nullptr, 1.0f/sqrtf(float(n_rot)), il);
-
-                cb(cur, "attn_out", il);
             }
-            if (il == n_layer - 1) {
-                // skip computing output for unused tokens
-                ggml_tensor * inp_out_ids = build_inp_out_ids();
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            cb(Kcur, "Kcur", il);
+            if (model.layers[il].bk) {
+                Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
+                cb(Kcur, "Kcur", il);
             }
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            cb(Vcur, "Vcur", il);
+            if (model.layers[il].bv) {
+                Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
+                cb(Vcur, "Vcur", il);
+            }
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_rot, n_head,    n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_rot, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_rot, n_head_kv, n_tokens);
 
-            cur = ffn_inp;
-            cur = build_norm(cur,
-                    model.layers[il].attn_post_norm, nullptr,
-                    LLM_NORM_RMS, il);
-            cb(cur, "attn_post_norm", il);
+            Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-            // MoE branch
-            cur = build_moe_ffn(cur,
-                    model.layers[il].ffn_gate_inp,  model.layers[il].ffn_gate_inp_b,
-                    model.layers[il].ffn_up_exps,   model.layers[il].ffn_up_exps_b,
-                    model.layers[il].ffn_gate_exps, model.layers[il].ffn_gate_exps_b,
-                    model.layers[il].ffn_down_exps, model.layers[il].ffn_down_exps_b,
-                    nullptr,
-                    n_expert, n_expert_used,
-                    LLM_FFN_SWIGLU_OAI_MOE, false,
-                    false, 0.0,
-                    LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX_WEIGHT,
-                    il);
-            cb(cur, "ffn_moe_out", il);
+            Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-            cur = ggml_add(ctx0, cur, ffn_inp);
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
 
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, model.layers[il].bo,
+                    Qcur, Kcur, Vcur, nullptr, model.layers[il].attn_sinks, nullptr, 1.0f/sqrtf(float(n_rot)), il);
 
-            // input for next layer
-            inpL = cur;
+            cb(cur, "attn_out", il);
         }
-        cur = inpL;
+        if (il == n_layer - 1) {
+            // skip computing output for unused tokens
+            ggml_tensor * inp_out_ids = build_inp_out_ids();
+            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+        }
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+        cb(ffn_inp, "ffn_inp", il);
 
+        cur = ffn_inp;
         cur = build_norm(cur,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, -1);
+                model.layers[il].attn_post_norm, nullptr,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_post_norm", il);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+        // MoE branch
+        cur = build_moe_ffn(cur,
+                model.layers[il].ffn_gate_inp,  model.layers[il].ffn_gate_inp_b,
+                model.layers[il].ffn_up_exps,   model.layers[il].ffn_up_exps_b,
+                model.layers[il].ffn_gate_exps, model.layers[il].ffn_gate_exps_b,
+                model.layers[il].ffn_down_exps, model.layers[il].ffn_down_exps_b,
+                nullptr,
+                n_expert, n_expert_used,
+                LLM_FFN_SWIGLU_OAI_MOE, false,
+                false, 0.0,
+                LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX_WEIGHT,
+                il);
+        cb(cur, "ffn_moe_out", il);
 
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
+        cur = ggml_add(ctx0, cur, ffn_inp);
 
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
 
-        ggml_build_forward_expand(gf, cur);
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
+
+    cur = build_norm(cur,
+            model.output_norm, NULL,
+            LLM_NORM_RMS, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/openelm.cpp

@@ -1,124 +1,124 @@
 #include "models.h"
 
 llm_build_openelm::llm_build_openelm(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_head = hparams.n_embd_head_v;
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
-        inpL = build_inp_embd(model.tok_embd);
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
+    inpL = build_inp_embd(model.tok_embd);
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        auto * inp_attn = build_attn_inp_kv();
+    auto * inp_attn = build_attn_inp_kv();
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            const int64_t n_head    = hparams.n_head(il);
-            const int64_t n_head_kv = hparams.n_head_kv(il);
-            const int64_t n_head_qkv = 2*n_head_kv + n_head;
+    for (int il = 0; il < n_layer; ++il) {
+        const int64_t n_head    = hparams.n_head(il);
+        const int64_t n_head_kv = hparams.n_head_kv(il);
+        const int64_t n_head_qkv = 2*n_head_kv + n_head;
 
-            cur = inpL;
-            ggml_tensor * residual = cur;
-
-            // norm
-            cur = build_norm(inpL,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "attn_norm", il);
-
-            // self-attention
-            {
-                cur = build_lora_mm(model.layers[il].wqkv, cur);
-                cb(cur, "wqkv", il);
-
-                cur = ggml_reshape_3d(ctx0, cur, n_embd_head_k, n_head_qkv, n_tokens);
-
-                ggml_tensor * Qcur = ggml_view_3d(ctx0, cur, n_embd_head, n_head,    n_tokens, cur->nb[1], cur->nb[2], 0);
-                cb(Qcur, "Qcur", il);
-
-                ggml_tensor * Kcur = ggml_view_3d(ctx0, cur, n_embd_head, n_head_kv, n_tokens, cur->nb[1], cur->nb[2], cur->nb[1]*n_head);
-                cb(Kcur, "Kcur", il);
-
-                ggml_tensor * Vcur = ggml_cont(ctx0, ggml_view_3d(ctx0, cur, n_embd_head, n_head_kv, n_tokens, cur->nb[1], cur->nb[2], cur->nb[1]*(n_head+n_head_kv)));
-                cb(Vcur, "Vcur", il);
-
-                Qcur = build_norm(Qcur,
-                        model.layers[il].attn_q_norm, NULL,
-                        LLM_NORM_RMS, il);
-                cb(Qcur, "Qcur", il);
-
-                Kcur = build_norm(Kcur,
-                        model.layers[il].attn_k_norm, NULL,
-                        LLM_NORM_RMS, il);
-                cb(Kcur, "Kcur", il);
-
-                Qcur = ggml_rope_ext(
-                        ctx0, Qcur, inp_pos, NULL,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-
-                Kcur = ggml_rope_ext(
-                        ctx0, Kcur, inp_pos, NULL,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Qcur, "Vcur", il);
-
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, NULL,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
-            }
-            if (il == n_layer - 1 && inp_out_ids) {
-                residual = ggml_get_rows(ctx0, residual, inp_out_ids);
-                cur      = ggml_get_rows(ctx0, cur,      inp_out_ids);
-            }
-            ggml_tensor * ffn_inp = ggml_add(ctx0, residual, cur);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // feed-forward network
-            {
-                cur = build_norm(ffn_inp,
-                        model.layers[il].ffn_norm, NULL,
-                        LLM_NORM_RMS, il);
-                cb(cur, "ffn_norm", il);
-
-                cur = build_ffn(cur,
-                        model.layers[il].ffn_up,   NULL, NULL,
-                        model.layers[il].ffn_gate, NULL, NULL,
-                        model.layers[il].ffn_down, NULL, NULL,
-                        NULL,
-                        LLM_FFN_SILU, LLM_FFN_PAR, il);
-                cb(cur, "ffn_out", il);
-            }
-            cur = ggml_add(ctx0, cur, ffn_inp);
-
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            inpL = cur;
-        }
         cur = inpL;
+        ggml_tensor * residual = cur;
 
         // norm
-        cur = build_norm(cur,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, -1);
+        cur = build_norm(inpL,
+                model.layers[il].attn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_norm", il);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+        // self-attention
+        {
+            cur = build_lora_mm(model.layers[il].wqkv, cur);
+            cb(cur, "wqkv", il);
 
-        cur = build_lora_mm(model.output, cur);
+            cur = ggml_reshape_3d(ctx0, cur, n_embd_head_k, n_head_qkv, n_tokens);
 
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+            ggml_tensor * Qcur = ggml_view_3d(ctx0, cur, n_embd_head, n_head,    n_tokens, cur->nb[1], cur->nb[2], 0);
+            cb(Qcur, "Qcur", il);
 
-        ggml_build_forward_expand(gf, cur);
+            ggml_tensor * Kcur = ggml_view_3d(ctx0, cur, n_embd_head, n_head_kv, n_tokens, cur->nb[1], cur->nb[2], cur->nb[1]*n_head);
+            cb(Kcur, "Kcur", il);
+
+            ggml_tensor * Vcur = ggml_cont(ctx0, ggml_view_3d(ctx0, cur, n_embd_head, n_head_kv, n_tokens, cur->nb[1], cur->nb[2], cur->nb[1]*(n_head+n_head_kv)));
+            cb(Vcur, "Vcur", il);
+
+            Qcur = build_norm(Qcur,
+                    model.layers[il].attn_q_norm, NULL,
+                    LLM_NORM_RMS, il);
+            cb(Qcur, "Qcur", il);
+
+            Kcur = build_norm(Kcur,
+                    model.layers[il].attn_k_norm, NULL,
+                    LLM_NORM_RMS, il);
+            cb(Kcur, "Kcur", il);
+
+            Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, NULL,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
+
+            Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, NULL,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
+
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Qcur, "Vcur", il);
+
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, NULL,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
+        }
+        if (il == n_layer - 1 && inp_out_ids) {
+            residual = ggml_get_rows(ctx0, residual, inp_out_ids);
+            cur      = ggml_get_rows(ctx0, cur,      inp_out_ids);
+        }
+        ggml_tensor * ffn_inp = ggml_add(ctx0, residual, cur);
+        cb(ffn_inp, "ffn_inp", il);
+
+        // feed-forward network
+        {
+            cur = build_norm(ffn_inp,
+                    model.layers[il].ffn_norm, NULL,
+                    LLM_NORM_RMS, il);
+            cb(cur, "ffn_norm", il);
+
+            cur = build_ffn(cur,
+                    model.layers[il].ffn_up,   NULL, NULL,
+                    model.layers[il].ffn_gate, NULL, NULL,
+                    model.layers[il].ffn_down, NULL, NULL,
+                    NULL,
+                    LLM_FFN_SILU, LLM_FFN_PAR, il);
+            cb(cur, "ffn_out", il);
+        }
+        cur = ggml_add(ctx0, cur, ffn_inp);
+
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
+
+        inpL = cur;
+    }
+    cur = inpL;
+
+    // norm
+    cur = build_norm(cur,
+            model.output_norm, NULL,
+            LLM_NORM_RMS, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    cur = build_lora_mm(model.output, cur);
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/orion.cpp

@@ -1,123 +1,123 @@
 #include "models.h"
 
 llm_build_orion::llm_build_orion(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_head = hparams.n_embd_head_v;
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        auto * inp_attn = build_attn_inp_kv();
+    auto * inp_attn = build_attn_inp_kv();
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            ggml_tensor * inpSA = inpL;
+    for (int il = 0; il < n_layer; ++il) {
+        ggml_tensor * inpSA = inpL;
 
-            // norm
-            cur = build_norm(inpL,
-                    model.layers[il].attn_norm, model.layers[il].attn_norm_b,
-                    LLM_NORM, il);
-            cb(cur, "attn_norm", il);
+        // norm
+        cur = build_norm(inpL,
+                model.layers[il].attn_norm, model.layers[il].attn_norm_b,
+                LLM_NORM, il);
+        cb(cur, "attn_norm", il);
 
-            // self-attention
-            {
-                // compute Q and K and RoPE them
-                ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
-                cb(Qcur, "Qcur", il);
-                // if (model.layers[il].bq) {
-                //     Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
-                //     cb(Qcur, "Qcur", il);
-                // }
+        // self-attention
+        {
+            // compute Q and K and RoPE them
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            cb(Qcur, "Qcur", il);
+            // if (model.layers[il].bq) {
+            //     Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
+            //     cb(Qcur, "Qcur", il);
+            // }
 
-                ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
-                // if (model.layers[il].bk) {
-                //     Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
-                //     cb(Kcur, "Kcur", il);
-                // }
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            cb(Kcur, "Kcur", il);
+            // if (model.layers[il].bk) {
+            //     Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
+            //     cb(Kcur, "Kcur", il);
+            // }
 
-                ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
-                // if (model.layers[il].bv) {
-                //     Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
-                //     cb(Vcur, "Vcur", il);
-                // }
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            cb(Vcur, "Vcur", il);
+            // if (model.layers[il].bv) {
+            //     Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
+            //     cb(Vcur, "Vcur", il);
+            // }
 
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
 
-                Qcur = ggml_rope_ext(
-                        ctx0, Qcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
+            Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-                Kcur = ggml_rope_ext(
-                        ctx0, Kcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
+            Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
 
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, NULL,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
-            }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // feed-forward network
-            cur = build_norm(ffn_inp,
-                    model.layers[il].ffn_norm, model.layers[il].ffn_norm_b,
-                    LLM_NORM, il);
-            cb(cur, "ffn_norm", il);
-
-            cur = build_ffn(cur,
-                    model.layers[il].ffn_up,   NULL, NULL,
-                    model.layers[il].ffn_gate, NULL, NULL,
-                    model.layers[il].ffn_down, NULL, NULL,
-                    NULL,
-                    LLM_FFN_SILU, LLM_FFN_PAR, il);
-            cb(cur, "ffn_out", il);
-
-            cur = ggml_add(ctx0, cur, ffn_inp);
-
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, NULL,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
         }
-        cur = inpL;
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+        }
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+        cb(ffn_inp, "ffn_inp", il);
 
-        cur = build_norm(cur,
-                model.output_norm, model.output_norm_b,
-                LLM_NORM, -1);
+        // feed-forward network
+        cur = build_norm(ffn_inp,
+                model.layers[il].ffn_norm, model.layers[il].ffn_norm_b,
+                LLM_NORM, il);
+        cb(cur, "ffn_norm", il);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+        cur = build_ffn(cur,
+                model.layers[il].ffn_up,   NULL, NULL,
+                model.layers[il].ffn_gate, NULL, NULL,
+                model.layers[il].ffn_down, NULL, NULL,
+                NULL,
+                LLM_FFN_SILU, LLM_FFN_PAR, il);
+        cb(cur, "ffn_out", il);
 
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
+        cur = ggml_add(ctx0, cur, ffn_inp);
 
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
 
-        ggml_build_forward_expand(gf, cur);
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
+
+    cur = build_norm(cur,
+            model.output_norm, model.output_norm_b,
+            LLM_NORM, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/pangu-embedded.cpp

@@ -0,0 +1,121 @@
+#include "models.h"
+
+
+llm_build_pangu_embedded::llm_build_pangu_embedded(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
+    const int64_t n_embd_head = hparams.n_embd_head_v;
+
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_rot);
+
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
+
+    inpL = build_inp_embd(model.tok_embd);
+
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
+
+    auto * inp_attn = build_attn_inp_kv();
+
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
+
+    for (int il = 0; il < n_layer; ++il) {
+        ggml_tensor * inpSA = inpL;
+
+        // norm
+        cur = build_norm(inpL,
+                model.layers[il].attn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_norm", il);
+
+        // self attention
+        {
+            // compute Q and K and RoPE them
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
+            cb(Qcur, "Qcur", il);
+
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
+            cb(Kcur, "Kcur", il);
+
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
+            cb(Vcur, "Vcur", il);
+
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
+
+            Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
+
+            Kcur = ggml_rope_ext(ctx0, Kcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
+
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
+
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, model.layers[il].bo,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
+        }
+
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+        }
+
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+        cb(ffn_inp, "ffn_inp", il);
+
+        // feed-forward network
+        cur = build_norm(ffn_inp,
+                model.layers[il].ffn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "ffn_norm", il);
+
+        cur = build_ffn(cur,
+                model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
+                model.layers[il].ffn_gate, model.layers[il].ffn_gate_b, NULL,
+                model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
+                NULL,
+                LLM_FFN_SILU, LLM_FFN_PAR, il);
+
+        cur = ggml_add(ctx0, cur, ffn_inp);
+        cb(cur, "ffn_out", il);
+
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
+
+        // input for next layer
+        inpL = cur;
+    }
+
+    cur = inpL;
+
+    cur = build_norm(cur,
+            model.output_norm, NULL,
+            LLM_NORM_RMS, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+
+    if (model.output_b != nullptr) {
+        cur = ggml_add(ctx0, cur, model.output_b);
+    }
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
+}

src/models/phi3.cpp

@@ -1,152 +1,151 @@
 #include "models.h"
 
-
 template<bool iswa>
 llm_build_phi3<iswa>::llm_build_phi3(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
-        const int64_t n_embd_gqa = hparams.n_embd_v_gqa();
+    const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_gqa = hparams.n_embd_v_gqa();
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        using inp_attn_type = std::conditional_t<iswa, llm_graph_input_attn_kv_iswa, llm_graph_input_attn_kv>;
-        inp_attn_type * inp_attn = nullptr;
+    using inp_attn_type = std::conditional_t<iswa, llm_graph_input_attn_kv_iswa, llm_graph_input_attn_kv>;
+    inp_attn_type * inp_attn = nullptr;
 
-        if constexpr (iswa) {
-            inp_attn = build_attn_inp_kv_iswa();
-        } else {
-            inp_attn = build_attn_inp_kv();
-        }
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
-
-        for (int il = 0; il < n_layer; ++il) {
-            auto * residual = inpL;
-
-            // self-attention
-            {
-                // rope freq factors for 128k context
-                ggml_tensor * rope_factors = model.get_rope_factors(cparams, il);
-
-                ggml_tensor* attn_norm_output = build_norm(inpL,
-                        model.layers[il].attn_norm,
-                        model.layers[il].attn_norm_b,
-                        LLM_NORM_RMS, il);
-                cb(attn_norm_output, "attn_norm", il);
-
-                ggml_tensor * Qcur = nullptr;
-                ggml_tensor * Kcur = nullptr;
-                ggml_tensor * Vcur = nullptr;
-
-                if (model.layers[il].wqkv) {
-                    cur = build_lora_mm(model.layers[il].wqkv, attn_norm_output);
-                    cb(cur, "wqkv", il);
-
-                    Qcur = ggml_view_3d(ctx0, cur, n_embd_head, n_head,    n_tokens, n_embd_head * sizeof(float), cur->nb[1], 0 * sizeof(float) * (n_embd));
-                    Kcur = ggml_view_3d(ctx0, cur, n_embd_head, n_head_kv, n_tokens, n_embd_head * sizeof(float), cur->nb[1], 1 * sizeof(float) * (n_embd));
-                    Vcur = ggml_view_3d(ctx0, cur, n_embd_head, n_head_kv, n_tokens, n_embd_head * sizeof(float), cur->nb[1], 1 * sizeof(float) * (n_embd + n_embd_gqa));
-                    }
-                    else {
-                    Qcur = ggml_add(ctx0, build_lora_mm(model.layers[il].wq, attn_norm_output), model.layers[il].bq);
-                    Kcur = ggml_add(ctx0, build_lora_mm(model.layers[il].wk, attn_norm_output), model.layers[il].bk);
-                    Vcur = ggml_add(ctx0, build_lora_mm(model.layers[il].wv, attn_norm_output), model.layers[il].bv);
-
-                    Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                    Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                    Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
-                }
-                Qcur = ggml_rope_ext(
-                        ctx0, Qcur, inp_pos, rope_factors,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-
-                Kcur = ggml_rope_ext(
-                        ctx0, Kcur, inp_pos, rope_factors,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
-
-                Qcur = ggml_scale(ctx0, Qcur, 1.0f / sqrtf(float(n_embd_head)));
-                cb(Qcur, "Qcur", il);
-
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f, il);
-            }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur      = ggml_get_rows(ctx0, cur,      inp_out_ids);
-                residual = ggml_get_rows(ctx0, residual, inp_out_ids);
-            }
-            cur = ggml_add(ctx0, cur, residual);
-            residual = cur;
-
-            cur = build_norm(cur,
-                    model.layers[il].ffn_norm, model.layers[il].ffn_norm_b,
-                    LLM_NORM_RMS, il);
-            cb(cur, "ffn_norm", il);
-
-            // feed-forward network
-            if (model.layers[il].ffn_gate_inp == nullptr) {
-                cur = build_ffn(cur,
-                        model.layers[il].ffn_up,   NULL, NULL,
-                        NULL,                      NULL, NULL,
-                        model.layers[il].ffn_down, NULL, NULL,
-                        NULL,
-                        LLM_FFN_SWIGLU, LLM_FFN_SEQ, il);
-                cb(cur, "ffn_out", il);
-            } else {
-                // MoE branch
-                cur = build_moe_ffn(cur,
-                        model.layers[il].ffn_gate_inp,
-                        model.layers[il].ffn_up_exps,
-                        model.layers[il].ffn_gate_exps,
-                        model.layers[il].ffn_down_exps,
-                        nullptr,
-                        n_expert, n_expert_used,
-                        LLM_FFN_SILU, true,
-                        false, 0.0,
-                        LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
-                        il);
-                cb(cur, "ffn_moe_out", il);
-            }
-            cur = ggml_add(ctx0, residual, cur);
-
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
-        }
-        cur = build_norm(inpL,
-                model.output_norm,
-                model.output_norm_b,
-                LLM_NORM_RMS, -1);
-
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
-
-        cur = build_lora_mm(model.output, cur);
-
-        if (model.output_b != nullptr) {
-            cb(cur, "result_output_no_bias", -1);
-            cur = ggml_add(ctx0, cur, model.output_b);
-        }
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
-
-        ggml_build_forward_expand(gf, cur);
+    if constexpr (iswa) {
+        inp_attn = build_attn_inp_kv_iswa();
+    } else {
+        inp_attn = build_attn_inp_kv();
     }
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
+
+    for (int il = 0; il < n_layer; ++il) {
+        auto * residual = inpL;
+
+        // self-attention
+        {
+            // rope freq factors for 128k context
+            ggml_tensor * rope_factors = model.get_rope_factors(cparams, il);
+
+            ggml_tensor* attn_norm_output = build_norm(inpL,
+                    model.layers[il].attn_norm,
+                    model.layers[il].attn_norm_b,
+                    LLM_NORM_RMS, il);
+            cb(attn_norm_output, "attn_norm", il);
+
+            ggml_tensor * Qcur = nullptr;
+            ggml_tensor * Kcur = nullptr;
+            ggml_tensor * Vcur = nullptr;
+
+            if (model.layers[il].wqkv) {
+                cur = build_lora_mm(model.layers[il].wqkv, attn_norm_output);
+                cb(cur, "wqkv", il);
+
+                Qcur = ggml_view_3d(ctx0, cur, n_embd_head, n_head,    n_tokens, n_embd_head * sizeof(float), cur->nb[1], 0 * sizeof(float) * (n_embd));
+                Kcur = ggml_view_3d(ctx0, cur, n_embd_head, n_head_kv, n_tokens, n_embd_head * sizeof(float), cur->nb[1], 1 * sizeof(float) * (n_embd));
+                Vcur = ggml_view_3d(ctx0, cur, n_embd_head, n_head_kv, n_tokens, n_embd_head * sizeof(float), cur->nb[1], 1 * sizeof(float) * (n_embd + n_embd_gqa));
+                }
+                else {
+                Qcur = ggml_add(ctx0, build_lora_mm(model.layers[il].wq, attn_norm_output), model.layers[il].bq);
+                Kcur = ggml_add(ctx0, build_lora_mm(model.layers[il].wk, attn_norm_output), model.layers[il].bk);
+                Vcur = ggml_add(ctx0, build_lora_mm(model.layers[il].wv, attn_norm_output), model.layers[il].bv);
+
+                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
+            }
+            Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, rope_factors,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
+
+            Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, rope_factors,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
+
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
+
+            Qcur = ggml_scale(ctx0, Qcur, 1.0f / sqrtf(float(n_embd_head)));
+            cb(Qcur, "Qcur", il);
+
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, model.layers[il].bo,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f, il);
+        }
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur      = ggml_get_rows(ctx0, cur,      inp_out_ids);
+            residual = ggml_get_rows(ctx0, residual, inp_out_ids);
+        }
+        cur = ggml_add(ctx0, cur, residual);
+        residual = cur;
+
+        cur = build_norm(cur,
+                model.layers[il].ffn_norm, model.layers[il].ffn_norm_b,
+                LLM_NORM_RMS, il);
+        cb(cur, "ffn_norm", il);
+
+        // feed-forward network
+        if (model.layers[il].ffn_gate_inp == nullptr) {
+            cur = build_ffn(cur,
+                    model.layers[il].ffn_up,   NULL, NULL,
+                    NULL,                      NULL, NULL,
+                    model.layers[il].ffn_down, NULL, NULL,
+                    NULL,
+                    LLM_FFN_SWIGLU, LLM_FFN_SEQ, il);
+            cb(cur, "ffn_out", il);
+        } else {
+            // MoE branch
+            cur = build_moe_ffn(cur,
+                    model.layers[il].ffn_gate_inp,
+                    model.layers[il].ffn_up_exps,
+                    model.layers[il].ffn_gate_exps,
+                    model.layers[il].ffn_down_exps,
+                    nullptr,
+                    n_expert, n_expert_used,
+                    LLM_FFN_SILU, true,
+                    false, 0.0,
+                    LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
+                    il);
+            cb(cur, "ffn_moe_out", il);
+        }
+        cur = ggml_add(ctx0, residual, cur);
+
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
+
+        // input for next layer
+        inpL = cur;
+    }
+    cur = build_norm(inpL,
+            model.output_norm,
+            model.output_norm_b,
+            LLM_NORM_RMS, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    cur = build_lora_mm(model.output, cur);
+
+    if (model.output_b != nullptr) {
+        cb(cur, "result_output_no_bias", -1);
+        cur = ggml_add(ctx0, cur, model.output_b);
+    }
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
+}
 
 // Explicit template instantiations
 template struct llm_build_phi3<false>;

src/models/plamo.cpp

@@ -1,110 +1,110 @@
 #include "models.h"
 
 llm_build_plamo::llm_build_plamo(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_head = hparams.n_embd_head_v;
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        auto * inp_attn = build_attn_inp_kv();
+    auto * inp_attn = build_attn_inp_kv();
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            // norm
-            cur = build_norm(inpL,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "attn_norm", il);
+    for (int il = 0; il < n_layer; ++il) {
+        // norm
+        cur = build_norm(inpL,
+                model.layers[il].attn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_norm", il);
 
-            ggml_tensor * sa_inp = cur;
+        ggml_tensor * sa_inp = cur;
 
-            // self-attention
-            {
-                // compute Q and K and RoPE them
-                ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
-                cb(Qcur, "Qcur", il);
+        // self-attention
+        {
+            // compute Q and K and RoPE them
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            cb(Qcur, "Qcur", il);
 
-                ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            cb(Kcur, "Kcur", il);
 
-                ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            cb(Vcur, "Vcur", il);
 
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
 
-                Qcur = ggml_rope_ext(
-                        ctx0, Qcur, inp_pos, nullptr,
-                        n_embd_head, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
+            Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, nullptr,
+                    n_embd_head, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-                Kcur = ggml_rope_ext(
-                        ctx0, Kcur, inp_pos, nullptr,
-                        n_embd_head, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
+            Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, nullptr,
+                    n_embd_head, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
 
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, NULL,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
-            }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur    = ggml_get_rows(ctx0,    cur, inp_out_ids);
-                sa_inp = ggml_get_rows(ctx0, sa_inp, inp_out_ids);
-                inpL   = ggml_get_rows(ctx0,   inpL, inp_out_ids);
-            }
-            ggml_tensor * sa_out = cur;
-
-            cur = sa_inp;
-
-            // feed-forward network
-            {
-                cur = build_ffn(cur,
-                        model.layers[il].ffn_up,   NULL, NULL,
-                        model.layers[il].ffn_gate, NULL, NULL,
-                        model.layers[il].ffn_down, NULL, NULL,
-                        NULL,
-                        LLM_FFN_SILU, LLM_FFN_PAR, il);
-                cb(cur, "ffn_out", il);
-            }
-            cur = ggml_add(ctx0, cur, sa_out);
-            cur = ggml_add(ctx0, cur, inpL);
-
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, NULL,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
         }
-        cur = inpL;
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur    = ggml_get_rows(ctx0,    cur, inp_out_ids);
+            sa_inp = ggml_get_rows(ctx0, sa_inp, inp_out_ids);
+            inpL   = ggml_get_rows(ctx0,   inpL, inp_out_ids);
+        }
+        ggml_tensor * sa_out = cur;
 
-        cur = build_norm(cur,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, -1);
+        cur = sa_inp;
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+        // feed-forward network
+        {
+            cur = build_ffn(cur,
+                    model.layers[il].ffn_up,   NULL, NULL,
+                    model.layers[il].ffn_gate, NULL, NULL,
+                    model.layers[il].ffn_down, NULL, NULL,
+                    NULL,
+                    LLM_FFN_SILU, LLM_FFN_PAR, il);
+            cb(cur, "ffn_out", il);
+        }
+        cur = ggml_add(ctx0, cur, sa_out);
+        cur = ggml_add(ctx0, cur, inpL);
 
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
 
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
 
-        ggml_build_forward_expand(gf, cur);
+    cur = build_norm(cur,
+            model.output_norm, NULL,
+            LLM_NORM_RMS, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/plm.cpp

@@ -1,168 +1,168 @@
 #include "models.h"
 
 llm_build_plm::llm_build_plm(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const float kq_scale = 1.0f/sqrtf(float(hparams.n_embd_head_k));
+    const float kq_scale = 1.0f/sqrtf(float(hparams.n_embd_head_k));
 
-        const uint32_t n_embd_head_qk_rope = hparams.n_rot;
-        const uint32_t n_embd_head_qk_nope = hparams.n_embd_head_k - hparams.n_rot;
-        const uint32_t kv_lora_rank = hparams.n_lora_kv;
+    const uint32_t n_embd_head_qk_rope = hparams.n_rot;
+    const uint32_t n_embd_head_qk_nope = hparams.n_embd_head_k - hparams.n_rot;
+    const uint32_t kv_lora_rank = hparams.n_lora_kv;
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        // {n_embd, n_tokens}
-        inpL = build_inp_embd(model.tok_embd);
+    // {n_embd, n_tokens}
+    inpL = build_inp_embd(model.tok_embd);
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        auto * inp_attn = build_attn_inp_kv();
+    auto * inp_attn = build_attn_inp_kv();
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            ggml_tensor * inpSA = inpL;
+    for (int il = 0; il < n_layer; ++il) {
+        ggml_tensor * inpSA = inpL;
 
-            // norm
-            cur = build_norm(inpL,
-                    model.layers[il].attn_norm, NULL,
+        // norm
+        cur = build_norm(inpL,
+                model.layers[il].attn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_norm", il);
+
+        // self_attention
+        {
+            ggml_tensor * q = NULL;
+            q = ggml_mul_mat(ctx0, model.layers[il].wq, cur);
+            cb(q, "q", il);
+
+            // split into {n_head * n_embd_head_qk_nope, n_tokens}
+            ggml_tensor * q_nope = ggml_view_3d(ctx0, q, n_embd_head_qk_nope, n_head, n_tokens,
+                    ggml_row_size(q->type, hparams.n_embd_head_k),
+                    ggml_row_size(q->type, hparams.n_embd_head_k * n_head),
+                    0);
+            cb(q_nope, "q_nope", il);
+
+            // and {n_head * n_embd_head_qk_rope, n_tokens}
+            ggml_tensor * q_pe = ggml_view_3d(ctx0, q, n_embd_head_qk_rope, n_head, n_tokens,
+                    ggml_row_size(q->type, hparams.n_embd_head_k),
+                    ggml_row_size(q->type, hparams.n_embd_head_k * n_head),
+                    ggml_row_size(q->type, n_embd_head_qk_nope));
+            cb(q_pe, "q_pe", il);
+
+            // {n_embd, kv_lora_rank + n_embd_head_qk_rope} * {n_embd, n_tokens} -> {kv_lora_rank + n_embd_head_qk_rope, n_tokens}
+            ggml_tensor * kv_pe_compresseed = ggml_mul_mat(ctx0, model.layers[il].wkv_a_mqa, cur);
+            cb(kv_pe_compresseed, "kv_pe_compresseed", il);
+
+            // split into {kv_lora_rank, n_tokens}
+            ggml_tensor * kv_compressed = ggml_view_2d(ctx0, kv_pe_compresseed, kv_lora_rank, n_tokens,
+                    kv_pe_compresseed->nb[1],
+                    0);
+            cb(kv_compressed, "kv_compressed", il);
+
+            // and {n_embd_head_qk_rope, n_tokens}
+            ggml_tensor * k_pe = ggml_view_3d(ctx0, kv_pe_compresseed, n_embd_head_qk_rope, 1, n_tokens,
+                    kv_pe_compresseed->nb[1],
+                    kv_pe_compresseed->nb[1],
+                    ggml_row_size(kv_pe_compresseed->type, kv_lora_rank));
+            cb(k_pe, "k_pe", il);
+
+            kv_compressed = build_norm(kv_compressed,
+                    model.layers[il].attn_kv_a_norm, NULL,
                     LLM_NORM_RMS, il);
-            cb(cur, "attn_norm", il);
+            cb(kv_compressed, "kv_compressed", il);
 
-            // self_attention
-            {
-                ggml_tensor * q = NULL;
-                q = ggml_mul_mat(ctx0, model.layers[il].wq, cur);
-                cb(q, "q", il);
+            // {kv_lora_rank, n_head * (n_embd_head_qk_nope + n_embd_head_v)} * {kv_lora_rank, n_tokens} -> {n_head * (n_embd_head_qk_nope + n_embd_head_v), n_tokens}
+            ggml_tensor * kv = ggml_mul_mat(ctx0, model.layers[il].wkv_b, kv_compressed);
+            cb(kv, "kv", il);
 
-                // split into {n_head * n_embd_head_qk_nope, n_tokens}
-                ggml_tensor * q_nope = ggml_view_3d(ctx0, q, n_embd_head_qk_nope, n_head, n_tokens,
-                        ggml_row_size(q->type, hparams.n_embd_head_k),
-                        ggml_row_size(q->type, hparams.n_embd_head_k * n_head),
-                        0);
-                cb(q_nope, "q_nope", il);
+            // split into {n_head * n_embd_head_qk_nope, n_tokens}
+            ggml_tensor * k_nope = ggml_view_3d(ctx0, kv, n_embd_head_qk_nope, n_head, n_tokens,
+                    ggml_row_size(kv->type, n_embd_head_qk_nope + hparams.n_embd_head_v),
+                    ggml_row_size(kv->type, n_head * (n_embd_head_qk_nope + hparams.n_embd_head_v)),
+                    0);
+            cb(k_nope, "k_nope", il);
 
-                // and {n_head * n_embd_head_qk_rope, n_tokens}
-                ggml_tensor * q_pe = ggml_view_3d(ctx0, q, n_embd_head_qk_rope, n_head, n_tokens,
-                        ggml_row_size(q->type, hparams.n_embd_head_k),
-                        ggml_row_size(q->type, hparams.n_embd_head_k * n_head),
-                        ggml_row_size(q->type, n_embd_head_qk_nope));
-                cb(q_pe, "q_pe", il);
+            // and {n_head * n_embd_head_v, n_tokens}
+            ggml_tensor * v_states = ggml_view_3d(ctx0, kv, hparams.n_embd_head_v, n_head, n_tokens,
+                    ggml_row_size(kv->type, (n_embd_head_qk_nope + hparams.n_embd_head_v)),
+                    ggml_row_size(kv->type, (n_embd_head_qk_nope + hparams.n_embd_head_v)*n_head),
+                    ggml_row_size(kv->type, (n_embd_head_qk_nope)));
+            cb(v_states, "v_states", il);
 
-                // {n_embd, kv_lora_rank + n_embd_head_qk_rope} * {n_embd, n_tokens} -> {kv_lora_rank + n_embd_head_qk_rope, n_tokens}
-                ggml_tensor * kv_pe_compresseed = ggml_mul_mat(ctx0, model.layers[il].wkv_a_mqa, cur);
-                cb(kv_pe_compresseed, "kv_pe_compresseed", il);
+            v_states = ggml_cont(ctx0, v_states);
+            cb(v_states, "v_states", il);
 
-                // split into {kv_lora_rank, n_tokens}
-                ggml_tensor * kv_compressed = ggml_view_2d(ctx0, kv_pe_compresseed, kv_lora_rank, n_tokens,
-                        kv_pe_compresseed->nb[1],
-                        0);
-                cb(kv_compressed, "kv_compressed", il);
+            v_states = ggml_view_2d(ctx0, v_states, hparams.n_embd_head_v * n_head, n_tokens,
+                    ggml_row_size(kv->type, hparams.n_embd_head_v * n_head),
+                    0);
+            cb(v_states, "v_states", il);
 
-                // and {n_embd_head_qk_rope, n_tokens}
-                ggml_tensor * k_pe = ggml_view_3d(ctx0, kv_pe_compresseed, n_embd_head_qk_rope, 1, n_tokens,
-                        kv_pe_compresseed->nb[1],
-                        kv_pe_compresseed->nb[1],
-                        ggml_row_size(kv_pe_compresseed->type, kv_lora_rank));
-                cb(k_pe, "k_pe", il);
+            q_pe = ggml_rope_ext(
+                    ctx0, q_pe, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
+            cb(q_pe, "q_pe", il);
 
-                kv_compressed = build_norm(kv_compressed,
-                        model.layers[il].attn_kv_a_norm, NULL,
-                        LLM_NORM_RMS, il);
-                cb(kv_compressed, "kv_compressed", il);
+            // shared RoPE key
+            k_pe = ggml_rope_ext(
+                    ctx0, k_pe, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
+            cb(k_pe, "k_pe", il);
 
-                // {kv_lora_rank, n_head * (n_embd_head_qk_nope + n_embd_head_v)} * {kv_lora_rank, n_tokens} -> {n_head * (n_embd_head_qk_nope + n_embd_head_v), n_tokens}
-                ggml_tensor * kv = ggml_mul_mat(ctx0, model.layers[il].wkv_b, kv_compressed);
-                cb(kv, "kv", il);
+            ggml_tensor * q_states = ggml_concat(ctx0, q_nope, q_pe, 0);
+            cb(q_states, "q_states", il);
 
-                // split into {n_head * n_embd_head_qk_nope, n_tokens}
-                ggml_tensor * k_nope = ggml_view_3d(ctx0, kv, n_embd_head_qk_nope, n_head, n_tokens,
-                        ggml_row_size(kv->type, n_embd_head_qk_nope + hparams.n_embd_head_v),
-                        ggml_row_size(kv->type, n_head * (n_embd_head_qk_nope + hparams.n_embd_head_v)),
-                        0);
-                cb(k_nope, "k_nope", il);
+            ggml_tensor * k_states = ggml_concat(ctx0, k_nope, ggml_repeat(ctx0, k_pe, q_pe), 0);
+            cb(k_states, "k_states", il);
 
-                // and {n_head * n_embd_head_v, n_tokens}
-                ggml_tensor * v_states = ggml_view_3d(ctx0, kv, hparams.n_embd_head_v, n_head, n_tokens,
-                        ggml_row_size(kv->type, (n_embd_head_qk_nope + hparams.n_embd_head_v)),
-                        ggml_row_size(kv->type, (n_embd_head_qk_nope + hparams.n_embd_head_v)*n_head),
-                        ggml_row_size(kv->type, (n_embd_head_qk_nope)));
-                cb(v_states, "v_states", il);
-
-                v_states = ggml_cont(ctx0, v_states);
-                cb(v_states, "v_states", il);
-
-                v_states = ggml_view_2d(ctx0, v_states, hparams.n_embd_head_v * n_head, n_tokens,
-                        ggml_row_size(kv->type, hparams.n_embd_head_v * n_head),
-                        0);
-                cb(v_states, "v_states", il);
-
-                q_pe = ggml_rope_ext(
-                        ctx0, q_pe, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-                cb(q_pe, "q_pe", il);
-
-                // shared RoPE key
-                k_pe = ggml_rope_ext(
-                        ctx0, k_pe, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-                cb(k_pe, "k_pe", il);
-
-                ggml_tensor * q_states = ggml_concat(ctx0, q_nope, q_pe, 0);
-                cb(q_states, "q_states", il);
-
-                ggml_tensor * k_states = ggml_concat(ctx0, k_nope, ggml_repeat(ctx0, k_pe, q_pe), 0);
-                cb(k_states, "k_states", il);
-
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, NULL,
-                        q_states, k_states, v_states, nullptr, nullptr, nullptr, kq_scale, il);
-            }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
-
-            cur = build_norm(ffn_inp,
-                    model.layers[il].ffn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "ffn_norm", il);
-
-            cur = build_ffn(cur,
-                    model.layers[il].ffn_up,   NULL, NULL,
-                    NULL, NULL, NULL,
-                    model.layers[il].ffn_down, NULL, NULL,
-                    NULL,
-                    LLM_FFN_RELU_SQR, LLM_FFN_SEQ, il);
-            cb(cur, "ffn_out", il);
-
-            cur = ggml_add(ctx0, cur, ffn_inp);
-
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, NULL,
+                    q_states, k_states, v_states, nullptr, nullptr, nullptr, kq_scale, il);
         }
-        cur = inpL;
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+        }
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+        cb(ffn_inp, "ffn_inp", il);
 
-        cur = build_norm(cur,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, -1);
+        cur = build_norm(ffn_inp,
+                model.layers[il].ffn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "ffn_norm", il);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+        cur = build_ffn(cur,
+                model.layers[il].ffn_up,   NULL, NULL,
+                NULL, NULL, NULL,
+                model.layers[il].ffn_down, NULL, NULL,
+                NULL,
+                LLM_FFN_RELU_SQR, LLM_FFN_SEQ, il);
+        cb(cur, "ffn_out", il);
 
-        cur = build_lora_mm(model.output, cur);
+        cur = ggml_add(ctx0, cur, ffn_inp);
 
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
 
-        ggml_build_forward_expand(gf, cur);
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
+
+    cur = build_norm(cur,
+            model.output_norm, NULL,
+            LLM_NORM_RMS, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    cur = build_lora_mm(model.output, cur);
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/qwen2.cpp

@@ -1,118 +1,117 @@
 #include "models.h"
 
-
 llm_build_qwen2::llm_build_qwen2(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_head = hparams.n_embd_head_v;
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        auto * inp_attn = build_attn_inp_kv();
+    auto * inp_attn = build_attn_inp_kv();
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            ggml_tensor * inpSA = inpL;
+    for (int il = 0; il < n_layer; ++il) {
+        ggml_tensor * inpSA = inpL;
 
-            // norm
-            cur = build_norm(inpL,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "attn_norm", il);
+        // norm
+        cur = build_norm(inpL,
+                model.layers[il].attn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_norm", il);
 
-            // self-attention
-            {
-                // compute Q and K and RoPE them
-                ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
-                Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
-                cb(Qcur, "Qcur", il);
+        // self-attention
+        {
+            // compute Q and K and RoPE them
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
+            cb(Qcur, "Qcur", il);
 
-                ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
-                Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
-                cb(Kcur, "Kcur", il);
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
+            cb(Kcur, "Kcur", il);
 
-                ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
-                Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
-                cb(Vcur, "Vcur", il);
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
+            cb(Vcur, "Vcur", il);
 
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
 
-                Qcur = ggml_rope_ext(
-                        ctx0, Qcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
+            Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-                Kcur = ggml_rope_ext(
-                        ctx0, Kcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
+            Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
 
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
-            }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // feed-forward network
-            cur = build_norm(ffn_inp,
-                    model.layers[il].ffn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "ffn_norm", il);
-
-            cur = build_ffn(cur,
-                    model.layers[il].ffn_up,   NULL, NULL,
-                    model.layers[il].ffn_gate, NULL, NULL,
-                    model.layers[il].ffn_down, NULL, NULL,
-                    NULL,
-                    LLM_FFN_SILU, LLM_FFN_PAR, il);
-            cb(cur, "ffn_out", il);
-
-            cur = ggml_add(ctx0, cur, ffn_inp);
-
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, model.layers[il].bo,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
         }
-        cur = inpL;
-
-        cur = build_norm(cur,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, -1);
-
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
-
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
-
-        if (model.output_b != nullptr) {
-            cur = ggml_add(ctx0, cur, model.output_b);
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
         }
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+        cb(ffn_inp, "ffn_inp", il);
 
-        ggml_build_forward_expand(gf, cur);
+        // feed-forward network
+        cur = build_norm(ffn_inp,
+                model.layers[il].ffn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "ffn_norm", il);
+
+        cur = build_ffn(cur,
+                model.layers[il].ffn_up,   NULL, NULL,
+                model.layers[il].ffn_gate, NULL, NULL,
+                model.layers[il].ffn_down, NULL, NULL,
+                NULL,
+                LLM_FFN_SILU, LLM_FFN_PAR, il);
+        cb(cur, "ffn_out", il);
+
+        cur = ggml_add(ctx0, cur, ffn_inp);
+
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
+
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
+
+    cur = build_norm(cur,
+            model.output_norm, NULL,
+            LLM_NORM_RMS, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+
+    if (model.output_b != nullptr) {
+        cur = ggml_add(ctx0, cur, model.output_b);
+    }
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/qwen2moe.cpp

@@ -1,151 +1,151 @@
 #include "models.h"
 
 llm_build_qwen2moe::llm_build_qwen2moe(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_head = hparams.n_embd_head_v;
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        auto * inp_attn = build_attn_inp_kv();
+    auto * inp_attn = build_attn_inp_kv();
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            ggml_tensor * inpSA = inpL;
+    for (int il = 0; il < n_layer; ++il) {
+        ggml_tensor * inpSA = inpL;
 
-            // norm
-            cur = build_norm(inpL,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "attn_norm", il);
+        // norm
+        cur = build_norm(inpL,
+                model.layers[il].attn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_norm", il);
 
-            // self_attention
-            {
-                // compute Q and K and RoPE them
-                ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+        // self_attention
+        {
+            // compute Q and K and RoPE them
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            cb(Qcur, "Qcur", il);
+            if (model.layers[il].bq) {
+                Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
                 cb(Qcur, "Qcur", il);
-                if (model.layers[il].bq) {
-                    Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
-                    cb(Qcur, "Qcur", il);
-                }
-                ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            }
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            cb(Kcur, "Kcur", il);
+            if (model.layers[il].bk) {
+                Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
                 cb(Kcur, "Kcur", il);
-                if (model.layers[il].bk) {
-                    Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
-                    cb(Kcur, "Kcur", il);
-                }
-                ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            }
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            cb(Vcur, "Vcur", il);
+            if (model.layers[il].bv) {
+                Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
                 cb(Vcur, "Vcur", il);
-                if (model.layers[il].bv) {
-                    Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
-                    cb(Vcur, "Vcur", il);
-                }
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
-
-                Qcur = ggml_rope_ext(
-                        ctx0, Qcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-
-                Kcur = ggml_rope_ext(
-                        ctx0, Kcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
-
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
             }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
 
-            // MoE branch
-            cur = build_norm(ffn_inp,
-                    model.layers[il].ffn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "ffn_norm", il);
+            Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-            ggml_tensor * moe_out =
-                build_moe_ffn(cur,
-                        model.layers[il].ffn_gate_inp,
-                        model.layers[il].ffn_up_exps,
-                        model.layers[il].ffn_gate_exps,
-                        model.layers[il].ffn_down_exps,
-                        nullptr,
-                        n_expert, n_expert_used,
-                        LLM_FFN_SILU, false,
-                        false, 0.0,
-                        LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
-                        il);
-            cb(moe_out, "ffn_moe_out", il);
+            Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-            // FFN shared expert
-            {
-                ggml_tensor * cur_gate_inp = build_lora_mm(model.layers[il].ffn_gate_inp_shexp, cur);
-                cb(cur_gate_inp, "ffn_shexp_gate_inp", il);
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
 
-                // sigmoid
-                ggml_tensor * cur_gate = ggml_div(ctx0, ggml_silu(ctx0, cur_gate_inp), cur_gate_inp);
-                cb(cur_gate, "ffn_shexp_gate", il);
-
-                ggml_tensor * cur_ffn = build_ffn(cur,
-                        model.layers[il].ffn_up_shexp,   NULL, NULL,
-                        model.layers[il].ffn_gate_shexp, NULL, NULL,
-                        model.layers[il].ffn_down_shexp, NULL, NULL,
-                        NULL,
-                        LLM_FFN_SILU, LLM_FFN_PAR, il);
-                cb(cur_ffn, "ffn_shexp", il);
-
-                ggml_tensor * ffn_shexp_out = ggml_mul(ctx0, cur_ffn, cur_gate);
-                cb(ffn_shexp_out, "ffn_shexp_out", il);
-
-                moe_out = ggml_add(ctx0, moe_out, ffn_shexp_out);
-                cb(moe_out, "ffn_out", il);
-
-                cur = moe_out;
-            }
-            cur = ggml_add(ctx0, cur, ffn_inp);
-
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, model.layers[il].bo,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
         }
-        cur = inpL;
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+        }
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+        cb(ffn_inp, "ffn_inp", il);
 
-        cur = build_norm(cur,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, -1);
+        // MoE branch
+        cur = build_norm(ffn_inp,
+                model.layers[il].ffn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "ffn_norm", il);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+        ggml_tensor * moe_out =
+            build_moe_ffn(cur,
+                    model.layers[il].ffn_gate_inp,
+                    model.layers[il].ffn_up_exps,
+                    model.layers[il].ffn_gate_exps,
+                    model.layers[il].ffn_down_exps,
+                    nullptr,
+                    n_expert, n_expert_used,
+                    LLM_FFN_SILU, false,
+                    false, 0.0,
+                    LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
+                    il);
+        cb(moe_out, "ffn_moe_out", il);
 
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
+        // FFN shared expert
+        {
+            ggml_tensor * cur_gate_inp = build_lora_mm(model.layers[il].ffn_gate_inp_shexp, cur);
+            cb(cur_gate_inp, "ffn_shexp_gate_inp", il);
 
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+            // sigmoid
+            ggml_tensor * cur_gate = ggml_div(ctx0, ggml_silu(ctx0, cur_gate_inp), cur_gate_inp);
+            cb(cur_gate, "ffn_shexp_gate", il);
 
-        ggml_build_forward_expand(gf, cur);
+            ggml_tensor * cur_ffn = build_ffn(cur,
+                    model.layers[il].ffn_up_shexp,   NULL, NULL,
+                    model.layers[il].ffn_gate_shexp, NULL, NULL,
+                    model.layers[il].ffn_down_shexp, NULL, NULL,
+                    NULL,
+                    LLM_FFN_SILU, LLM_FFN_PAR, il);
+            cb(cur_ffn, "ffn_shexp", il);
+
+            ggml_tensor * ffn_shexp_out = ggml_mul(ctx0, cur_ffn, cur_gate);
+            cb(ffn_shexp_out, "ffn_shexp_out", il);
+
+            moe_out = ggml_add(ctx0, moe_out, ffn_shexp_out);
+            cb(moe_out, "ffn_out", il);
+
+            cur = moe_out;
+        }
+        cur = ggml_add(ctx0, cur, ffn_inp);
+
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
+
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
+
+    cur = build_norm(cur,
+            model.output_norm, NULL,
+            LLM_NORM_RMS, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/qwen2vl.cpp

@@ -1,117 +1,117 @@
 #include "models.h"
 
 llm_build_qwen2vl::llm_build_qwen2vl(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_head = hparams.n_embd_head_v;
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        auto * inp_attn = build_attn_inp_kv();
+    auto * inp_attn = build_attn_inp_kv();
 
-        int sections[4];
-        std::copy(std::begin(hparams.rope_sections), std::begin(hparams.rope_sections) + 4, sections);
+    int sections[4];
+    std::copy(std::begin(hparams.rope_sections), std::begin(hparams.rope_sections) + 4, sections);
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            ggml_tensor * inpSA = inpL;
+    for (int il = 0; il < n_layer; ++il) {
+        ggml_tensor * inpSA = inpL;
 
-            // norm
-            cur = build_norm(inpL,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "attn_norm", il);
+        // norm
+        cur = build_norm(inpL,
+                model.layers[il].attn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_norm", il);
 
-            // self-attention
-            {
-                // compute Q and K and RoPE them
-                ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
-                Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
-                cb(Qcur, "Qcur", il);
+        // self-attention
+        {
+            // compute Q and K and RoPE them
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
+            cb(Qcur, "Qcur", il);
 
-                ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
-                Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
-                cb(Kcur, "Kcur", il);
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
+            cb(Kcur, "Kcur", il);
 
-                ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
-                Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
-                cb(Vcur, "Vcur", il);
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
+            cb(Vcur, "Vcur", il);
 
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
 
-                Qcur = ggml_rope_multi(
-                        ctx0, Qcur, inp_pos, nullptr,
-                        n_rot, sections, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
+            Qcur = ggml_rope_multi(
+                    ctx0, Qcur, inp_pos, nullptr,
+                    n_rot, sections, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-                Kcur = ggml_rope_multi(
-                        ctx0, Kcur, inp_pos, nullptr,
-                        n_rot, sections, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
+            Kcur = ggml_rope_multi(
+                    ctx0, Kcur, inp_pos, nullptr,
+                    n_rot, sections, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
 
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
-            }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // feed-forward network
-            cur = build_norm(ffn_inp,
-                    model.layers[il].ffn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "ffn_norm", il);
-
-            cur = build_ffn(cur,
-                    model.layers[il].ffn_up,   NULL, NULL,
-                    model.layers[il].ffn_gate, NULL, NULL,
-                    model.layers[il].ffn_down, NULL, NULL,
-                    NULL,
-                    LLM_FFN_SILU, LLM_FFN_PAR, il);
-            cb(cur, "ffn_out", il);
-
-            cur = ggml_add(ctx0, cur, ffn_inp);
-
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, model.layers[il].bo,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
         }
-        cur = inpL;
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+        }
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+        cb(ffn_inp, "ffn_inp", il);
 
-        cur = build_norm(cur,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, -1);
+        // feed-forward network
+        cur = build_norm(ffn_inp,
+                model.layers[il].ffn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "ffn_norm", il);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+        cur = build_ffn(cur,
+                model.layers[il].ffn_up,   NULL, NULL,
+                model.layers[il].ffn_gate, NULL, NULL,
+                model.layers[il].ffn_down, NULL, NULL,
+                NULL,
+                LLM_FFN_SILU, LLM_FFN_PAR, il);
+        cb(cur, "ffn_out", il);
 
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
+        cur = ggml_add(ctx0, cur, ffn_inp);
 
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
 
-        ggml_build_forward_expand(gf, cur);
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
+
+    cur = build_norm(cur,
+            model.output_norm, NULL,
+            LLM_NORM_RMS, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/qwen3.cpp

@@ -1,117 +1,117 @@
 #include "models.h"
 
 llm_build_qwen3::llm_build_qwen3(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_head = hparams.n_embd_head_v;
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        auto * inp_attn = build_attn_inp_kv();
+    auto * inp_attn = build_attn_inp_kv();
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            ggml_tensor * inpSA = inpL;
+    for (int il = 0; il < n_layer; ++il) {
+        ggml_tensor * inpSA = inpL;
 
-            // norm
-            cur = build_norm(inpL,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "attn_norm", il);
+        // norm
+        cur = build_norm(inpL,
+                model.layers[il].attn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_norm", il);
 
-            // self-attention
-            {
-                // compute Q and K and RoPE them
-                ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
-                cb(Qcur, "Qcur", il);
+        // self-attention
+        {
+            // compute Q and K and RoPE them
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            cb(Qcur, "Qcur", il);
 
-                ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            cb(Kcur, "Kcur", il);
 
-                ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            cb(Vcur, "Vcur", il);
 
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
 
-                Qcur = build_norm(Qcur, model.layers[il].attn_q_norm, NULL, LLM_NORM_RMS, il);
-                cb(Qcur, "Qcur_normed", il);
+            Qcur = build_norm(Qcur, model.layers[il].attn_q_norm, NULL, LLM_NORM_RMS, il);
+            cb(Qcur, "Qcur_normed", il);
 
-                Qcur = ggml_rope_ext(
-                        ctx0, Qcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
+            Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-                Kcur = build_norm(Kcur, model.layers[il].attn_k_norm, NULL, LLM_NORM_RMS, il);
-                cb(Kcur, "Kcur_normed", il);
+            Kcur = build_norm(Kcur, model.layers[il].attn_k_norm, NULL, LLM_NORM_RMS, il);
+            cb(Kcur, "Kcur_normed", il);
 
-                Kcur = ggml_rope_ext(
-                        ctx0, Kcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
+            Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
 
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
-            }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // feed-forward network
-            cur = build_norm(ffn_inp,
-                    model.layers[il].ffn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "ffn_norm", il);
-
-            cur = build_ffn(cur,
-                    model.layers[il].ffn_up,   NULL, NULL,
-                    model.layers[il].ffn_gate, NULL, NULL,
-                    model.layers[il].ffn_down, NULL, NULL,
-                    NULL,
-                    LLM_FFN_SILU, LLM_FFN_PAR, il);
-            cb(cur, "ffn_out", il);
-
-            cur = ggml_add(ctx0, cur, ffn_inp);
-
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, model.layers[il].bo,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
         }
-        cur = inpL;
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+        }
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+        cb(ffn_inp, "ffn_inp", il);
 
-        cur = build_norm(cur,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, -1);
+        // feed-forward network
+        cur = build_norm(ffn_inp,
+                model.layers[il].ffn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "ffn_norm", il);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+        cur = build_ffn(cur,
+                model.layers[il].ffn_up,   NULL, NULL,
+                model.layers[il].ffn_gate, NULL, NULL,
+                model.layers[il].ffn_down, NULL, NULL,
+                NULL,
+                LLM_FFN_SILU, LLM_FFN_PAR, il);
+        cb(cur, "ffn_out", il);
 
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
+        cur = ggml_add(ctx0, cur, ffn_inp);
 
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
 
-        ggml_build_forward_expand(gf, cur);
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
+
+    cur = build_norm(cur,
+            model.output_norm, NULL,
+            LLM_NORM_RMS, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/qwen3moe.cpp

@@ -1,124 +1,124 @@
 #include "models.h"
 
 llm_build_qwen3moe::llm_build_qwen3moe(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_head = hparams.n_embd_head_v;
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        auto * inp_attn = build_attn_inp_kv();
+    auto * inp_attn = build_attn_inp_kv();
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            ggml_tensor * inpSA = inpL;
+    for (int il = 0; il < n_layer; ++il) {
+        ggml_tensor * inpSA = inpL;
 
-            // norm
-            cur = build_norm(inpL,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "attn_norm", il);
+        // norm
+        cur = build_norm(inpL,
+                model.layers[il].attn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_norm", il);
 
-            // self_attention
-            {
-                // compute Q and K and RoPE them
-                ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
-                cb(Qcur, "Qcur", il);
+        // self_attention
+        {
+            // compute Q and K and RoPE them
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            cb(Qcur, "Qcur", il);
 
-                ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            cb(Kcur, "Kcur", il);
 
-                ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            cb(Vcur, "Vcur", il);
 
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
 
-                Qcur = build_norm(Qcur, model.layers[il].attn_q_norm, NULL, LLM_NORM_RMS, il);
-                cb(Qcur, "Qcur_normed", il);
+            Qcur = build_norm(Qcur, model.layers[il].attn_q_norm, NULL, LLM_NORM_RMS, il);
+            cb(Qcur, "Qcur_normed", il);
 
-                Qcur = ggml_rope_ext(
-                        ctx0, Qcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
+            Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-                Kcur = build_norm(Kcur, model.layers[il].attn_k_norm, NULL, LLM_NORM_RMS, il);
-                cb(Kcur, "Kcur_normed", il);
+            Kcur = build_norm(Kcur, model.layers[il].attn_k_norm, NULL, LLM_NORM_RMS, il);
+            cb(Kcur, "Kcur_normed", il);
 
-                Kcur = ggml_rope_ext(
-                        ctx0, Kcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
+            Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
 
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
-            }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // MoE branch
-            cur = build_norm(ffn_inp,
-                    model.layers[il].ffn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "ffn_norm", il);
-
-            ggml_tensor * moe_out =
-                build_moe_ffn(cur,
-                        model.layers[il].ffn_gate_inp,
-                        model.layers[il].ffn_up_exps,
-                        model.layers[il].ffn_gate_exps,
-                        model.layers[il].ffn_down_exps,
-                        nullptr,
-                        n_expert, n_expert_used,
-                        LLM_FFN_SILU, true,
-                        false, 0.0,
-                        LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
-                        il);
-            cb(moe_out, "ffn_moe_out", il);
-            cur = moe_out;
-
-            cur = ggml_add(ctx0, cur, ffn_inp);
-
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, model.layers[il].bo,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
         }
-        cur = inpL;
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+        }
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+        cb(ffn_inp, "ffn_inp", il);
 
-        cur = build_norm(cur,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, -1);
+        // MoE branch
+        cur = build_norm(ffn_inp,
+                model.layers[il].ffn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "ffn_norm", il);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+        ggml_tensor * moe_out =
+            build_moe_ffn(cur,
+                    model.layers[il].ffn_gate_inp,
+                    model.layers[il].ffn_up_exps,
+                    model.layers[il].ffn_gate_exps,
+                    model.layers[il].ffn_down_exps,
+                    nullptr,
+                    n_expert, n_expert_used,
+                    LLM_FFN_SILU, true,
+                    false, 0.0,
+                    LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
+                    il);
+        cb(moe_out, "ffn_moe_out", il);
+        cur = moe_out;
 
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
+        cur = ggml_add(ctx0, cur, ffn_inp);
 
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
 
-        ggml_build_forward_expand(gf, cur);
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
+
+    cur = build_norm(cur,
+            model.output_norm, NULL,
+            LLM_NORM_RMS, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/refact.cpp

@@ -1,94 +1,94 @@
 #include "models.h"
 
 llm_build_refact::llm_build_refact(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_head = hparams.n_embd_head_v;
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        auto * inp_attn = build_attn_inp_kv();
+    auto * inp_attn = build_attn_inp_kv();
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            ggml_tensor * inpSA = inpL;
+    for (int il = 0; il < n_layer; ++il) {
+        ggml_tensor * inpSA = inpL;
 
-            cur = build_norm(inpL,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "attn_norm", il);
+        cur = build_norm(inpL,
+                model.layers[il].attn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_norm", il);
 
-            // self-attention
-            {
-                ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
-                cb(Qcur, "Qcur", il);
+        // self-attention
+        {
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            cb(Qcur, "Qcur", il);
 
-                ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            cb(Kcur, "Kcur", il);
 
-                ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            cb(Vcur, "Vcur", il);
 
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
 
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
 
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, NULL,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
-            }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // feed-forward network
-            {
-                cur = build_norm(ffn_inp,
-                        model.layers[il].ffn_norm, NULL,
-                        LLM_NORM_RMS, il);
-                cb(cur, "ffn_norm", il);
-
-                cur = build_ffn(cur,
-                        model.layers[il].ffn_up,   NULL, NULL,
-                        model.layers[il].ffn_gate, NULL, NULL,
-                        model.layers[il].ffn_down, NULL, NULL,
-                        NULL,
-                        LLM_FFN_SILU, LLM_FFN_PAR, il);
-                cb(cur, "ffn_out", il);
-            }
-            cur = ggml_add(ctx0, cur, ffn_inp);
-
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, NULL,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
         }
-        cur = inpL;
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+        }
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+        cb(ffn_inp, "ffn_inp", il);
 
-        cur = build_norm(cur,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, -1);
+        // feed-forward network
+        {
+            cur = build_norm(ffn_inp,
+                    model.layers[il].ffn_norm, NULL,
+                    LLM_NORM_RMS, il);
+            cb(cur, "ffn_norm", il);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+            cur = build_ffn(cur,
+                    model.layers[il].ffn_up,   NULL, NULL,
+                    model.layers[il].ffn_gate, NULL, NULL,
+                    model.layers[il].ffn_down, NULL, NULL,
+                    NULL,
+                    LLM_FFN_SILU, LLM_FFN_PAR, il);
+            cb(cur, "ffn_out", il);
+        }
+        cur = ggml_add(ctx0, cur, ffn_inp);
 
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
 
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
 
-        ggml_build_forward_expand(gf, cur);
+    cur = build_norm(cur,
+            model.output_norm, NULL,
+            LLM_NORM_RMS, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/seed-oss.cpp

@@ -1,124 +1,124 @@
 #include "models.h"
 
 llm_build_seed_oss::llm_build_seed_oss(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_head = hparams.n_embd_head_v;
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        auto * inp_attn = build_attn_inp_kv();
+    auto * inp_attn = build_attn_inp_kv();
 
-        const float kq_scale = hparams.f_attention_scale == 0.0f ? 1.0f/sqrtf(float(n_embd_head)) : hparams.f_attention_scale;
+    const float kq_scale = hparams.f_attention_scale == 0.0f ? 1.0f/sqrtf(float(n_embd_head)) : hparams.f_attention_scale;
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            ggml_tensor * inpSA = inpL;
+    for (int il = 0; il < n_layer; ++il) {
+        ggml_tensor * inpSA = inpL;
 
-            // norm
-            cur = build_norm(inpL,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "attn_norm", il);
+        // norm
+        cur = build_norm(inpL,
+                model.layers[il].attn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_norm", il);
 
-            // self-attention
-            {
-                // compute Q and K and RoPE them
-                ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+        // self-attention
+        {
+            // compute Q and K and RoPE them
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            cb(Qcur, "Qcur", il);
+            if (model.layers[il].bq) {
+                Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
                 cb(Qcur, "Qcur", il);
-                if (model.layers[il].bq) {
-                    Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
-                    cb(Qcur, "Qcur", il);
-                }
-                ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
-                if (model.layers[il].bk) {
-                    Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
-                    cb(Kcur, "Kcur", il);
-                }
-                ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
-                if (model.layers[il].bv) {
-                    Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
-                    cb(Vcur, "Vcur", il);
-                }
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
-
-                Qcur = ggml_rope_ext(
-                        ctx0, Qcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-
-                Kcur = ggml_rope_ext(
-                        ctx0, Kcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
-
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, kq_scale, il);
-                cb(cur, "attn_out", il);
             }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            cb(Kcur, "Kcur", il);
+            if (model.layers[il].bk) {
+                Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
+                cb(Kcur, "Kcur", il);
             }
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            cb(Vcur, "Vcur", il);
+            if (model.layers[il].bv) {
+                Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
+                cb(Vcur, "Vcur", il);
+            }
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
 
-            // feed-forward network
-            cur = build_norm(ffn_inp,
-                    model.layers[il].attn_post_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "attn_post_norm", il);
+            Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-            cur = build_ffn(cur,
-                    model.layers[il].ffn_up,   NULL, NULL,
-                    model.layers[il].ffn_gate, NULL, NULL,
-                    model.layers[il].ffn_down, NULL, NULL,
-                    NULL,
-                    LLM_FFN_SILU, LLM_FFN_PAR, il);
-            cb(cur, "ffn_out", il);
+            Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-            cur = ggml_add(ctx0, cur, ffn_inp);
-            cb(cur, "ffn_out", il);
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
 
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, model.layers[il].bo,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, kq_scale, il);
+            cb(cur, "attn_out", il);
         }
-        cur = inpL;
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+        }
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+        cb(ffn_inp, "ffn_inp", il);
 
-        cur = build_norm(cur,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, -1);
+        // feed-forward network
+        cur = build_norm(ffn_inp,
+                model.layers[il].attn_post_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_post_norm", il);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+        cur = build_ffn(cur,
+                model.layers[il].ffn_up,   NULL, NULL,
+                model.layers[il].ffn_gate, NULL, NULL,
+                model.layers[il].ffn_down, NULL, NULL,
+                NULL,
+                LLM_FFN_SILU, LLM_FFN_PAR, il);
+        cb(cur, "ffn_out", il);
 
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
+        cur = ggml_add(ctx0, cur, ffn_inp);
+        cb(cur, "ffn_out", il);
 
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
 
-        ggml_build_forward_expand(gf, cur);
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
+
+    cur = build_norm(cur,
+            model.output_norm, NULL,
+            LLM_NORM_RMS, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/smallthinker.cpp

@@ -2,118 +2,118 @@
 
 template <bool iswa>
 llm_build_smallthinker<iswa>::llm_build_smallthinker(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params){
-        const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_head = hparams.n_embd_head_v;
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        using inp_attn_type = std::conditional_t<iswa, llm_graph_input_attn_kv_iswa, llm_graph_input_attn_kv>;
-        inp_attn_type * inp_attn = nullptr;
+    using inp_attn_type = std::conditional_t<iswa, llm_graph_input_attn_kv_iswa, llm_graph_input_attn_kv>;
+    inp_attn_type * inp_attn = nullptr;
 
-        if constexpr (iswa) {
-            inp_attn = build_attn_inp_kv_iswa();
-        } else {
-            inp_attn = build_attn_inp_kv();
-        }
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
-
-        for (int il = 0; il < n_layer; ++il) {
-            ggml_tensor * inpSA  = inpL;
-            ggml_tensor * probs  = nullptr;
-
-            probs = build_lora_mm(model.layers[il].ffn_gate_inp, inpL);  // [n_expert, n_tokens]
-            cb(probs, "ffn_moe_logits", il);
-
-            // norm
-            cur = build_norm(inpL,model.layers[il].attn_norm, NULL, LLM_NORM_RMS, il);
-            cb(cur, "attn_norm", il);
-
-            // self_attention
-            {
-                // compute Q and K and RoPE them
-                struct ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
-                cb(Qcur, "Qcur", il);
-
-                struct ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
-
-                struct ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
-
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
-
-                if (hparams.n_no_rope_layer_step == n_layer || il % hparams.n_no_rope_layer_step != 0) {
-                    Qcur = ggml_rope_ext(ctx0, Qcur, inp_pos, nullptr, n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                                     ext_factor, attn_factor, beta_fast, beta_slow);
-
-                    Kcur = ggml_rope_ext(ctx0, Kcur, inp_pos, nullptr, n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                                     ext_factor, attn_factor, beta_fast, beta_slow);
-                }
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f / sqrtf(float(n_embd_head)), il);
-            }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur = ggml_get_rows(ctx0, cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-                probs = ggml_get_rows(ctx0, probs, inp_out_ids);
-            }
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // MoE branch
-            cur = build_norm(ffn_inp, model.layers[il].ffn_norm, NULL, LLM_NORM_RMS, il);
-            cb(cur, "ffn_norm", il);
-
-            ggml_tensor * ffn_out =
-                build_moe_ffn(cur,
-                        nullptr,
-                        model.layers[il].ffn_up_exps,
-                        model.layers[il].ffn_gate_exps,
-                        model.layers[il].ffn_down_exps,
-                        nullptr,
-                        n_expert, n_expert_used,
-                        LLM_FFN_RELU, true,
-                        false, 0.0,
-                        static_cast<llama_expert_gating_func_type>(hparams.expert_gating_func),
-                        il, probs);
-
-            cb(ffn_out, "ffn_out", il);
-            cur = ffn_out;
-
-            cur = ggml_add(ctx0, cur, ffn_inp);
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
-        }
-        cur = inpL;
-
-        cur = build_norm(cur, model.output_norm, NULL, LLM_NORM_RMS, -1);
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
-
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
-
-        ggml_build_forward_expand(gf, cur);
+    if constexpr (iswa) {
+        inp_attn = build_attn_inp_kv_iswa();
+    } else {
+        inp_attn = build_attn_inp_kv();
     }
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
+
+    for (int il = 0; il < n_layer; ++il) {
+        ggml_tensor * inpSA  = inpL;
+        ggml_tensor * probs  = nullptr;
+
+        probs = build_lora_mm(model.layers[il].ffn_gate_inp, inpL);  // [n_expert, n_tokens]
+        cb(probs, "ffn_moe_logits", il);
+
+        // norm
+        cur = build_norm(inpL,model.layers[il].attn_norm, NULL, LLM_NORM_RMS, il);
+        cb(cur, "attn_norm", il);
+
+        // self_attention
+        {
+            // compute Q and K and RoPE them
+            struct ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            cb(Qcur, "Qcur", il);
+
+            struct ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            cb(Kcur, "Kcur", il);
+
+            struct ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            cb(Vcur, "Vcur", il);
+
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
+
+            if (hparams.n_no_rope_layer_step == n_layer || il % hparams.n_no_rope_layer_step != 0) {
+                Qcur = ggml_rope_ext(ctx0, Qcur, inp_pos, nullptr, n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                                    ext_factor, attn_factor, beta_fast, beta_slow);
+
+                Kcur = ggml_rope_ext(ctx0, Kcur, inp_pos, nullptr, n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                                    ext_factor, attn_factor, beta_fast, beta_slow);
+            }
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, model.layers[il].bo,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f / sqrtf(float(n_embd_head)), il);
+        }
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur = ggml_get_rows(ctx0, cur, inp_out_ids);
+            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+            probs = ggml_get_rows(ctx0, probs, inp_out_ids);
+        }
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+        cb(ffn_inp, "ffn_inp", il);
+
+        // MoE branch
+        cur = build_norm(ffn_inp, model.layers[il].ffn_norm, NULL, LLM_NORM_RMS, il);
+        cb(cur, "ffn_norm", il);
+
+        ggml_tensor * ffn_out =
+            build_moe_ffn(cur,
+                    nullptr,
+                    model.layers[il].ffn_up_exps,
+                    model.layers[il].ffn_gate_exps,
+                    model.layers[il].ffn_down_exps,
+                    nullptr,
+                    n_expert, n_expert_used,
+                    LLM_FFN_RELU, true,
+                    false, 0.0,
+                    static_cast<llama_expert_gating_func_type>(hparams.expert_gating_func),
+                    il, probs);
+
+        cb(ffn_out, "ffn_out", il);
+        cur = ffn_out;
+
+        cur = ggml_add(ctx0, cur, ffn_inp);
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
+
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
+
+    cur = build_norm(cur, model.output_norm, NULL, LLM_NORM_RMS, -1);
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
+}
 
 // Explicit template instantiations
 template struct llm_build_smallthinker<false>;

src/models/smollm3.cpp

@@ -1,128 +1,128 @@
 #include "models.h"
 
 llm_build_smollm3::llm_build_smollm3(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_head = hparams.n_embd_head_v;
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        auto * inp_attn = build_attn_inp_kv();
+    auto * inp_attn = build_attn_inp_kv();
 
-        const float kq_scale = hparams.f_attention_scale == 0.0f ? 1.0f/sqrtf(float(n_embd_head)) : hparams.f_attention_scale;
+    const float kq_scale = hparams.f_attention_scale == 0.0f ? 1.0f/sqrtf(float(n_embd_head)) : hparams.f_attention_scale;
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            ggml_tensor * inpSA = inpL;
+    for (int il = 0; il < n_layer; ++il) {
+        ggml_tensor * inpSA = inpL;
 
-            const bool use_rope = (il + 1) % hparams.n_no_rope_layer_step != 0;
+        const bool use_rope = (il + 1) % hparams.n_no_rope_layer_step != 0;
 
-            // norm
-            cur = build_norm(inpL,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "attn_norm", il);
+        // norm
+        cur = build_norm(inpL,
+                model.layers[il].attn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_norm", il);
 
-            // self-attention
-            {
-                // compute Q and K and RoPE them
-                ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+        // self-attention
+        {
+            // compute Q and K and RoPE them
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            cb(Qcur, "Qcur", il);
+            if (model.layers[il].bq) {
+                Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
                 cb(Qcur, "Qcur", il);
-                if (model.layers[il].bq) {
-                    Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
-                    cb(Qcur, "Qcur", il);
-                }
-                ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            }
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            cb(Kcur, "Kcur", il);
+            if (model.layers[il].bk) {
+                Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
                 cb(Kcur, "Kcur", il);
-                if (model.layers[il].bk) {
-                    Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
-                    cb(Kcur, "Kcur", il);
-                }
-                ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            }
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            cb(Vcur, "Vcur", il);
+            if (model.layers[il].bv) {
+                Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
                 cb(Vcur, "Vcur", il);
-                if (model.layers[il].bv) {
-                    Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
-                    cb(Vcur, "Vcur", il);
-                }
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
-
-                if (use_rope) {
-                    Qcur = ggml_rope_ext(
-                            ctx0, Qcur, inp_pos, nullptr,
-                            n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                            ext_factor, attn_factor, beta_fast, beta_slow
-                            );
-
-                    Kcur = ggml_rope_ext(
-                            ctx0, Kcur, inp_pos, nullptr,
-                            n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                            ext_factor, attn_factor, beta_fast, beta_slow
-                            );
-                }
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
-
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, kq_scale, il);
-                cb(cur, "attn_out", il);
             }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
+
+            if (use_rope) {
+                Qcur = ggml_rope_ext(
+                        ctx0, Qcur, inp_pos, nullptr,
+                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                        ext_factor, attn_factor, beta_fast, beta_slow
+                        );
+
+                Kcur = ggml_rope_ext(
+                        ctx0, Kcur, inp_pos, nullptr,
+                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                        ext_factor, attn_factor, beta_fast, beta_slow
+                        );
             }
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
 
-            // feed-forward network
-            {
-                cur = build_norm(ffn_inp,
-                        model.layers[il].ffn_norm, NULL,
-                        LLM_NORM_RMS, il);
-                cb(cur, "ffn_norm", il);
-
-                cur = build_ffn(cur,
-                        model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
-                        model.layers[il].ffn_gate, model.layers[il].ffn_gate_b, NULL,
-                        model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
-                        NULL,
-                        LLM_FFN_SILU, LLM_FFN_PAR, il);
-                cb(cur, "ffn_out", il);
-            }
-            cur = ggml_add(ctx0, cur, ffn_inp);
-            cb(cur, "ffn_out", il);
-
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, model.layers[il].bo,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, kq_scale, il);
+            cb(cur, "attn_out", il);
         }
-        cur = inpL;
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+        }
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+        cb(ffn_inp, "ffn_inp", il);
 
-        cur = build_norm(cur,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, -1);
+        // feed-forward network
+        {
+            cur = build_norm(ffn_inp,
+                    model.layers[il].ffn_norm, NULL,
+                    LLM_NORM_RMS, il);
+            cb(cur, "ffn_norm", il);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+            cur = build_ffn(cur,
+                    model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
+                    model.layers[il].ffn_gate, model.layers[il].ffn_gate_b, NULL,
+                    model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
+                    NULL,
+                    LLM_FFN_SILU, LLM_FFN_PAR, il);
+            cb(cur, "ffn_out", il);
+        }
+        cur = ggml_add(ctx0, cur, ffn_inp);
+        cb(cur, "ffn_out", il);
 
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
 
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
 
-        ggml_build_forward_expand(gf, cur);
+    cur = build_norm(cur,
+            model.output_norm, NULL,
+            LLM_NORM_RMS, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/starcoder.cpp

@@ -1,100 +1,100 @@
 #include "models.h"
 
 llm_build_starcoder::llm_build_starcoder(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
-        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
+    const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        auto * inp_attn = build_attn_inp_kv();
+    auto * inp_attn = build_attn_inp_kv();
 
-        ggml_tensor * pos = ggml_get_rows(ctx0, model.pos_embd, inp_pos);
-        cb(pos, "pos_embd", -1);
+    ggml_tensor * pos = ggml_get_rows(ctx0, model.pos_embd, inp_pos);
+    cb(pos, "pos_embd", -1);
 
-        inpL = ggml_add(ctx0, inpL, pos);
-        cb(inpL, "inpL", -1);
+    inpL = ggml_add(ctx0, inpL, pos);
+    cb(inpL, "inpL", -1);
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            cur = build_norm(inpL,
-                    model.layers[il].attn_norm,
-                    model.layers[il].attn_norm_b,
-                    LLM_NORM, il);
-            cb(cur, "attn_norm", il);
-
-            // self-attention
-            {
-                cur = build_lora_mm(model.layers[il].wqkv, cur);
-                cb(cur, "wqkv", il);
-
-                cur = ggml_add(ctx0, cur, model.layers[il].bqkv);
-                cb(cur, "bqkv", il);
-
-                ggml_tensor * Qcur = ggml_view_3d(ctx0, cur, n_embd_head, n_head,    n_tokens, n_embd_head*sizeof(float), cur->nb[1], 0*sizeof(float)*(n_embd));
-                ggml_tensor * Kcur = ggml_view_3d(ctx0, cur, n_embd_head, n_head_kv, n_tokens, n_embd_head*sizeof(float), cur->nb[1], 1*sizeof(float)*(n_embd));
-                ggml_tensor * Vcur = ggml_view_3d(ctx0, cur, n_embd_head, n_head_kv, n_tokens, n_embd_head*sizeof(float), cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa));
-
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
-
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
-            }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
-                inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
-            }
-            // add the input
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpL);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // FF
-            {
-                cur = build_norm(ffn_inp,
-                        model.layers[il].ffn_norm,
-                        model.layers[il].ffn_norm_b,
-                        LLM_NORM, il);
-                cb(cur, "ffn_norm", il);
-
-                cur = build_ffn(cur,
-                        model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
-                        NULL,                      NULL,                        NULL,
-                        model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
-                        NULL,
-                        LLM_FFN_GELU, LLM_FFN_SEQ, il);
-                cb(cur, "ffn_out", il);
-            }
-            cur = ggml_add(ctx0, cur, ffn_inp);
-
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
-        }
+    for (int il = 0; il < n_layer; ++il) {
         cur = build_norm(inpL,
-                model.output_norm,
-                model.output_norm_b,
-                LLM_NORM, -1);
+                model.layers[il].attn_norm,
+                model.layers[il].attn_norm_b,
+                LLM_NORM, il);
+        cb(cur, "attn_norm", il);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+        // self-attention
+        {
+            cur = build_lora_mm(model.layers[il].wqkv, cur);
+            cb(cur, "wqkv", il);
 
-        cur = build_lora_mm(model.output, cur);
+            cur = ggml_add(ctx0, cur, model.layers[il].bqkv);
+            cb(cur, "bqkv", il);
 
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+            ggml_tensor * Qcur = ggml_view_3d(ctx0, cur, n_embd_head, n_head,    n_tokens, n_embd_head*sizeof(float), cur->nb[1], 0*sizeof(float)*(n_embd));
+            ggml_tensor * Kcur = ggml_view_3d(ctx0, cur, n_embd_head, n_head_kv, n_tokens, n_embd_head*sizeof(float), cur->nb[1], 1*sizeof(float)*(n_embd));
+            ggml_tensor * Vcur = ggml_view_3d(ctx0, cur, n_embd_head, n_head_kv, n_tokens, n_embd_head*sizeof(float), cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa));
 
-        ggml_build_forward_expand(gf, cur);
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
+
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, model.layers[il].bo,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
+        }
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
+            inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
+        }
+        // add the input
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpL);
+        cb(ffn_inp, "ffn_inp", il);
+
+        // FF
+        {
+            cur = build_norm(ffn_inp,
+                    model.layers[il].ffn_norm,
+                    model.layers[il].ffn_norm_b,
+                    LLM_NORM, il);
+            cb(cur, "ffn_norm", il);
+
+            cur = build_ffn(cur,
+                    model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
+                    NULL,                      NULL,                        NULL,
+                    model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
+                    NULL,
+                    LLM_FFN_GELU, LLM_FFN_SEQ, il);
+            cb(cur, "ffn_out", il);
+        }
+        cur = ggml_add(ctx0, cur, ffn_inp);
+
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
+
+        // input for next layer
+        inpL = cur;
+    }
+    cur = build_norm(inpL,
+            model.output_norm,
+            model.output_norm_b,
+            LLM_NORM, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    cur = build_lora_mm(model.output, cur);
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/starcoder2.cpp

@@ -1,121 +1,121 @@
 #include "models.h"
 
 llm_build_starcoder2::llm_build_starcoder2(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_head = hparams.n_embd_head_v;
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        auto * inp_attn = build_attn_inp_kv();
+    auto * inp_attn = build_attn_inp_kv();
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            ggml_tensor * inpSA = inpL;
+    for (int il = 0; il < n_layer; ++il) {
+        ggml_tensor * inpSA = inpL;
 
-            // norm
-            cur = build_norm(inpL,
-                    model.layers[il].attn_norm, model.layers[il].attn_norm_b,
-                    LLM_NORM, il);
-            cb(cur, "attn_norm", il);
+        // norm
+        cur = build_norm(inpL,
+                model.layers[il].attn_norm, model.layers[il].attn_norm_b,
+                LLM_NORM, il);
+        cb(cur, "attn_norm", il);
 
-            // self-attention
-            {
-                // compute Q and K and RoPE them
-                ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+        // self-attention
+        {
+            // compute Q and K and RoPE them
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            cb(Qcur, "Qcur", il);
+            if (model.layers[il].bq) {
+                Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
                 cb(Qcur, "Qcur", il);
-                if (model.layers[il].bq) {
-                    Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
-                    cb(Qcur, "Qcur", il);
-                }
-                ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
-                if (model.layers[il].bk) {
-                    Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
-                    cb(Kcur, "Kcur", il);
-                }
-                ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
-                if (model.layers[il].bv) {
-                    Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
-                    cb(Vcur, "Vcur", il);
-                }
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
-
-                Qcur = ggml_rope_ext(
-                        ctx0, Qcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-
-                Kcur = ggml_rope_ext(
-                        ctx0, Kcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
-
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
             }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            cb(Kcur, "Kcur", il);
+            if (model.layers[il].bk) {
+                Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
+                cb(Kcur, "Kcur", il);
             }
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            cb(Vcur, "Vcur", il);
+            if (model.layers[il].bv) {
+                Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
+                cb(Vcur, "Vcur", il);
+            }
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
 
-            // feed-forward network
+            Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-            cur = build_norm(ffn_inp,
-                    model.layers[il].ffn_norm, model.layers[il].ffn_norm_b,
-                    LLM_NORM, il);
-            cb(cur, "ffn_norm", il);
+            Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-            cur = build_ffn(cur,
-                    model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
-                    NULL,                      NULL,                        NULL,
-                    model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
-                    NULL,
-                    LLM_FFN_GELU, LLM_FFN_SEQ, il);
-            cb(cur, "ffn_out", il);
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
 
-            cur = ggml_add(ctx0, cur, ffn_inp);
-
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, model.layers[il].bo,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
         }
-        cur = inpL;
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+        }
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+        cb(ffn_inp, "ffn_inp", il);
 
-        cur = build_norm(cur,
-                model.output_norm, model.output_norm_b,
-                LLM_NORM, -1);
+        // feed-forward network
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+        cur = build_norm(ffn_inp,
+                model.layers[il].ffn_norm, model.layers[il].ffn_norm_b,
+                LLM_NORM, il);
+        cb(cur, "ffn_norm", il);
 
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
+        cur = build_ffn(cur,
+                model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
+                NULL,                      NULL,                        NULL,
+                model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
+                NULL,
+                LLM_FFN_GELU, LLM_FFN_SEQ, il);
+        cb(cur, "ffn_out", il);
 
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+        cur = ggml_add(ctx0, cur, ffn_inp);
 
-        ggml_build_forward_expand(gf, cur);
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
+
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
+
+    cur = build_norm(cur,
+            model.output_norm, model.output_norm_b,
+            LLM_NORM, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/t5-dec.cpp

@@ -1,166 +1,166 @@
 #include "models.h"
 
 llm_build_t5_dec::llm_build_t5_dec(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
-        //const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
+    const int64_t n_embd_head = hparams.n_embd_head_v;
+    //const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        ggml_tensor * embd_enc       = build_inp_cross_embd();
-        ggml_tensor * pos_bucket_dec = build_inp_pos_bucket_dec();
+    ggml_tensor * embd_enc       = build_inp_cross_embd();
+    ggml_tensor * pos_bucket_dec = build_inp_pos_bucket_dec();
 
-        const int64_t n_outputs_enc = embd_enc->ne[1];
+    const int64_t n_outputs_enc = embd_enc->ne[1];
 
-        auto * inp_attn_self  = build_attn_inp_kv();
-        auto * inp_attn_cross = build_attn_inp_cross();
+    auto * inp_attn_self  = build_attn_inp_kv();
+    auto * inp_attn_cross = build_attn_inp_cross();
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        const int64_t dec_n_layer = hparams.dec_n_layer;
+    const int64_t dec_n_layer = hparams.dec_n_layer;
 
-        for (int il = 0; il < dec_n_layer; ++il) {
-            ggml_tensor * inpSA = inpL;
+    for (int il = 0; il < dec_n_layer; ++il) {
+        ggml_tensor * inpSA = inpL;
 
-            // norm
-            cur = build_norm(inpL,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "attn_norm", il);
+        // norm
+        cur = build_norm(inpL,
+                model.layers[il].attn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_norm", il);
 
-            // self-attention
-            {
-                ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
-                cb(Qcur, "Qcur", il);
+        // self-attention
+        {
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            cb(Qcur, "Qcur", il);
 
-                ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            cb(Kcur, "Kcur", il);
 
-                ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            cb(Vcur, "Vcur", il);
 
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
 
-                ggml_tensor * attn_rel_b = model.layers[il].attn_rel_b ? model.layers[il].attn_rel_b : model.layers[0].attn_rel_b;
-                ggml_tensor * kq_b = build_pos_bias(pos_bucket_dec, attn_rel_b);
+            ggml_tensor * attn_rel_b = model.layers[il].attn_rel_b ? model.layers[il].attn_rel_b : model.layers[0].attn_rel_b;
+            ggml_tensor * kq_b = build_pos_bias(pos_bucket_dec, attn_rel_b);
 
-                cur = build_attn(inp_attn_self,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Qcur, Kcur, Vcur, kq_b, nullptr, nullptr, 1.0f, il);
-                cb(cur, "kqv_out", il);
-            }
-            cur = ggml_add(ctx0, cur, inpSA);
-            cb(cur, "cross_inp", il);
-
-            ggml_tensor * inpCA = cur;
-
-            // norm
-            cur = build_norm(cur,
-                    model.layers[il].attn_norm_cross, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "attn_norm_cross", il);
-
-            // cross-attention
-            {
-                ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq_cross, cur);
-                cb(Qcur, "Qcur", il);
-
-                ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk_cross, embd_enc);
-                cb(Kcur, "Kcur", il);
-
-                ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv_cross, embd_enc);
-                cb(Vcur, "Vcur", il);
-
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_outputs_enc);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_outputs_enc);
-
-                cur = build_attn(inp_attn_cross,
-                        model.layers[il].wo_cross, nullptr,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f, il);
-                cb(cur, "kqv_out", il);
-
-                //ggml_tensor * q =                 ggml_permute(ctx0, Qcur, 0, 2, 1, 3);
-                //ggml_tensor * k = ggml_cont(ctx0, ggml_permute(ctx0, Kcur, 0, 2, 1, 3));
-
-                //ggml_tensor * kq = ggml_mul_mat(ctx0, k, q);
-                //cb(kq, "kq", il);
-
-                //kq = ggml_soft_max_ext(ctx0, kq, KQ_mask_cross, 1.0f, hparams.f_max_alibi_bias);
-                //cb(kq, "kq_soft_max_ext", il);
-
-                //ggml_tensor * v = ggml_cont(ctx0, ggml_transpose(ctx0, ggml_reshape_2d(ctx0, Vcur, n_embd_gqa, n_outputs_enc)));
-                //cb(v, "v", il);
-
-                //ggml_tensor * kqv = ggml_mul_mat(ctx0, ggml_reshape_3d(ctx0, v, n_outputs_enc, n_embd_head, n_head_kv), kq);
-                //cb(kqv, "kqv", il);
-
-                //ggml_tensor * kqv_merged = ggml_permute(ctx0, kqv, 0, 2, 1, 3);
-                //cb(kqv_merged, "kqv_merged", il);
-
-                //cur = ggml_cont_2d(ctx0, kqv_merged, n_embd_gqa, n_tokens);
-                //cb(cur, "kqv_merged_cont", il);
-
-                //ggml_build_forward_expand(gf, cur);
-
-                //cur = build_lora_mm(model.layers[il].wo_cross, cur);
-                //cb(cur, "kqv_out", il);
-            }
-            if (il == dec_n_layer - 1 && inp_out_ids) {
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpCA = ggml_get_rows(ctx0, inpCA, inp_out_ids);
-            }
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpCA);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // feed-forward network
-            {
-                cur = build_norm(ffn_inp,
-                        model.layers[il].ffn_norm, NULL,
-                        LLM_NORM_RMS, il);
-                cb(cur, "ffn_norm", il);
-
-                // T5 uses relu, flan-T5 uses gelu-gated
-                cur = build_ffn(cur,
-                        model.layers[il].ffn_up,   NULL, NULL,
-                        model.layers[il].ffn_gate, NULL, NULL,
-                        model.layers[il].ffn_down, NULL, NULL,
-                        NULL,
-                        model.layers[il].ffn_gate ? LLM_FFN_GELU : LLM_FFN_RELU,
-                        model.layers[il].ffn_gate ? LLM_FFN_PAR : LLM_FFN_SEQ,
-                        il);
-                cb(cur, "ffn_out", il);
-            }
-            cur = ggml_add(ctx0, cur, ffn_inp);
-            cb(cur, "ffn_out", il);
-
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
+            cur = build_attn(inp_attn_self,
+                    model.layers[il].wo, model.layers[il].bo,
+                    Qcur, Kcur, Vcur, kq_b, nullptr, nullptr, 1.0f, il);
+            cb(cur, "kqv_out", il);
         }
-        cur = inpL;
-        cb(cur, "result_embd", -1);
+        cur = ggml_add(ctx0, cur, inpSA);
+        cb(cur, "cross_inp", il);
 
+        ggml_tensor * inpCA = cur;
+
+        // norm
         cur = build_norm(cur,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, -1);
+                model.layers[il].attn_norm_cross, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_norm_cross", il);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+        // cross-attention
+        {
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq_cross, cur);
+            cb(Qcur, "Qcur", il);
 
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk_cross, embd_enc);
+            cb(Kcur, "Kcur", il);
 
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv_cross, embd_enc);
+            cb(Vcur, "Vcur", il);
 
-        ggml_build_forward_expand(gf, cur);
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_outputs_enc);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_outputs_enc);
+
+            cur = build_attn(inp_attn_cross,
+                    model.layers[il].wo_cross, nullptr,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f, il);
+            cb(cur, "kqv_out", il);
+
+            //ggml_tensor * q =                 ggml_permute(ctx0, Qcur, 0, 2, 1, 3);
+            //ggml_tensor * k = ggml_cont(ctx0, ggml_permute(ctx0, Kcur, 0, 2, 1, 3));
+
+            //ggml_tensor * kq = ggml_mul_mat(ctx0, k, q);
+            //cb(kq, "kq", il);
+
+            //kq = ggml_soft_max_ext(ctx0, kq, KQ_mask_cross, 1.0f, hparams.f_max_alibi_bias);
+            //cb(kq, "kq_soft_max_ext", il);
+
+            //ggml_tensor * v = ggml_cont(ctx0, ggml_transpose(ctx0, ggml_reshape_2d(ctx0, Vcur, n_embd_gqa, n_outputs_enc)));
+            //cb(v, "v", il);
+
+            //ggml_tensor * kqv = ggml_mul_mat(ctx0, ggml_reshape_3d(ctx0, v, n_outputs_enc, n_embd_head, n_head_kv), kq);
+            //cb(kqv, "kqv", il);
+
+            //ggml_tensor * kqv_merged = ggml_permute(ctx0, kqv, 0, 2, 1, 3);
+            //cb(kqv_merged, "kqv_merged", il);
+
+            //cur = ggml_cont_2d(ctx0, kqv_merged, n_embd_gqa, n_tokens);
+            //cb(cur, "kqv_merged_cont", il);
+
+            //ggml_build_forward_expand(gf, cur);
+
+            //cur = build_lora_mm(model.layers[il].wo_cross, cur);
+            //cb(cur, "kqv_out", il);
+        }
+        if (il == dec_n_layer - 1 && inp_out_ids) {
+            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+            inpCA = ggml_get_rows(ctx0, inpCA, inp_out_ids);
+        }
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpCA);
+        cb(ffn_inp, "ffn_inp", il);
+
+        // feed-forward network
+        {
+            cur = build_norm(ffn_inp,
+                    model.layers[il].ffn_norm, NULL,
+                    LLM_NORM_RMS, il);
+            cb(cur, "ffn_norm", il);
+
+            // T5 uses relu, flan-T5 uses gelu-gated
+            cur = build_ffn(cur,
+                    model.layers[il].ffn_up,   NULL, NULL,
+                    model.layers[il].ffn_gate, NULL, NULL,
+                    model.layers[il].ffn_down, NULL, NULL,
+                    NULL,
+                    model.layers[il].ffn_gate ? LLM_FFN_GELU : LLM_FFN_RELU,
+                    model.layers[il].ffn_gate ? LLM_FFN_PAR : LLM_FFN_SEQ,
+                    il);
+            cb(cur, "ffn_out", il);
+        }
+        cur = ggml_add(ctx0, cur, ffn_inp);
+        cb(cur, "ffn_out", il);
+
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
+
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
+    cb(cur, "result_embd", -1);
+
+    cur = build_norm(cur,
+            model.output_norm, NULL,
+            LLM_NORM_RMS, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/t5-enc.cpp

@@ -1,96 +1,96 @@
 #include "models.h"
 
 llm_build_t5_enc::llm_build_t5_enc(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_head = hparams.n_embd_head_v;
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        ggml_tensor * pos_bucket_enc = build_inp_pos_bucket_enc();
+    ggml_tensor * pos_bucket_enc = build_inp_pos_bucket_enc();
 
-        auto * inp_attn = build_attn_inp_no_cache();
+    auto * inp_attn = build_attn_inp_no_cache();
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            ggml_tensor * inpSA = inpL;
+    for (int il = 0; il < n_layer; ++il) {
+        ggml_tensor * inpSA = inpL;
 
-            // norm
-            cur = build_norm(inpL,
-                    model.layers[il].attn_norm_enc, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "attn_norm", il);
+        // norm
+        cur = build_norm(inpL,
+                model.layers[il].attn_norm_enc, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_norm", il);
 
-            // self-attention
-            {
-                ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq_enc, cur);
-                cb(Qcur, "Qcur", il);
+        // self-attention
+        {
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq_enc, cur);
+            cb(Qcur, "Qcur", il);
 
-                ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk_enc, cur);
-                cb(Kcur, "Kcur", il);
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk_enc, cur);
+            cb(Kcur, "Kcur", il);
 
-                ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv_enc, cur);
-                cb(Vcur, "Vcur", il);
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv_enc, cur);
+            cb(Vcur, "Vcur", il);
 
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
 
-                ggml_tensor * attn_rel_b = model.layers[il].attn_rel_b_enc ? model.layers[il].attn_rel_b_enc : model.layers[0].attn_rel_b_enc;
-                ggml_tensor * kq_b = build_pos_bias(pos_bucket_enc, attn_rel_b);
+            ggml_tensor * attn_rel_b = model.layers[il].attn_rel_b_enc ? model.layers[il].attn_rel_b_enc : model.layers[0].attn_rel_b_enc;
+            ggml_tensor * kq_b = build_pos_bias(pos_bucket_enc, attn_rel_b);
 
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo_enc, nullptr,
-                        Qcur, Kcur, Vcur, kq_b, nullptr, nullptr, 1.0f, il);
-                cb(cur, "kqv_out", il);
-            }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // feed-forward network
-            {
-                cur = build_norm(ffn_inp,
-                        model.layers[il].ffn_norm_enc, NULL,
-                        LLM_NORM_RMS, il);
-                cb(cur, "ffn_norm", il);
-
-                // T5 uses relu, flan-T5 uses gelu-gated
-                cur = build_ffn(cur,
-                        model.layers[il].ffn_up_enc,   NULL, NULL,
-                        model.layers[il].ffn_gate_enc, NULL, NULL,
-                        model.layers[il].ffn_down_enc, NULL, NULL,
-                        NULL,
-                        model.layers[il].ffn_gate_enc ? LLM_FFN_GELU : LLM_FFN_RELU,
-                        model.layers[il].ffn_gate_enc ? LLM_FFN_PAR  : LLM_FFN_SEQ,
-                        il);
-                cb(cur, "ffn_out", il);
-            }
-            cur = ggml_add(ctx0, cur, ffn_inp);
-            cb(cur, "ffn_out", il);
-
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo_enc, nullptr,
+                    Qcur, Kcur, Vcur, kq_b, nullptr, nullptr, 1.0f, il);
+            cb(cur, "kqv_out", il);
         }
-        cur = inpL;
-        cb(cur, "result_embd", -1);
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+        }
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+        cb(ffn_inp, "ffn_inp", il);
 
-        cur = build_norm(cur,
-                model.output_norm_enc, NULL,
-                LLM_NORM_RMS, -1);
+        // feed-forward network
+        {
+            cur = build_norm(ffn_inp,
+                    model.layers[il].ffn_norm_enc, NULL,
+                    LLM_NORM_RMS, il);
+            cb(cur, "ffn_norm", il);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+            // T5 uses relu, flan-T5 uses gelu-gated
+            cur = build_ffn(cur,
+                    model.layers[il].ffn_up_enc,   NULL, NULL,
+                    model.layers[il].ffn_gate_enc, NULL, NULL,
+                    model.layers[il].ffn_down_enc, NULL, NULL,
+                    NULL,
+                    model.layers[il].ffn_gate_enc ? LLM_FFN_GELU : LLM_FFN_RELU,
+                    model.layers[il].ffn_gate_enc ? LLM_FFN_PAR  : LLM_FFN_SEQ,
+                    il);
+            cb(cur, "ffn_out", il);
+        }
+        cur = ggml_add(ctx0, cur, ffn_inp);
+        cb(cur, "ffn_out", il);
 
-        ggml_build_forward_expand(gf, cur);
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
+
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
+    cb(cur, "result_embd", -1);
+
+    cur = build_norm(cur,
+            model.output_norm_enc, NULL,
+            LLM_NORM_RMS, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/wavtokenizer-dec.cpp

@@ -1,149 +1,149 @@
 #include "models.h"
 
 llm_build_wavtokenizer_dec::llm_build_wavtokenizer_dec(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        cur = ggml_cont(ctx0, ggml_transpose(ctx0, inpL));
+    cur = ggml_cont(ctx0, ggml_transpose(ctx0, inpL));
 
-        cur = ggml_conv_1d_ph(ctx0, model.conv1d, cur, 1, 1);
-        cur = ggml_add(ctx0, cur, model.conv1d_b);
+    cur = ggml_conv_1d_ph(ctx0, model.conv1d, cur, 1, 1);
+    cur = ggml_add(ctx0, cur, model.conv1d_b);
 
-        // posnet
-        for (uint32_t il = 0; il < hparams.posnet.n_layer; ++il) {
-            const auto & layer = model.layers[il].posnet;
-
-            inpL = cur;
-
-            switch (il) {
-                case 0:
-                case 1:
-                case 3:
-                case 4:
-                    {
-                        cur = build_norm(cur,
-                                layer.norm1,
-                                layer.norm1_b,
-                                LLM_NORM_GROUP, 0);
-
-                        cur = ggml_mul(ctx0, ggml_sigmoid(ctx0, cur), cur);
-
-                        cur = ggml_conv_1d_ph(ctx0, layer.conv1, cur, 1, 1);
-                        cur = ggml_add(ctx0, cur, layer.conv1_b);
-
-                        cur = build_norm(cur,
-                                layer.norm2,
-                                layer.norm2_b,
-                                LLM_NORM_GROUP, 0);
-
-                        cur = ggml_mul(ctx0, ggml_sigmoid(ctx0, cur), cur);
-
-                        cur = ggml_conv_1d_ph(ctx0, layer.conv2, cur, 1, 1);
-                        cur = ggml_add(ctx0, cur, layer.conv2_b);
-
-                        cur = ggml_add(ctx0, cur, inpL);
-                    } break;
-                case 2:
-                    {
-                        cur = build_norm(cur,
-                                layer.attn_norm,
-                                layer.attn_norm_b,
-                                LLM_NORM_GROUP, 0);
-
-                        ggml_tensor * q;
-                        ggml_tensor * k;
-                        ggml_tensor * v;
-
-                        q = ggml_conv_1d_ph(ctx0, layer.attn_q, cur, 1, 1);
-                        k = ggml_conv_1d_ph(ctx0, layer.attn_k, cur, 1, 1);
-                        v = ggml_conv_1d_ph(ctx0, layer.attn_v, cur, 1, 1);
-
-                        q = ggml_add(ctx0, q, layer.attn_q_b);
-                        k = ggml_add(ctx0, k, layer.attn_k_b);
-                        v = ggml_add(ctx0, v, layer.attn_v_b);
-
-                        q = ggml_cont(ctx0, ggml_transpose(ctx0, q));
-                        k = ggml_cont(ctx0, ggml_transpose(ctx0, k));
-
-                        ggml_tensor * kq = ggml_mul_mat(ctx0, k, q);
-
-                        kq = ggml_soft_max_ext(ctx0, kq, nullptr, 1.0f/sqrtf(float(hparams.posnet.n_embd)), 0.0f);
-
-                        cur = ggml_mul_mat(ctx0, kq, v);
-
-                        cur = ggml_conv_1d_ph(ctx0, layer.attn_o, cur, 1, 1);
-                        cur = ggml_add(ctx0, cur, layer.attn_o_b);
-
-                        cur = ggml_add(ctx0, cur, inpL);
-                    } break;
-                case 5:
-                    {
-                        cur = build_norm(cur,
-                                layer.norm,
-                                layer.norm_b,
-                                LLM_NORM_GROUP, 0);
-                    } break;
-                default: GGML_ABORT("unknown posnet layer");
-            };
-        }
-        cur = ggml_cont(ctx0, ggml_transpose(ctx0, cur));
-
-        cur = build_norm(cur,
-                model.tok_norm,
-                model.tok_norm_b,
-                LLM_NORM, -1);
-
-        cur = ggml_cont(ctx0, ggml_transpose(ctx0, cur));
+    // posnet
+    for (uint32_t il = 0; il < hparams.posnet.n_layer; ++il) {
+        const auto & layer = model.layers[il].posnet;
 
         inpL = cur;
 
-        // convnext
-        for (uint32_t il = 0; il < hparams.convnext.n_layer; ++il) {
-            const auto & layer = model.layers[il].convnext;
+        switch (il) {
+            case 0:
+            case 1:
+            case 3:
+            case 4:
+                {
+                    cur = build_norm(cur,
+                            layer.norm1,
+                            layer.norm1_b,
+                            LLM_NORM_GROUP, 0);
 
-            cur = inpL;
+                    cur = ggml_mul(ctx0, ggml_sigmoid(ctx0, cur), cur);
 
-            cur = ggml_conv_1d_dw_ph(ctx0, layer.dw, cur, 1, 1);
-            cur = ggml_add(ctx0, cur, layer.dw_b);
+                    cur = ggml_conv_1d_ph(ctx0, layer.conv1, cur, 1, 1);
+                    cur = ggml_add(ctx0, cur, layer.conv1_b);
 
-            cur = ggml_cont(ctx0, ggml_transpose(ctx0, cur));
+                    cur = build_norm(cur,
+                            layer.norm2,
+                            layer.norm2_b,
+                            LLM_NORM_GROUP, 0);
 
-            cur = build_norm(cur,
-                    layer.norm,
-                    layer.norm_b,
-                    LLM_NORM, -1);
+                    cur = ggml_mul(ctx0, ggml_sigmoid(ctx0, cur), cur);
 
-            cur = build_ffn(cur,
-                    layer.pw1, layer.pw1_b, NULL,
-                    NULL,      NULL,        NULL,
-                    layer.pw2, layer.pw2_b, NULL,
-                    NULL,
-                    LLM_FFN_GELU, LLM_FFN_SEQ, il);
+                    cur = ggml_conv_1d_ph(ctx0, layer.conv2, cur, 1, 1);
+                    cur = ggml_add(ctx0, cur, layer.conv2_b);
 
-            cur = ggml_mul(ctx0, cur, layer.gamma);
+                    cur = ggml_add(ctx0, cur, inpL);
+                } break;
+            case 2:
+                {
+                    cur = build_norm(cur,
+                            layer.attn_norm,
+                            layer.attn_norm_b,
+                            LLM_NORM_GROUP, 0);
 
-            cur = ggml_cont(ctx0, ggml_transpose(ctx0, cur));
+                    ggml_tensor * q;
+                    ggml_tensor * k;
+                    ggml_tensor * v;
+
+                    q = ggml_conv_1d_ph(ctx0, layer.attn_q, cur, 1, 1);
+                    k = ggml_conv_1d_ph(ctx0, layer.attn_k, cur, 1, 1);
+                    v = ggml_conv_1d_ph(ctx0, layer.attn_v, cur, 1, 1);
+
+                    q = ggml_add(ctx0, q, layer.attn_q_b);
+                    k = ggml_add(ctx0, k, layer.attn_k_b);
+                    v = ggml_add(ctx0, v, layer.attn_v_b);
+
+                    q = ggml_cont(ctx0, ggml_transpose(ctx0, q));
+                    k = ggml_cont(ctx0, ggml_transpose(ctx0, k));
+
+                    ggml_tensor * kq = ggml_mul_mat(ctx0, k, q);
+
+                    kq = ggml_soft_max_ext(ctx0, kq, nullptr, 1.0f/sqrtf(float(hparams.posnet.n_embd)), 0.0f);
+
+                    cur = ggml_mul_mat(ctx0, kq, v);
+
+                    cur = ggml_conv_1d_ph(ctx0, layer.attn_o, cur, 1, 1);
+                    cur = ggml_add(ctx0, cur, layer.attn_o_b);
+
+                    cur = ggml_add(ctx0, cur, inpL);
+                } break;
+            case 5:
+                {
+                    cur = build_norm(cur,
+                            layer.norm,
+                            layer.norm_b,
+                            LLM_NORM_GROUP, 0);
+                } break;
+            default: GGML_ABORT("unknown posnet layer");
+        };
+    }
+    cur = ggml_cont(ctx0, ggml_transpose(ctx0, cur));
+
+    cur = build_norm(cur,
+            model.tok_norm,
+            model.tok_norm_b,
+            LLM_NORM, -1);
+
+    cur = ggml_cont(ctx0, ggml_transpose(ctx0, cur));
+
+    inpL = cur;
+
+    // convnext
+    for (uint32_t il = 0; il < hparams.convnext.n_layer; ++il) {
+        const auto & layer = model.layers[il].convnext;
 
-            inpL = ggml_add(ctx0, cur, inpL);
-        }
         cur = inpL;
 
+        cur = ggml_conv_1d_dw_ph(ctx0, layer.dw, cur, 1, 1);
+        cur = ggml_add(ctx0, cur, layer.dw_b);
+
         cur = ggml_cont(ctx0, ggml_transpose(ctx0, cur));
 
         cur = build_norm(cur,
-                model.output_norm,
-                model.output_norm_b,
+                layer.norm,
+                layer.norm_b,
                 LLM_NORM, -1);
 
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
+        cur = build_ffn(cur,
+                layer.pw1, layer.pw1_b, NULL,
+                NULL,      NULL,        NULL,
+                layer.pw2, layer.pw2_b, NULL,
+                NULL,
+                LLM_FFN_GELU, LLM_FFN_SEQ, il);
 
-        cur = ggml_add(ctx0, cur, model.output_b);
+        cur = ggml_mul(ctx0, cur, layer.gamma);
 
-        cb(cur, "result_embd", -1);
-        res->t_embd = cur;
+        cur = ggml_cont(ctx0, ggml_transpose(ctx0, cur));
 
-        ggml_build_forward_expand(gf, cur);
+        inpL = ggml_add(ctx0, cur, inpL);
+    }
+    cur = inpL;
+
+    cur = ggml_cont(ctx0, ggml_transpose(ctx0, cur));
+
+    cur = build_norm(cur,
+            model.output_norm,
+            model.output_norm_b,
+            LLM_NORM, -1);
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+
+    cur = ggml_add(ctx0, cur, model.output_b);
+
+    cb(cur, "result_embd", -1);
+    res->t_embd = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }

src/models/xverse.cpp

@@ -1,108 +1,108 @@
 #include "models.h"
 
 llm_build_xverse::llm_build_xverse(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
-        const int64_t n_embd_head = hparams.n_embd_head_v;
+    const int64_t n_embd_head = hparams.n_embd_head_v;
 
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
+    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-        ggml_tensor * cur;
-        ggml_tensor * inpL;
+    ggml_tensor * cur;
+    ggml_tensor * inpL;
 
-        inpL = build_inp_embd(model.tok_embd);
+    inpL = build_inp_embd(model.tok_embd);
 
-        // inp_pos - contains the positions
-        ggml_tensor * inp_pos = build_inp_pos();
+    // inp_pos - contains the positions
+    ggml_tensor * inp_pos = build_inp_pos();
 
-        auto * inp_attn = build_attn_inp_kv();
+    auto * inp_attn = build_attn_inp_kv();
 
-        ggml_tensor * inp_out_ids = build_inp_out_ids();
+    ggml_tensor * inp_out_ids = build_inp_out_ids();
 
-        for (int il = 0; il < n_layer; ++il) {
-            ggml_tensor * inpSA = inpL;
+    for (int il = 0; il < n_layer; ++il) {
+        ggml_tensor * inpSA = inpL;
 
-            cur = build_norm(inpL,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, il);
-            cb(cur, "attn_norm", il);
+        cur = build_norm(inpL,
+                model.layers[il].attn_norm, NULL,
+                LLM_NORM_RMS, il);
+        cb(cur, "attn_norm", il);
 
-            // self-attention
-            {
-                ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
-                cb(Qcur, "Qcur", il);
+        // self-attention
+        {
+            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
+            cb(Qcur, "Qcur", il);
 
-                ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
+            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
+            cb(Kcur, "Kcur", il);
 
-                ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
+            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);
+            cb(Vcur, "Vcur", il);
 
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
+            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
 
-                Qcur = ggml_rope_ext(
-                        ctx0, Qcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
+            Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-                Kcur = ggml_rope_ext(
-                        ctx0, Kcur, inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow
-                        );
+            Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                    );
 
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
+            cb(Qcur, "Qcur", il);
+            cb(Kcur, "Kcur", il);
+            cb(Vcur, "Vcur", il);
 
-                cur = build_attn(inp_attn,
-                        model.layers[il].wo, NULL,
-                        Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
-            }
-            if (il == n_layer - 1 && inp_out_ids) {
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
-            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // feed-forward network
-            {
-                cur = build_norm(ffn_inp,
-                        model.layers[il].ffn_norm, NULL,
-                        LLM_NORM_RMS, il);
-                cb(cur, "ffn_norm", il);
-
-                cur = build_ffn(cur,
-                        model.layers[il].ffn_up,   NULL, NULL,
-                        model.layers[il].ffn_gate, NULL, NULL,
-                        model.layers[il].ffn_down, NULL, NULL,
-                        NULL,
-                        LLM_FFN_SILU, LLM_FFN_PAR, il);
-                cb(cur, "ffn_out", il);
-            }
-            cur = ggml_add(ctx0, cur, ffn_inp);
-
-            cur = build_cvec(cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
+            cur = build_attn(inp_attn,
+                    model.layers[il].wo, NULL,
+                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
         }
-        cur = inpL;
+        if (il == n_layer - 1 && inp_out_ids) {
+            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+        }
+        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+        cb(ffn_inp, "ffn_inp", il);
 
-        cur = build_norm(cur, model.output_norm, NULL, LLM_NORM_RMS, -1);
+        // feed-forward network
+        {
+            cur = build_norm(ffn_inp,
+                    model.layers[il].ffn_norm, NULL,
+                    LLM_NORM_RMS, il);
+            cb(cur, "ffn_norm", il);
 
-        cb(cur, "result_norm", -1);
-        res->t_embd = cur;
+            cur = build_ffn(cur,
+                    model.layers[il].ffn_up,   NULL, NULL,
+                    model.layers[il].ffn_gate, NULL, NULL,
+                    model.layers[il].ffn_down, NULL, NULL,
+                    NULL,
+                    LLM_FFN_SILU, LLM_FFN_PAR, il);
+            cb(cur, "ffn_out", il);
+        }
+        cur = ggml_add(ctx0, cur, ffn_inp);
 
-        // lm_head
-        cur = build_lora_mm(model.output, cur);
+        cur = build_cvec(cur, il);
+        cb(cur, "l_out", il);
 
-        cb(cur, "result_output", -1);
-        res->t_logits = cur;
+        // input for next layer
+        inpL = cur;
+    }
+    cur = inpL;
 
-        ggml_build_forward_expand(gf, cur);
+    cur = build_norm(cur, model.output_norm, NULL, LLM_NORM_RMS, -1);
+
+    cb(cur, "result_norm", -1);
+    res->t_embd = cur;
+
+    // lm_head
+    cur = build_lora_mm(model.output, cur);
+
+    cb(cur, "result_output", -1);
+    res->t_logits = cur;
+
+    ggml_build_forward_expand(gf, cur);
 }