llama : add IBM Granite MoE architecture (#9438 )

* feat(gguf-py): Add granitemoe architecture This includes the addition of new tensor names for the new moe layers. These may not be correct at this point due to the need for the hack in gguf_writer.py to double-check the length of the shape for these layers. Branch: GraniteMoE Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * feat(convert_hf_to_gguf): Add GraniteMoeModel GraniteMoe has the same configuration deltas as Granite Branch: GraniteMoE Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix(granitemoe convert): Split the double-sized input layer into gate and up After a lot of staring and squinting, it's clear that the standard mixtral expert implementation is equivalent to the vectorized parallel experts in granite. The difference is that in granite, the w1 and w3 are concatenated into a single tensor "input_linear." Rather than reimplementing all of the math on the llama.cpp side, the much simpler route is to just split this tensor during conversion and follow the standard mixtral route. Branch: GraniteMoE Co-Authored-By: alex.brooks@ibm.com Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * feat(granitemoe): Implement granitemoe GraniteMoE follows the mixtral architecture (once the input_linear layers are split into gate_exps/up_exps). The main delta is the addition of the same four multipliers used in Granite. Branch: GraniteMoE Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * Typo fix in docstring Co-Authored-By: ggerganov@gmail.com Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix(conversion): Simplify tensor name mapping in conversion Branch: GraniteMoE Co-Authored-By: git@compilade.net Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix(convert): Remove unused tensor name mappings Branch: GraniteMoE Co-Authored-By: git@compilade.net Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix(convert): Sanity check on merged FFN tensor sizes Branch: GraniteMoE Co-Authored-By: git@compilade.net Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix: Allow "output" layer in granite moe architecture (convert and cpp) Branch: GraniteMoE Co-Authored-By: git@compilade.net Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix(granite): Add missing 'output' tensor for Granite This is a fix for the previous `granite` architecture PR. Recent snapshots have included this (`lm_head.weights`) as part of the architecture Branch: GraniteMoE Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> --------- Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
cann: fix crash when llama-bench is running on multiple cann devices (#9627 )
2024-09-25 10:06:52 +03:00 · 2024-09-25 11:30:38 +08:00 · 2024-09-24 11:03:21 +03:00 · 2024-09-24 11:01:18 +03:00 · 2024-09-24 11:00:52 +03:00 · 2024-09-24 10:16:06 +03:00
32 changed files with 1020 additions and 146 deletions
@@ -112,6 +112,7 @@ Typically finetunes of the base models below are supported as well.
 - Go: [go-skynet/go-llama.cpp](https://github.com/go-skynet/go-llama.cpp)
 - Node.js: [withcatai/node-llama-cpp](https://github.com/withcatai/node-llama-cpp)
 - JS/TS (llama.cpp server client): [lgrammel/modelfusion](https://modelfusion.dev/integration/model-provider/llamacpp)
+- JS/TS (Programmable Prompt Engine CLI): [offline-ai/cli](https://github.com/offline-ai/cli)
 - JavaScript/Wasm (works in browser): [tangledgroup/llama-cpp-wasm](https://github.com/tangledgroup/llama-cpp-wasm)
 - Typescript/Wasm (nicer API, available on npm): [ngxson/wllama](https://github.com/ngxson/wllama)
 - Ruby: [yoshoku/llama_cpp.rb](https://github.com/yoshoku/llama_cpp.rb)
@@ -691,7 +691,7 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
        [](gpt_params & params) {
            params.ctx_shift = false;
        }
-    ).set_examples({LLAMA_EXAMPLE_MAIN}));
+    ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_SERVER}));
    add_opt(llama_arg(
        {"--chunks"}, "N",
        format("max number of chunks to process (default: %d, -1 = all)", params.n_chunks),
@@ -82,7 +82,7 @@ struct gpt_log_entry {
            }
        }

-        if (level != GGML_LOG_LEVEL_NONE && prefix) {
+        if (level != GGML_LOG_LEVEL_NONE && level != GGML_LOG_LEVEL_CONT && prefix) {
            if (timestamp) {
                // [M.s.ms.us]
                fprintf(fcur, "%s%d.%02d.%03d.%03d%s ",
@@ -83,8 +83,10 @@ void gpt_log_set_timestamps(struct gpt_log * log,       bool   timestamps); // w
 #define LOG_WRN(...) LOG_TMPL(GGML_LOG_LEVEL_WARN,  0,                 __VA_ARGS__)
 #define LOG_ERR(...) LOG_TMPL(GGML_LOG_LEVEL_ERROR, 0,                 __VA_ARGS__)
 #define LOG_DBG(...) LOG_TMPL(GGML_LOG_LEVEL_DEBUG, LOG_DEFAULT_DEBUG, __VA_ARGS__)
+#define LOG_CNT(...) LOG_TMPL(GGML_LOG_LEVEL_CONT,  0,                 __VA_ARGS__)

 #define LOG_INFV(verbosity, ...) LOG_TMPL(GGML_LOG_LEVEL_INFO,  verbosity, __VA_ARGS__)
 #define LOG_WRNV(verbosity, ...) LOG_TMPL(GGML_LOG_LEVEL_WARN,  verbosity, __VA_ARGS__)
 #define LOG_ERRV(verbosity, ...) LOG_TMPL(GGML_LOG_LEVEL_ERROR, verbosity, __VA_ARGS__)
 #define LOG_DBGV(verbosity, ...) LOG_TMPL(GGML_LOG_LEVEL_DEBUG, verbosity, __VA_ARGS__)
+#define LOG_CNTV(verbosity, ...) LOG_TMPL(GGML_LOG_LEVEL_CONT,  verbosity, __VA_ARGS__)
@@ -209,7 +209,15 @@ struct gpt_sampler * gpt_sampler_init(const struct llama_model * model, const st
            GGML_ASSERT(false && "unknown mirostat version");
        }
    } else {
-        llama_sampler_chain_add(result->chain, llama_sampler_init_softmax());
+        if (params.n_probs > 0) {
+            // some use cases require to sample greedily, but still obtain the probabilities of the top tokens
+            // ref: https://github.com/ggerganov/llama.cpp/pull/9605
+            //
+            // the following will not produce exactly the same probs as applyging softmax to the full vocabulary, but
+            // it is much faster, since we avoid sorting all tokens and should give a good approximation
+            llama_sampler_chain_add(result->chain, llama_sampler_init_top_k(params.n_probs));
+            llama_sampler_chain_add(result->chain, llama_sampler_init_softmax());
+        }
        llama_sampler_chain_add(result->chain, llama_sampler_init_greedy());
    }

@@ -4102,16 +4102,45 @@ class GraniteModel(LlamaModel):
        #   consistency
        if attention_scale := self.hparams.get("attention_multiplier"):
            self.gguf_writer.add_attention_scale(attention_scale)
+            logger.info("gguf: (granite) attention_scale = %s", attention_scale)
        if embedding_scale := self.hparams.get("embedding_multiplier"):
            self.gguf_writer.add_embedding_scale(embedding_scale)
+            logger.info("gguf: (granite) embedding_scale = %s", embedding_scale)
        if residual_scale := self.hparams.get("residual_multiplier"):
            self.gguf_writer.add_residual_scale(residual_scale)
-        if logits_scaling := self.hparams.get("logits_scaling"):
-            self.gguf_writer.add_logit_scale(logits_scaling)
+            logger.info("gguf: (granite) residual_scale = %s", residual_scale)
+        if logits_scale := self.hparams.get("logits_scaling"):
+            self.gguf_writer.add_logit_scale(logits_scale)
+            logger.info("gguf: (granite) logits_scale = %s", logits_scale)
+
+
+@Model.register("GraniteMoeForCausalLM")
+class GraniteMoeModel(GraniteModel):
+    """Conversion for IBM's GraniteMoeForCausalLM"""
+    model_arch = gguf.MODEL_ARCH.GRANITE_MOE
+
+    def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
+        """In modeling_granitemoe, the JetMoe implementation of parallel experts
+        is used. This essentially merges w1 and w3 into a single tensor with 2x
+        the hidden size that is then split during forward. To keep compatibility
+        with existing mixtral support, we pull them apart here.
+        """
+
+        if name.endswith("block_sparse_moe.input_linear.weight"):
+            ffn_dim = self.hparams["intermediate_size"]
+            assert data_torch.shape[-2] == 2 * ffn_dim, "Merged FFN tensor size must be 2 * intermediate_size"
+            gate, up = data_torch[..., :ffn_dim, :], data_torch[..., ffn_dim:, :]
+            return [
+                (self.format_tensor_name(gguf.MODEL_TENSOR.FFN_GATE_EXP, bid), gate),
+                (self.format_tensor_name(gguf.MODEL_TENSOR.FFN_UP_EXP, bid), up),
+            ]
+
+        return super().modify_tensors(data_torch, name, bid)


 ###### CONVERSION LOGIC ######

+
 # tree of lazy tensors
 class LazyTorchTensor(gguf.LazyBase):
    _tensor_type = torch.Tensor
@@ -263,9 +263,9 @@ int main(int argc, char ** argv) {
        if (params.n_keep > 0) {
        LOG_INF("%s: static prompt based on n_keep: '", __func__);
            for (int i = 0; i < params.n_keep; i++) {
-                LOG("%s", llama_token_to_piece(ctx, embd_inp[i]).c_str());
+                LOG_CNT("%s", llama_token_to_piece(ctx, embd_inp[i]).c_str());
            }
-            LOG("'\n");
+            LOG_CNT("'\n");
        }
        LOG_INF("\n");
    }
@@ -306,8 +306,8 @@ int main(int argc, char ** argv) {

    LOG_INF("generate: n_ctx = %d, n_batch = %d, n_predict = %d, n_keep = %d\n", n_ctx, params.n_batch, params.n_predict, params.n_keep);

-    LOG("\n");
-    LOG("\n#####  Infill mode  #####\n\n");
+    LOG_INF("\n");
+    LOG_INF("\n#####  Infill mode  #####\n\n");
    if (params.interactive) {
        const char *control_message;
        if (params.multiline_input) {
@@ -318,11 +318,11 @@ int main(int argc, char ** argv) {
                              " - To return control without starting a new line, end your input with '/'.\n"
                              " - If you want to submit another line, end your input with '\\'.\n";
        }
-        LOG("== Running in interactive mode. ==\n");
+        LOG_INF("== Running in interactive mode. ==\n");
 #if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__)) || defined (_WIN32)
-        LOG(       " - Press Ctrl+C to interject at any time.\n");
+        LOG_INF(       " - Press Ctrl+C to interject at any time.\n");
 #endif
-        LOG(       "%s\n", control_message);
+        LOG_INF(       "%s\n", control_message);

        is_interacting = params.interactive_first;
    }
@@ -385,9 +385,9 @@ int main(int argc, char ** argv) {
        if (params.n_keep > add_bos) {
            LOG_INF("%s: static prompt based on n_keep: '", __func__);
            for (int i = 0; i < params.n_keep; i++) {
-                LOG("%s", llama_token_to_piece(ctx, embd_inp[i]).c_str());
+                LOG_CNT("%s", llama_token_to_piece(ctx, embd_inp[i]).c_str());
            }
-            LOG("'\n");
+            LOG_CNT("'\n");
        }
        LOG_INF("\n");
    }
@@ -409,40 +409,40 @@ int main(int argc, char ** argv) {
    }

    if (params.interactive) {
-        LOG("%s: interactive mode on.\n", __func__);
+        LOG_INF("%s: interactive mode on.\n", __func__);

        if (!params.antiprompt.empty()) {
            for (const auto & antiprompt : params.antiprompt) {
-                LOG("Reverse prompt: '%s'\n", antiprompt.c_str());
+                LOG_INF("Reverse prompt: '%s'\n", antiprompt.c_str());
                if (params.verbose_prompt) {
                    auto tmp = ::llama_tokenize(ctx, antiprompt, false, true);
                    for (int i = 0; i < (int) tmp.size(); i++) {
-                        LOG("%6d -> '%s'\n", tmp[i], llama_token_to_piece(ctx, tmp[i]).c_str());
+                        LOG_INF("%6d -> '%s'\n", tmp[i], llama_token_to_piece(ctx, tmp[i]).c_str());
                    }
                }
            }
        }

        if (params.input_prefix_bos) {
-            LOG("Input prefix with BOS\n");
+            LOG_INF("Input prefix with BOS\n");
        }

        if (!params.input_prefix.empty()) {
-            LOG("Input prefix: '%s'\n", params.input_prefix.c_str());
+            LOG_INF("Input prefix: '%s'\n", params.input_prefix.c_str());
            if (params.verbose_prompt) {
                auto tmp = ::llama_tokenize(ctx, params.input_prefix, true, true);
                for (int i = 0; i < (int) tmp.size(); i++) {
-                    LOG("%6d -> '%s'\n", tmp[i], llama_token_to_piece(ctx, tmp[i]).c_str());
+                    LOG_INF("%6d -> '%s'\n", tmp[i], llama_token_to_piece(ctx, tmp[i]).c_str());
                }
            }
        }

        if (!params.input_suffix.empty()) {
-            LOG("Input suffix: '%s'\n", params.input_suffix.c_str());
+            LOG_INF("Input suffix: '%s'\n", params.input_suffix.c_str());
            if (params.verbose_prompt) {
                auto tmp = ::llama_tokenize(ctx, params.input_suffix, false, true);
                for (int i = 0; i < (int) tmp.size(); i++) {
-                    LOG("%6d -> '%s'\n", tmp[i], llama_token_to_piece(ctx, tmp[i]).c_str());
+                    LOG_INF("%6d -> '%s'\n", tmp[i], llama_token_to_piece(ctx, tmp[i]).c_str());
                }
            }
        }
@@ -474,7 +474,7 @@ int main(int argc, char ** argv) {
      //GGML_ASSERT(n_ctx >= n_ctx_train * ga_n && "n_ctx must be at least n_ctx_train * grp_attn_n"); // NOLINT
        LOG_INF("self-extend: n_ctx_train = %d, grp_attn_n = %d, grp_attn_w = %d\n", n_ctx_train, ga_n, ga_w);
    }
-    LOG("\n");
+    LOG_INF("\n");

    if (params.interactive) {
        const char * control_message;
@@ -486,11 +486,11 @@ int main(int argc, char ** argv) {
                              " - To return control without starting a new line, end your input with '/'.\n"
                              " - If you want to submit another line, end your input with '\\'.\n";
        }
-        LOG("== Running in interactive mode. ==\n");
+        LOG_INF("== Running in interactive mode. ==\n");
 #if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__)) || defined (_WIN32)
-        LOG(       " - Press Ctrl+C to interject at any time.\n");
+        LOG_INF(       " - Press Ctrl+C to interject at any time.\n");
 #endif
-        LOG(       "%s\n", control_message);
+        LOG_INF(       "%s\n", control_message);

        is_interacting = params.interactive_first;
    }
@@ -21,8 +21,6 @@ The project is under active development, and we are [looking for feedback and co
 | -------- | ----------- |
 | `-h, --help, --usage` | print usage and exit |
 | `--version` | show version and build info |
-| `-v, --verbose` | print verbose information |
-| `--verbosity N` | set specific verbosity level (default: 0) |
 | `-t, --threads N` | number of threads to use during generation (default: -1)<br/>(env: LLAMA_ARG_THREADS) |
 | `-tb, --threads-batch N` | number of threads to use during batch and prompt processing (default: same as --threads) |
 | `-C, --cpu-mask M` | CPU affinity mask: arbitrarily long hex. Complements cpu-range (default: "") |
@@ -40,15 +38,18 @@ The project is under active development, and we are [looking for feedback and co
 | `-b, --batch-size N` | logical maximum batch size (default: 2048)<br/>(env: LLAMA_ARG_BATCH) |
 | `-ub, --ubatch-size N` | physical maximum batch size (default: 512)<br/>(env: LLAMA_ARG_UBATCH) |
 | `--keep N` | number of tokens to keep from the initial prompt (default: 0, -1 = all) |
+| `--no-context-shift` | disables context shift on inifinite text generation (default: disabled) |
 | `-fa, --flash-attn` | enable Flash Attention (default: disabled)<br/>(env: LLAMA_ARG_FLASH_ATTN) |
 | `-p, --prompt PROMPT` | prompt to start generation with |
+| `--no-perf` | disable internal libllama performance timings (default: false)<br/>(env: LLAMA_ARG_NO_PERF) |
 | `-f, --file FNAME` | a file containing the prompt (default: none) |
 | `-bf, --binary-file FNAME` | binary file containing the prompt (default: none) |
 | `-e, --escape` | process escapes sequences (\n, \r, \t, \', \", \\) (default: true) |
 | `--no-escape` | do not process escape sequences |
+| `-sp, --special` | special tokens output enabled (default: false) |
 | `--spm-infill` | use Suffix/Prefix/Middle pattern for infill (instead of Prefix/Suffix/Middle) as some models prefer this. (default: disabled) |
 | `--samplers SAMPLERS` | samplers that will be used for generation in the order, separated by ';'<br/>(default: top_k;tfs_z;typ_p;top_p;min_p;temperature) |
-| `-s, --seed SEED` | RNG seed (default: -1, use random seed for < 0) |
+| `-s, --seed SEED` | RNG seed (default: 4294967295, use random seed for 4294967295) |
 | `--sampling-seq SEQUENCE` | simplified sequence for samplers that will be used (default: kfypmt) |
 | `--ignore-eos` | ignore end of stream token and continue generating (implies --logit-bias EOS-inf) |
 | `--penalize-nl` | penalize newline tokens (default: false) |
@@ -87,7 +88,7 @@ The project is under active development, and we are [looking for feedback and co
 | `-ctk, --cache-type-k TYPE` | KV cache data type for K (default: f16) |
 | `-ctv, --cache-type-v TYPE` | KV cache data type for V (default: f16) |
 | `-dt, --defrag-thold N` | KV cache defragmentation threshold (default: -1.0, < 0 - disabled)<br/>(env: LLAMA_ARG_DEFRAG_THOLD) |
-| `-np, --parallel N` | number of parallel sequences to decode (default: 1)<br/>(env:  LLAMA_ARG_N_PARALLEL) |
+| `-np, --parallel N` | number of parallel sequences to decode (default: 1)<br/>(env: LLAMA_ARG_N_PARALLEL) |
 | `-cb, --cont-batching` | enable continuous batching (a.k.a dynamic batching) (default: enabled)<br/>(env: LLAMA_ARG_CONT_BATCHING) |
 | `-nocb, --no-cont-batching` | disable continuous batching<br/>(env: LLAMA_ARG_NO_CONT_BATCHING) |
 | `--mlock` | force system to keep model in RAM rather than swapping or compressing |
@@ -128,12 +129,13 @@ The project is under active development, and we are [looking for feedback and co
 | `-sps, --slot-prompt-similarity SIMILARITY` | how much the prompt of a request must match the prompt of a slot in order to use that slot (default: 0.50, 0.0 = disabled)<br/> |
 | `--lora-init-without-apply` | load LoRA adapters without applying them (apply later via POST /lora-adapters) (default: disabled) |
 | `-ld, --logdir LOGDIR` | path under which to save YAML logs (no logging if unset) |
-| `--log-test` | Log test |
 | `--log-disable` | Log disable |
-| `--log-enable` | Log enable |
-| `--log-new` | Log new |
-| `--log-append` | Log append |
-| `--log-file FNAME` | Log file |
+| `--log-file FNAME` | Log to file |
+| `--log-colors` | Enable colored logging<br/>(env: LLAMA_LOG_COLORS) |
+| `-v, --verbose, --log-verbose` | Set verbosity level to infinity (i.e. log all messages, useful for debugging) |
+| `-lv, --verbosity, --log-verbosity N` | Set the verbosity threshold. Messages with a higher verbosity will be ignored.<br/>(env: LLAMA_LOG_VERBOSITY) |
+| `--log-prefix` | Enable prefx in log messages<br/>(env: LLAMA_LOG_PREFIX) |
+| `--log-timestamps` | Enable timestamps in log messages<br/>(env: LLAMA_LOG_TIMESTAMPS) |

 Note: If both command line argument and environment variable are both set for the same param, the argument will take precedence over env var.

@@ -1180,6 +1180,15 @@ struct server_context {
            SLT_DBG(slot, "stopped by limit, n_decoded = %d, n_predict = %d\n", slot.n_decoded, slot.params.n_predict);
        }

+        // if context shift is disabled, we stop when it reaches the context limit
+        if (slot.n_decoded >= slot.n_ctx) {
+            slot.truncated      = true;
+            slot.stopped_limit  = true;
+            slot.has_next_token = false;
+
+            SLT_DBG(slot, "stopped due to running out of context capacity, n_decoded = %d, n_ctx = %d\n", slot.n_decoded, slot.n_ctx);
+        }
+
        if (llama_token_is_eog(model, result.tok)) {
            slot.stopped_eos    = true;
            slot.has_next_token = false;
@@ -1480,7 +1489,7 @@ struct server_context {
            if (result.error) {
                error_handler(result.data);
                cancel_tasks(id_tasks);
-                break;
+                return;
            }

            size_t idx = result.data["index"];
@@ -1827,6 +1836,14 @@ struct server_context {
        for (server_slot & slot : slots) {
            if (slot.ga_n == 1) {
                if (slot.is_processing() && (int) system_tokens.size() + slot.n_past >= slot.n_ctx - 1) {
+                    if (!params.ctx_shift) {
+                        // this check is redundant (for good)
+                        // we should never get here, because generation should already stopped in process_token()
+                        slot.release();
+                        send_error(slot, "context shift is disabled", ERROR_TYPE_SERVER);
+                        continue;
+                    }
+
                    // Shift context
                    const int n_keep    = slot.params.n_keep + add_bos_token;
                    const int n_left    = (int) system_tokens.size() + slot.n_past - n_keep;
@@ -1961,6 +1978,14 @@ struct server_context {
                                continue;
                            }
                        } else {
+                            if (!params.ctx_shift) {
+                                // if context shift is disabled, we make sure prompt size is smaller than KV size
+                                if ((int) system_tokens.size() + slot.n_prompt_tokens >= slot.n_ctx) {
+                                    slot.release();
+                                    send_error(slot, "the request exceeds the available context size. try increasing the context size or enable context shift", ERROR_TYPE_INVALID_REQUEST);
+                                    continue;
+                                }
+                            }
                            if (slot.params.n_keep < 0) {
                                slot.params.n_keep = slot.n_prompt_tokens;
                            }
@@ -3154,7 +3179,7 @@ int main(int argc, char ** argv) {
    }

    // print sample chat example to make it clear which template is used
-    LOG_INF("%s: chat template, built_in: %d, chat_example: '%s\n'", __func__, params.chat_template.empty(), llama_chat_format_example(ctx_server.model, params.chat_template).c_str());
+    LOG_INF("%s: chat template, built_in: %d, chat_example: '%s'\n", __func__, params.chat_template.empty(), llama_chat_format_example(ctx_server.model, params.chat_template).c_str());

    ctx_server.queue_tasks.on_new_task(std::bind(
                &server_context::process_single_task, &ctx_server, std::placeholders::_1));
@@ -0,0 +1,62 @@
+@llama.cpp
+@ctx_shift
+Feature: llama.cpp server
+
+  Background: Server startup
+    Given a server listening on localhost:8080
+    And   a model file tinyllamas/stories260K.gguf from HF repo ggml-org/models
+    And   a model file test-model.gguf
+    And   a model alias tinyllama-2
+    And   BOS token is 1
+    And   42 as server seed
+    And   256 KV cache size
+    And   32 as batch size
+    And   2 slots
+
+  Scenario: Inference with context shift
+    And   64 server max tokens to predict
+    Then  the server is starting
+    Then  the server is healthy
+    Given a prompt:
+    """
+    Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua.
+    Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat.
+    Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur.
+    Excepteur sint occaecat cupidatat non proident, sunt in culpa qui officia deserunt mollit anim id est laborum.
+    """
+    And   a completion request with no api error
+    Then  64 tokens are predicted matching fun|Annaks|popcorns|pictry|bowl
+    And   the completion is  truncated
+    And   109 prompt tokens are processed
+
+  Scenario Outline: Inference without context shift
+    And   <n_predict> server max tokens to predict
+    And   disable context shifting
+    Then  the server is starting
+    Then  the server is healthy
+    Given a prompt:
+    """
+    Hi how are you
+    """
+    And   a completion request with no api error
+    Then  <n_token_output> tokens are predicted matching twind|Anna
+    And   the completion is <truncated> truncated
+    And   8 prompt tokens are processed
+    Examples:
+      | n_predict | n_token_output | truncated |
+      | 64        | 64             | not       |
+      | -1        | 120            |           |
+
+  Scenario: Inference without context shift (expected error: prompt too long)
+    And   disable context shifting
+    Then  the server is starting
+    Then  the server is healthy
+    Given a prompt:
+    """
+    Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua.
+    Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat.
+    Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur.
+    Excepteur sint occaecat cupidatat non proident, sunt in culpa qui officia deserunt mollit anim id est laborum.
+    """
+    And   a completion request with 400 api error
+
@@ -10,11 +10,11 @@ Feature: llama.cpp server
    And   42 as server seed
    And   2 slots
    # the bert-bge-small model has context size of 512
-    # since the generated prompts are as big as the batch size, we need to set the batch size to 512
+    # since the generated prompts are as big as the batch size, we need to set the batch size to <= 512
    # ref: https://huggingface.co/BAAI/bge-small-en-v1.5/blob/5c38ec7c405ec4b44b94cc5a9bb96e735b38267a/config.json#L20
-    And   512 as batch size
-    And   512 as ubatch size
-    And   2048 KV cache size
+    And   128 as batch size
+    And   128 as ubatch size
+    And   512 KV cache size
    And   embeddings extraction
    Then  the server is starting
    Then  the server is healthy
@@ -26,6 +26,20 @@ Feature: llama.cpp server
    """
    Then embeddings are generated

+  Scenario: Embedding (error: prompt too long)
+    When embeddings are computed for:
+    """
+    Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua.
+    Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat.
+    Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur.
+    Excepteur sint occaecat cupidatat non proident, sunt in culpa qui officia deserunt mollit anim id est laborum.
+    Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua.
+    Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat.
+    Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur.
+    Excepteur sint occaecat cupidatat non proident, sunt in culpa qui officia deserunt mollit anim id est laborum.
+    """
+    And  embeddings request with 500 api error
+
  Scenario: OAI Embeddings compatibility
    Given a model bert-bge-small
    When an OAI compatible embeddings computation request for:
@@ -77,6 +77,7 @@ def step_server_config(context, server_fqdn: str, server_port: str):
    context.response_format = None
    context.temperature = None
    context.lora_file = None
+    context.disable_ctx_shift = False

    context.tasks_result = []
    context.concurrent_tasks = []
@@ -148,7 +149,7 @@ def step_n_slots(context, n_slots: int):

@step('{n_predict:d} server max tokens to predict')
 def step_server_n_predict(context, n_predict: int):
-    context.n_server_predict = n_predict
+    context.n_server_predict = n_predict if n_predict > 0 else None


@step('{slot_save_path} as slot save path')
@@ -180,6 +181,9 @@ def step_server_embeddings(context):
 def step_server_metrics(context):
    context.server_metrics = True

+@step('disable context shifting')
+def step_server_disable_ctx_shift(context):
+    context.disable_ctx_shift = True

@step("the server is starting")
 def step_start_server(context):
@@ -257,7 +261,7 @@ async def step_all_slots_status(context, expected_slot_status_string: Literal['i
@step('a completion request with {api_error} api error')
@async_run_until_complete
 async def step_request_completion(context, api_error: Literal['raised'] | str):
-    expect_api_error = api_error == 'raised'
+    expect_api_error = api_error == 'raised' or api_error != 'no'
    seeds = await completions_seed(context, num_seeds=1)
    completion = await request_completion(context.prompts.pop(),
                                          seeds[0] if seeds is not None else seeds,
@@ -272,8 +276,11 @@ async def step_request_completion(context, api_error: Literal['raised'] | str):
    context.tasks_result.append(completion)
    if context.debug:
        print(f"Completion response: {completion}")
-    if expect_api_error:
+    if api_error == 'raised':
        assert completion == 401, f"completion must be an 401 status code: {completion}"
+    elif api_error.isdigit():
+        api_error_code = int(api_error)
+        assert completion == api_error_code, f"completion must be an {api_error_code} status code: {completion}"


@step('{predicted_n:d} tokens are predicted matching {re_content}')
@@ -645,6 +652,9 @@ def step_assert_embeddings(context):
    for embedding in context.embeddings:
        assert_embeddings(embedding)

+@step('embeddings request with {api_error_code:d} api error')
+def step_assert_embeddings(context, api_error_code: int):
+    assert context.embeddings == api_error_code, f"embeddings request must return code {api_error_code}, but got {context.embeddings}"

@step('an OAI compatible embeddings computation request for')
@async_run_until_complete
@@ -1089,15 +1099,17 @@ async def oai_chat_completions(user_prompt,
    return completion_response


-async def request_embedding(content, seed, base_url=None) -> list[list[float]]:
+async def request_embedding(content, seed, base_url=None) -> list[list[float]] | int:
    async with aiohttp.ClientSession(timeout=DEFAULT_TIMEOUT_SECONDS) as session:
        async with session.post(f'{base_url}/embedding',
                                json={
                                    "content": content,
                                }) as response:
-            assert response.status == 200
-            response_json = await response.json()
-            return [response_json['embedding']]
+            if response.status == 200:
+                response_json = await response.json()
+                return [response_json['embedding']]
+            else:
+                return response.status


 async def request_oai_embeddings(input, seed,
@@ -1372,6 +1384,8 @@ def start_server_background(context):
        server_args.append('--verbose')
    if context.lora_file:
        server_args.extend(['--lora', context.lora_file])
+    if context.disable_ctx_shift:
+        server_args.extend(['--no-context-shift'])

    args = [str(arg) for arg in [context.server_path, *server_args]]
    print(f"bench: starting server with: {' '.join(args)}")
@@ -32,6 +32,9 @@ struct seq_draft {
 int main(int argc, char ** argv) {
    gpt_params params;

+    // needed to get candidate probs even for temp <= 0.0
+    params.sparams.n_probs = 128;
+
    if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_SPECULATIVE)) {
        return 1;
    }
@@ -49,7 +52,7 @@ int main(int argc, char ** argv) {
    // probability threshold for splitting a draft branch (only for n_seq_dft > 1)
    const float p_split  = params.p_split;

-    std::default_random_engine rng(params.sparams.seed);
+    std::default_random_engine rng(params.sparams.seed == LLAMA_DEFAULT_SEED ? std::random_device()() : params.sparams.seed);
    std::uniform_real_distribution<> u_dist;

    // init llama.cpp
@@ -20,11 +20,11 @@
    },
    "nixpkgs": {
      "locked": {
-        "lastModified": 1726062873,
-        "narHash": "sha256-IiA3jfbR7K/B5+9byVi9BZGWTD4VSbWe8VLpp9B/iYk=",
+        "lastModified": 1726755586,
+        "narHash": "sha256-PmUr/2GQGvFTIJ6/Tvsins7Q43KTMvMFhvG6oaYK+Wk=",
        "owner": "NixOS",
        "repo": "nixpkgs",
-        "rev": "4f807e8940284ad7925ebd0a0993d2a1791acb2f",
+        "rev": "c04d5652cfa9742b1d519688f65d1bbccea9eb7e",
        "type": "github"
      },
      "original": {
@@ -570,6 +570,7 @@ extern "C" {
        GGML_LOG_LEVEL_WARN  = 2,
        GGML_LOG_LEVEL_ERROR = 3,
        GGML_LOG_LEVEL_DEBUG = 4,
+        GGML_LOG_LEVEL_CONT  = 5, // continue previous log
    };

    // this tensor...
@@ -1979,6 +1980,9 @@ extern "C" {
    typedef void (*ggml_custom2_op_t)(struct ggml_tensor * dst , const struct ggml_tensor * a, const struct ggml_tensor * b, int ith, int nth, void * userdata);
    typedef void (*ggml_custom3_op_t)(struct ggml_tensor * dst , const struct ggml_tensor * a, const struct ggml_tensor * b, const struct ggml_tensor * c, int ith, int nth, void * userdata);

+#define GGML_N_TASKS_MAX (-1)
+    // n_tasks == GGML_N_TASKS_MAX means to use max number of tasks
+
    GGML_API struct ggml_tensor * ggml_map_custom1(
            struct ggml_context   * ctx,
            struct ggml_tensor    * a,
@@ -1186,6 +1186,7 @@ elseif (CMAKE_OSX_ARCHITECTURES STREQUAL "x86_64" OR CMAKE_GENERATOR_PLATFORM_LW
        endif()
        if (GGML_AVX512)
            list(APPEND ARCH_FLAGS -mavx512f)
+            list(APPEND ARCH_FLAGS -mavx512dq)
            list(APPEND ARCH_FLAGS -mavx512bw)
        endif()
        if (GGML_AVX512_VBMI)
@@ -39,11 +39,44 @@
 //
 #if defined(__AVX__)
 #if defined(__F16C__)
+#if defined(__AVX512F__)
+#define GGML_F32Cx8x2_LOAD(x, y)     _mm512_cvtph_ps(_mm256_set_m128i(_mm_loadu_si128((const __m128i *)(y)), _mm_loadu_si128((const __m128i *)(x))))
+#define GGML_F32Cx16_REPEAT_LOAD(x)  _mm512_cvtph_ps(_mm256_set_m128i(x, x))
+#endif
 // the  _mm256_cvt intrinsics require F16C
 #define GGML_F32Cx8_LOAD(x)     _mm256_cvtph_ps(_mm_loadu_si128((const __m128i *)(x)))
 #define GGML_F32Cx8_REPEAT_LOAD(x, loadMask)     _mm256_cvtph_ps(_mm_shuffle_epi32(_mm_maskload_epi32((int const*)(x), loadMask), 68))
 #define GGML_F32Cx8_REARRANGE_LOAD(x, arrangeMask)     _mm256_cvtph_ps(_mm_shuffle_epi8(_mm_loadu_si128((const __m128i *) x), arrangeMask))
 #else
+#if defined(__AVX512F__)
+static inline __m512 __avx512_f32cx8x2_load(ggml_fp16_t *x, ggml_fp16_t *y) {
+    float tmp[16];
+
+    for (int i = 0; i < 8; i++) {
+        tmp[i] = GGML_FP16_TO_FP32(x[i]);
+    }
+
+    for (int i = 0; i < 8; i++) {
+        tmp[i + 8] = GGML_FP16_TO_FP32(y[i]);
+    }
+
+    return _mm512_loadu_ps(tmp);
+}
+static inline __m512 __avx512_repeat_f32cx16_load(__m128i x) {
+    float tmp[16];
+    uint16_t tmphalf[8];
+    _mm_storeu_si128((__m128i*)tmphalf, x);
+
+    for (int i = 0; i < 4; i++) {
+        tmp[i] = GGML_FP16_TO_FP32(tmphalf[i]);
+        tmp[i + 4] = GGML_FP16_TO_FP32(tmphalf[i]);
+        tmp[i + 8] = GGML_FP16_TO_FP32(tmphalf[i]);
+        tmp[i + 12] = GGML_FP16_TO_FP32(tmphalf[i]);
+    }
+
+    return _mm512_loadu_ps(tmp);
+}
+#endif
 static inline __m256 __avx_f32cx8_load(ggml_fp16_t *x) {
    float tmp[8];

@@ -78,30 +111,65 @@ static inline __m256 __avx_rearranged_f32cx8_load(ggml_fp16_t *x, __m128i arrang
 #define GGML_F32Cx8_LOAD(x)     __avx_f32cx8_load(x)
 #define GGML_F32Cx8_REPEAT_LOAD(x, loadMask)     __avx_repeat_f32cx8_load(x)
 #define GGML_F32Cx8_REARRANGE_LOAD(x, arrangeMask)     __avx_rearranged_f32cx8_load(x, arrangeMask)
+#if defined(__AVX512F__)
+#define GGML_F32Cx8x2_LOAD(x, y)     __avx512_f32cx8x2_load(x, y)
+#define GGML_F32Cx16_REPEAT_LOAD(x)  __avx512_repeat_f32cx16_load(x)
+#endif
 #endif
 #endif


 #if defined(__AVX2__) || defined(__AVX512F__)
-static inline __m256i sum_i16_pairs_int(const __m256i x) {
+#if defined(__AVX512F__)
+// add int16_t pairwise and return as 512 bit int vector
+static inline __m512i sum_i16_pairs_int_32x16(const __m512i x) {
+    const __m512i ones = _mm512_set1_epi16(1);
+    return _mm512_madd_epi16(ones, x);
+}
+
+static inline __m512i mul_sum_us8_pairs_int32x16(const __m512i ax, const __m512i sy) {
+#if defined(__AVXVNNI__) || (defined(__AVX512VNNI__) && defined(__AVX512VL__))
+    const __m512i zero = _mm512_setzero_si512();
+    return _mm512_dpbusd_epi32(zero, ax, sy);
+#else
+    // Perform multiplication and create 16-bit values
+    const __m512i dot = _mm512_maddubs_epi16(ax, sy);
+    return sum_i16_pairs_int_32x16(dot);
+#endif
+}
+
+// multiply int8_t, add results pairwise twice and return as 512 bit int vector
+static inline __m512i mul_sum_i8_pairs_int32x16(const __m512i x, const __m512i y) {
+    const __m512i zero = _mm512_setzero_si512();
+    // Get absolute values of x vectors
+    const __m512i ax = _mm512_abs_epi8(x);
+    // Sign the values of the y vectors
+    __mmask64 blt0 = _mm512_movepi8_mask(x);
+    const __m512i sy = _mm512_mask_sub_epi8(y, blt0, zero, y);
+    return mul_sum_us8_pairs_int32x16(ax, sy);
+}
+#endif
+
+// add int16_t pairwise and return as 256 bit int vector
+static inline __m256i sum_i16_pairs_int32x8(const __m256i x) {
    const __m256i ones = _mm256_set1_epi16(1);
    return _mm256_madd_epi16(ones, x);
 }

-static inline __m256i mul_sum_us8_pairs_int(const __m256i ax, const __m256i sy) {
+static inline __m256i mul_sum_us8_pairs_int32x8(const __m256i ax, const __m256i sy) {
 #if defined(__AVXVNNI__) || (defined(__AVX512VNNI__) && defined(__AVX512VL__))
    const __m256i zero = _mm256_setzero_si256();
    return _mm256_dpbusd_epi32(zero, ax, sy);
 #else
    // Perform multiplication and create 16-bit values
    const __m256i dot = _mm256_maddubs_epi16(ax, sy);
-    return sum_i16_pairs_int(dot);
+    return sum_i16_pairs_int32x8(dot);
 #endif
 }

 // Integer variant of the function defined in ggml-quants.c
-// multiply int8_t, add results pairwise twice and return as float vector
-static inline __m256i mul_sum_i8_pairs_int(const __m256i x, const __m256i y) {
+// multiply int8_t, add results pairwise twice and return as 256 bit int vector
+static inline __m256i mul_sum_i8_pairs_int32x8(const __m256i x, const __m256i y) {
 #if __AVXVNNIINT8__
    const __m256i zero = _mm256_setzero_si256();
    return _mm256_dpbssd_epi32(zero, x, y);
@@ -110,7 +178,7 @@ static inline __m256i mul_sum_i8_pairs_int(const __m256i x, const __m256i y) {
    const __m256i ax = _mm256_sign_epi8(x, x);
    // Sign the values of the y vectors
    const __m256i sy = _mm256_sign_epi8(y, x);
-    return mul_sum_us8_pairs_int(ax, sy);
+    return mul_sum_us8_pairs_int32x8(ax, sy);
 #endif
 }
 #endif
@@ -929,17 +997,17 @@ void ggml_gemv_q4_0_8x8_q8_0(int n, float * restrict s, size_t bs, const void *
                // ...........................................................................
                // B0(28-31) B4(28-31) B1(28-31) B5(28-31) B2(28-31) B6(28-31) B3(28-31) B7(28-31) with A0(28-31)

-                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int(_mm256_blend_epi32(rhs_vec_0123_0 ,_mm256_shuffle_epi32(rhs_vec_4567_0, 177), 170), _mm256_shuffle_epi32(lhs_vec_0, 0)));
-                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int(_mm256_blend_epi32(_mm256_shuffle_epi32(rhs_vec_0123_0, 177) ,rhs_vec_4567_0, 170), _mm256_shuffle_epi32(lhs_vec_0, 85)));
+                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int32x8(_mm256_blend_epi32(rhs_vec_0123_0 ,_mm256_shuffle_epi32(rhs_vec_4567_0, 177), 170), _mm256_shuffle_epi32(lhs_vec_0, 0)));
+                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int32x8(_mm256_blend_epi32(_mm256_shuffle_epi32(rhs_vec_0123_0, 177) ,rhs_vec_4567_0, 170), _mm256_shuffle_epi32(lhs_vec_0, 85)));

-                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int(_mm256_blend_epi32(rhs_vec_0123_1 ,_mm256_shuffle_epi32(rhs_vec_4567_1, 177), 170), _mm256_shuffle_epi32(lhs_vec_0, 170)));
-                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int(_mm256_blend_epi32(_mm256_shuffle_epi32(rhs_vec_0123_1, 177) ,rhs_vec_4567_1, 170), _mm256_shuffle_epi32(lhs_vec_0, 255)));
+                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int32x8(_mm256_blend_epi32(rhs_vec_0123_1 ,_mm256_shuffle_epi32(rhs_vec_4567_1, 177), 170), _mm256_shuffle_epi32(lhs_vec_0, 170)));
+                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int32x8(_mm256_blend_epi32(_mm256_shuffle_epi32(rhs_vec_0123_1, 177) ,rhs_vec_4567_1, 170), _mm256_shuffle_epi32(lhs_vec_0, 255)));

-                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int(_mm256_blend_epi32(rhs_vec_0123_2 ,_mm256_shuffle_epi32(rhs_vec_4567_2, 177), 170), _mm256_shuffle_epi32(lhs_vec_1, 0)));
-                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int(_mm256_blend_epi32(_mm256_shuffle_epi32(rhs_vec_0123_2, 177) ,rhs_vec_4567_2, 170), _mm256_shuffle_epi32(lhs_vec_1, 85)));
+                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int32x8(_mm256_blend_epi32(rhs_vec_0123_2 ,_mm256_shuffle_epi32(rhs_vec_4567_2, 177), 170), _mm256_shuffle_epi32(lhs_vec_1, 0)));
+                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int32x8(_mm256_blend_epi32(_mm256_shuffle_epi32(rhs_vec_0123_2, 177) ,rhs_vec_4567_2, 170), _mm256_shuffle_epi32(lhs_vec_1, 85)));

-                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int(_mm256_blend_epi32(rhs_vec_0123_3 ,_mm256_shuffle_epi32(rhs_vec_4567_3, 177), 170), _mm256_shuffle_epi32(lhs_vec_1, 170)));
-                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int(_mm256_blend_epi32(_mm256_shuffle_epi32(rhs_vec_0123_3, 177) ,rhs_vec_4567_3, 170), _mm256_shuffle_epi32(lhs_vec_1, 255)));
+                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int32x8(_mm256_blend_epi32(rhs_vec_0123_3 ,_mm256_shuffle_epi32(rhs_vec_4567_3, 177), 170), _mm256_shuffle_epi32(lhs_vec_1, 170)));
+                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int32x8(_mm256_blend_epi32(_mm256_shuffle_epi32(rhs_vec_0123_3, 177) ,rhs_vec_4567_3, 170), _mm256_shuffle_epi32(lhs_vec_1, 255)));

                // Accumulated values multipled with appropriate scales
                acc_row = _mm256_fmadd_ps(_mm256_cvtepi32_ps(iacc), _mm256_mul_ps(col_scale_f32, row_scale_f32), acc_row);
@@ -2421,10 +2489,411 @@ void ggml_gemm_q4_0_8x8_q8_0(int n, float * restrict s, size_t bs, const void *
    __m256i signextendlut = _mm256_castsi128_si256(_mm_set_epi8(-1, -2, -3, -4, -5, -6, -7, -8, 7, 6, 5, 4, 3, 2, 1, 0));
    signextendlut = _mm256_permute2f128_si256(signextendlut, signextendlut, 0);
    // Permute mask used for easier vector processing at later stages
-    __m256i requiredOrder = _mm256_set_epi32(3 ,2 ,1 ,0, 7 ,6, 5, 4);
+    __m256i requiredOrder = _mm256_set_epi32(3, 2, 1, 0, 7, 6, 5, 4);
+    int64_t xstart = 0;
+    int anr = nr - nr%16; // Used to align nr with boundary of 16
+#ifdef __AVX512F__
+    int anc = nc - nc%16; // Used to align nc with boundary of 16
+    // Mask to mask out nibbles from packed bytes expanded to 512 bit length
+    const __m512i m4bexpanded = _mm512_set1_epi8(0x0F);
+    // Lookup table to convert signed nibbles to signed bytes expanded to 512 bit length
+    __m512i signextendlutexpanded = _mm512_inserti32x8(_mm512_castsi256_si512(signextendlut), signextendlut, 1);
+
+    // Take group of four block_q8_0x4 structures at each pass of the loop and perform dot product operation
+    for (; y < anr / 4; y += 4) {
+
+        const block_q8_0x4 * a_ptrs[4];
+
+        a_ptrs[0] = a_ptr_start + (y * nb);
+        for (int i = 0; i < 3; ++i) {
+            a_ptrs[i + 1] = a_ptrs[i] + nb;
+        }
+
+        // Take group of two block_q4_0x8 structures at each pass of the loop and perform dot product operation
+        for (int64_t x = 0; x < anc / 8; x += 2) {
+
+            const block_q4_0x8 * b_ptr_0 = b_ptr_start + ((x)     * b_nb);
+            const block_q4_0x8 * b_ptr_1 = b_ptr_start + ((x + 1) * b_nb);
+
+            // Master FP accumulators
+            __m512 acc_rows[16];
+            for (int i = 0; i < 16; i++) {
+                acc_rows[i] = _mm512_setzero_ps();
+            }
+
+            for (int64_t b = 0; b < nb; b++) {
+                // Load the sixteen block_q4_0 quantized values interleaved with each other in chunks of eight - B0,B1 ....BE,BF
+                const __m256i rhs_raw_mat_0123_0 = _mm256_loadu_si256((const __m256i *)(b_ptr_0[b].qs));
+                const __m256i rhs_raw_mat_4567_0 = _mm256_loadu_si256((const __m256i *)(b_ptr_0[b].qs + 32));
+                const __m256i rhs_raw_mat_0123_1 = _mm256_loadu_si256((const __m256i *)(b_ptr_0[b].qs + 64));
+                const __m256i rhs_raw_mat_4567_1 = _mm256_loadu_si256((const __m256i *)(b_ptr_0[b].qs + 96));
+
+                const __m256i rhs_raw_mat_89AB_0 = _mm256_loadu_si256((const __m256i *)(b_ptr_1[b].qs));
+                const __m256i rhs_raw_mat_CDEF_0 = _mm256_loadu_si256((const __m256i *)(b_ptr_1[b].qs + 32));
+                const __m256i rhs_raw_mat_89AB_1 = _mm256_loadu_si256((const __m256i *)(b_ptr_1[b].qs + 64));
+                const __m256i rhs_raw_mat_CDEF_1 = _mm256_loadu_si256((const __m256i *)(b_ptr_1[b].qs + 96));
+
+                // Save the values in the following vectors in the formats B0B1B4B5B8B9BCBD, B2B3B6B7BABBBEBF for further processing and storing of values
+                const __m256i rhs_raw_mat_0145_0 = _mm256_blend_epi32(rhs_raw_mat_0123_0, _mm256_permutevar8x32_epi32(rhs_raw_mat_4567_0, requiredOrder), 240);
+                const __m256i rhs_raw_mat_2367_0 = _mm256_blend_epi32(_mm256_permutevar8x32_epi32(rhs_raw_mat_0123_0, requiredOrder), rhs_raw_mat_4567_0, 240);
+                const __m256i rhs_raw_mat_0145_1 = _mm256_blend_epi32(rhs_raw_mat_0123_1, _mm256_permutevar8x32_epi32(rhs_raw_mat_4567_1, requiredOrder), 240);
+                const __m256i rhs_raw_mat_2367_1 = _mm256_blend_epi32(_mm256_permutevar8x32_epi32(rhs_raw_mat_0123_1, requiredOrder), rhs_raw_mat_4567_1, 240);
+
+                const __m256i rhs_raw_mat_89CD_0 = _mm256_blend_epi32(rhs_raw_mat_89AB_0, _mm256_permutevar8x32_epi32(rhs_raw_mat_CDEF_0, requiredOrder), 240);
+                const __m256i rhs_raw_mat_ABEF_0 = _mm256_blend_epi32(_mm256_permutevar8x32_epi32(rhs_raw_mat_89AB_0, requiredOrder), rhs_raw_mat_CDEF_0, 240);
+                const __m256i rhs_raw_mat_89CD_1 = _mm256_blend_epi32(rhs_raw_mat_89AB_1, _mm256_permutevar8x32_epi32(rhs_raw_mat_CDEF_1, requiredOrder), 240);
+                const __m256i rhs_raw_mat_ABEF_1 = _mm256_blend_epi32(_mm256_permutevar8x32_epi32(rhs_raw_mat_89AB_1, requiredOrder), rhs_raw_mat_CDEF_1, 240);
+
+                const __m512i rhs_raw_mat_014589CD_0 = _mm512_inserti32x8(_mm512_castsi256_si512(rhs_raw_mat_0145_0), rhs_raw_mat_89CD_0, 1);
+                const __m512i rhs_raw_mat_2367ABEF_0 = _mm512_inserti32x8(_mm512_castsi256_si512(rhs_raw_mat_2367_0), rhs_raw_mat_ABEF_0, 1);
+                const __m512i rhs_raw_mat_014589CD_1 = _mm512_inserti32x8(_mm512_castsi256_si512(rhs_raw_mat_0145_1), rhs_raw_mat_89CD_1, 1);
+                const __m512i rhs_raw_mat_2367ABEF_1 = _mm512_inserti32x8(_mm512_castsi256_si512(rhs_raw_mat_2367_1), rhs_raw_mat_ABEF_1, 1);
+
+                // 4-bit -> 8-bit - Sign is maintained
+                const __m512i rhs_mat_014589CD_0 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(rhs_raw_mat_014589CD_0, m4bexpanded)); //B0(0-7) B1(0-7) B4(0-7) B5(0-7) B8(0-7) B9(0-7) BC(0-7) BD(0-7)
+                const __m512i rhs_mat_2367ABEF_0 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(rhs_raw_mat_2367ABEF_0, m4bexpanded)); //B2(0-7) B3(0-7) B6(0-7) B7(0-7) BA(0-7) BB(0-7) BE(0-7) BF(0-7)
+
+                const __m512i rhs_mat_014589CD_1 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(rhs_raw_mat_014589CD_1, m4bexpanded)); //B0(8-15) B1(8-15) B4(8-15) B5(8-15) B8(8-15) B9(8-15) BC(8-15) BD(8-15)
+                const __m512i rhs_mat_2367ABEF_1 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(rhs_raw_mat_2367ABEF_1, m4bexpanded)); //B2(8-15) B3(8-15) B6(8-15) B7(8-15) BA(8-15) BB(8-15) BE(8-15) BF(8-15)
+
+                const __m512i rhs_mat_014589CD_2 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(_mm512_srli_epi16(rhs_raw_mat_014589CD_0, 4), m4bexpanded)); //B0(16-23) B1(16-23) B4(16-23) B5(16-23) B8(16-23) B9(16-23) BC(16-23) BD(16-23)
+                const __m512i rhs_mat_2367ABEF_2 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(_mm512_srli_epi16(rhs_raw_mat_2367ABEF_0, 4), m4bexpanded)); //B2(16-23) B3(16-23) B6(16-23) B7(16-23) BA(16-23) BB(16-23) BE(16-23) BF(16-23)
+
+                const __m512i rhs_mat_014589CD_3 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(_mm512_srli_epi16(rhs_raw_mat_014589CD_1, 4), m4bexpanded)); //B0(24-31) B1(24-31) B4(24-31) B5(24-31) B8(24-31) B9(24-31) BC(24-31) BD(24-31)
+                const __m512i rhs_mat_2367ABEF_3 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(_mm512_srli_epi16(rhs_raw_mat_2367ABEF_1, 4), m4bexpanded)); //B2(24-31) B3(24-31) B6(24-31) B7(24-31) BA(24-31) BB(24-31) BE(24-31) BF(24-31)
+
+                // Shuffle pattern one - right side input
+                const __m512i rhs_mat_014589CD_0_sp1 = _mm512_shuffle_epi32(rhs_mat_014589CD_0, 136); //B0(0-3) B1(0-3) B0(0-3) B1(0-3) B4(0-3) B5(0-3) B4(0-3) B5(0-3) B8(0-3) B9(0-3) B8(0-3) B9(0-3) BC(0-3) BD(0-3) BC(0-3) BD(0-3)
+                const __m512i rhs_mat_2367ABEF_0_sp1 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_0, 136); //B2(0-3) B3(0-3) B2(0-3) B3(0-3) B6(0-3) B7(0-3) B6(0-3) B7(0-3) BA(0-3) BB(0-3) BA(0-3) BB(0-3) BE(0-3) BF(0-3) BE(0-3) BF(0-3)
+
+                const __m512i rhs_mat_014589CD_1_sp1 = _mm512_shuffle_epi32(rhs_mat_014589CD_1, 136); //B0(8-11) B1(8-11) B0(8-11) B1(8-11) B4(8-11) B5(8-11) B4(8-11) B5(8-11) B8(8-11) B9(8-11) B8(8-11) B9(8-11) BC(8-11) BD(8-11) BC(8-11) BD(8-11)
+                const __m512i rhs_mat_2367ABEF_1_sp1 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_1, 136); //B2(8-11) B3(8-11) B2(8-11) B3(8-11) B6(8-11) B7(8-11) B6(8-11) B7(8-11) BA(8-11) BB(8-11) BA(8-11) BB(8-11) BE(8-11) BF(8-11) BE(8-11) BF(8-11)
+
+                const __m512i rhs_mat_014589CD_2_sp1 = _mm512_shuffle_epi32(rhs_mat_014589CD_2, 136); //B0(16-19) B1(16-19) B0(16-19) B1(16-19) B4(16-19) B5(16-19) B4(16-19) B5(16-19) B8(16-19) B9(16-19) B8(16-19) B9(16-19) BC(16-19) BD(16-19) BC(16-19) BD(16-19)
+                const __m512i rhs_mat_2367ABEF_2_sp1 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_2, 136); //B2(16-19) B3(16-19) B2(16-19) B3(16-19) B6(16-19) B7(16-19) B6(16-19) B7(16-19) BA(16-19) BB(16-19) BA(16-19) BB(16-19) BE(16-19) BF(16-19) BE(16-19) BF(16-19)
+
+                const __m512i rhs_mat_014589CD_3_sp1 = _mm512_shuffle_epi32(rhs_mat_014589CD_3, 136); //B0(24-27) B1(24-27) B0(24-27) B1(24-27) B4(24-27) B5(24-27) B4(24-27) B5(24-27) B8(24-27) B9(24-27) B8(24-27) B9(24-27) BC(24-27) BD(24-27) BC(24-27) BD(24-27)
+                const __m512i rhs_mat_2367ABEF_3_sp1 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_3, 136); //B2(24-27) B3(24-27) B2(24-27) B3(24-27) B6(24-27) B7(24-27) B6(24-27) B7(24-27) BA(24-27) BB(24-27) BA(24-27) BB(24-27) BE(24-27) BF(24-27) BE(24-27) BF(24-27)
+
+                // Shuffle pattern two - right side input
+
+                const __m512i rhs_mat_014589CD_0_sp2 = _mm512_shuffle_epi32(rhs_mat_014589CD_0, 221); //B0(4-7) B1(4-7) B0(4-7) B1(4-7) B4(4-7) B5(4-7) B4(4-7) B5(4-7) B8(4-7) B9(4-7) B8(4-7) B9(4-7) BC(4-7) BD(4-7) BC(4-7) BD(4-7)
+                const __m512i rhs_mat_2367ABEF_0_sp2 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_0, 221); //B2(4-7) B3(4-7) B2(4-7) B3(4-7) B6(4-7) B7(4-7) B6(4-7) B7(4-7) BA(4-7) BB(4-7) BA(4-7) BB(4-7) BE(4-7) BF(4-7) BE(4-7) BF(4-7)
+
+                const __m512i rhs_mat_014589CD_1_sp2 = _mm512_shuffle_epi32(rhs_mat_014589CD_1, 221); //B0(12-15) B1(12-15) B0(12-15) B1(12-15) B4(12-15) B5(12-15) B4(12-15) B5(12-15) B8(12-15) B9(12-15) B8(12-15) B9(12-15) BC(12-15) BD(12-15) BC(12-15) BD(12-15)
+                const __m512i rhs_mat_2367ABEF_1_sp2 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_1, 221); //B2(12-15) B3(12-15) B2(12-15) B3(12-15) B6(12-15) B7(12-15) B6(12-15) B7(12-15) BA(12-15) BB(12-15) BA(12-15) BB(12-15) BE(12-15) BF(12-15) BE(12-15) BF(12-15)
+
+                const __m512i rhs_mat_014589CD_2_sp2 = _mm512_shuffle_epi32(rhs_mat_014589CD_2, 221); //B0(20-23) B1(20-23) B0(20-23) B1(20-23) B4(20-23) B5(20-23) B4(20-23) B5(20-23) B8(20-23) B9(20-23) B8(20-23) B9(20-23) BC(20-23) BD(20-23) BC(20-23) BD(20-23)
+                const __m512i rhs_mat_2367ABEF_2_sp2 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_2, 221); //B2(20-23) B3(20-23) B2(20-23) B3(20-23) B6(20-23) B7(20-23) B6(20-23) B7(20-23) BA(20-23) BB(20-23) BA(20-23) BB(20-23) BE(20-23) BF(20-23) BE(20-23) BF(20-23)
+
+                const __m512i rhs_mat_014589CD_3_sp2 = _mm512_shuffle_epi32(rhs_mat_014589CD_3, 221); //B0(28-31) B1(28-31) B0(28-31) B1(28-31) B4(28-31) B5(28-31) B4(28-31) B5(28-31) B8(28-31) B9(28-31) B8(28-31) B9(28-31) BC(28-31) BD(28-31) BC(28-31) BD(28-31)
+                const __m512i rhs_mat_2367ABEF_3_sp2 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_3, 221); //B2(28-31) B3(28-31) B2(28-31) B3(28-31) B6(28-31) B7(28-31) B6(28-31) B7(28-31) BA(28-31) BB(28-31) BA(28-31) BB(28-31) BE(28-31) BF(28-31) BE(28-31) BF(28-31)
+
+                // Scale values - Load the weight scale values of two block_q4_0x8
+                const __m512 col_scale_f32 = GGML_F32Cx8x2_LOAD(b_ptr_0[b].d, b_ptr_1[b].d);
+
+                // Process LHS in pairs of rows
+                for (int rp = 0; rp < 4; rp++) {
+
+                    // Load the four block_q4_0 quantized values interleaved with each other in chunks of eight - A0,A1,A2,A3
+                    // Loaded as set of 128 bit vectors and repeated and stored into a 256 bit vector before again repeating into 512 bit vector
+                    __m256i lhs_mat_ymm_0123_0 = _mm256_loadu_si256((const __m256i *)((a_ptrs[rp][b].qs)));
+                    __m256i lhs_mat_ymm_01_0 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_0, lhs_mat_ymm_0123_0, 0);
+                    __m256i lhs_mat_ymm_23_0 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_0, lhs_mat_ymm_0123_0, 17);
+                    __m256i lhs_mat_ymm_0123_1 = _mm256_loadu_si256((const __m256i *)((a_ptrs[rp][b].qs + 32)));
+                    __m256i lhs_mat_ymm_01_1 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_1, lhs_mat_ymm_0123_1, 0);
+                    __m256i lhs_mat_ymm_23_1 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_1, lhs_mat_ymm_0123_1, 17);
+                    __m256i lhs_mat_ymm_0123_2 = _mm256_loadu_si256((const __m256i *)((a_ptrs[rp][b].qs + 64)));
+                    __m256i lhs_mat_ymm_01_2 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_2, lhs_mat_ymm_0123_2, 0);
+                    __m256i lhs_mat_ymm_23_2 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_2, lhs_mat_ymm_0123_2, 17);
+                    __m256i lhs_mat_ymm_0123_3 = _mm256_loadu_si256((const __m256i *)((a_ptrs[rp][b].qs + 96)));
+                    __m256i lhs_mat_ymm_01_3 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_3, lhs_mat_ymm_0123_3, 0);
+                    __m256i lhs_mat_ymm_23_3 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_3, lhs_mat_ymm_0123_3, 17);
+
+                    __m512i lhs_mat_01_0 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_01_0), lhs_mat_ymm_01_0, 1);
+                    __m512i lhs_mat_23_0 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_23_0), lhs_mat_ymm_23_0, 1);
+                    __m512i lhs_mat_01_1 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_01_1), lhs_mat_ymm_01_1, 1);
+                    __m512i lhs_mat_23_1 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_23_1), lhs_mat_ymm_23_1, 1);
+                    __m512i lhs_mat_01_2 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_01_2), lhs_mat_ymm_01_2, 1);
+                    __m512i lhs_mat_23_2 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_23_2), lhs_mat_ymm_23_2, 1);
+                    __m512i lhs_mat_01_3 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_01_3), lhs_mat_ymm_01_3, 1);
+                    __m512i lhs_mat_23_3 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_23_3), lhs_mat_ymm_23_3, 1);
+
+                    // Shuffle pattern one - left side input
+
+                    const __m512i lhs_mat_01_0_sp1 = _mm512_shuffle_epi32(lhs_mat_01_0, 160);  //A0(0-3) A0(0-3) A1(0-3) A1(0-3) A0(0-3) A0(0-3) A1(0-3) A1(0-3) A0(0-3) A0(0-3) A1(0-3) A1(0-3) A0(0-3) A0(0-3) A1(0-3) A1(0-3)
+                    const __m512i lhs_mat_23_0_sp1 = _mm512_shuffle_epi32(lhs_mat_23_0, 160);  //A2(0-3) A2(0-3) A3(0-3) A3(0-3) A2(0-3) A2(0-3) A3(0-3) A3(0-3) A2(0-3) A2(0-3) A3(0-3) A3(0-3) A2(0-3) A2(0-3) A3(0-3) A3(0-3)
+
+                    const __m512i lhs_mat_01_1_sp1 = _mm512_shuffle_epi32(lhs_mat_01_1, 160);  //A0(8-11) A0(8-11) A1(8-11) A1(8-11) A0(8-11) A0(8-11) A1(8-11) A1(8-11) A0(8-11) A0(8-11) A1(8-11) A1(8-11) A0(8-11) A0(8-11) A1(8-11) A1(8-11)
+                    const __m512i lhs_mat_23_1_sp1 = _mm512_shuffle_epi32(lhs_mat_23_1, 160);  //A2(8-11) A2(8-11) A3(8-11) A3(8-11) A2(8-11) A2(8-11) A3(8-11) A3(8-11) A2(8-11) A2(8-11) A3(8-11) A3(8-11) A2(8-11) A2(8-11) A3(8-11) A3(8-11)
+
+                    const __m512i lhs_mat_01_2_sp1 = _mm512_shuffle_epi32(lhs_mat_01_2, 160);  //A0(16-19) A0(16-19) A1(16-19) A1(16-19) A0(16-19) A0(16-19) A1(16-19) A1(16-19) A0(16-19) A0(16-19) A1(16-19) A1(16-19) A0(16-19) A0(16-19) A1(16-19) A1(16-19)
+                    const __m512i lhs_mat_23_2_sp1 = _mm512_shuffle_epi32(lhs_mat_23_2, 160);  //A2(16-19) A2(16-19) A3(16-19) A3(16-19) A2(16-19) A2(16-19) A3(16-19) A3(16-19) A2(16-19) A2(16-19) A3(16-19) A3(16-19) A2(16-19) A2(16-19) A3(16-19) A3(16-19)
+
+                    const __m512i lhs_mat_01_3_sp1 = _mm512_shuffle_epi32(lhs_mat_01_3, 160);  //A0(24-27) A0(24-27) A1(24-27) A1(24-27) A0(24-27) A0(24-27) A1(24-27) A1(24-27) A0(24-27) A0(24-27) A1(24-27) A1(24-27) A0(24-27) A0(24-27) A1(24-27) A1(24-27)
+                    const __m512i lhs_mat_23_3_sp1 = _mm512_shuffle_epi32(lhs_mat_23_3, 160);  //A2(24-27) A2(24-27) A3(24-27) A3(24-27) A2(24-27) A2(24-27) A3(24-27) A3(24-27) A2(24-27) A2(24-27) A3(24-27) A3(24-27) A2(24-27) A2(24-27) A3(24-27) A3(24-27)
+
+                    // Shuffle pattern two - left side input
+
+                    const __m512i lhs_mat_01_0_sp2 = _mm512_shuffle_epi32(lhs_mat_01_0, 245);  //A0(4-7) A0(4-7) A1(4-7) A1(4-7) A0(4-7) A0(4-7) A1(4-7) A1(4-7) A0(4-7) A0(4-7) A1(4-7) A1(4-7) A0(4-7) A0(4-7) A1(4-7) A1(4-7)
+                    const __m512i lhs_mat_23_0_sp2 = _mm512_shuffle_epi32(lhs_mat_23_0, 245);  //A2(4-7) A2(4-7) A3(4-7) A3(4-7) A2(4-7) A2(4-7) A3(4-7) A3(4-7) A2(4-7) A2(4-7) A3(4-7) A3(4-7) A2(4-7) A2(4-7) A3(4-7) A3(4-7)
+
+                    const __m512i lhs_mat_01_1_sp2 = _mm512_shuffle_epi32(lhs_mat_01_1, 245);  //A0(12-15) A0(12-15) A1(12-15) A1(12-15) A0(12-15) A0(12-15) A1(12-15) A1(12-15) A0(12-15) A0(12-15) A1(12-15) A1(12-15) A0(12-15) A0(12-15) A1(12-15) A1(12-15)
+                    const __m512i lhs_mat_23_1_sp2 = _mm512_shuffle_epi32(lhs_mat_23_1, 245);  //A2(12-15) A2(12-15) A3(12-15) A3(12-15) A2(12-15) A2(12-15) A3(12-15) A3(12-15) A2(12-15) A2(12-15) A3(12-15) A3(12-15) A2(12-15) A2(12-15) A3(12-15) A3(12-15)
+
+                    const __m512i lhs_mat_01_2_sp2 = _mm512_shuffle_epi32(lhs_mat_01_2, 245);  //A0(20-23) A0(20-23) A1(20-23) A1(20-23) A0(20-23) A0(20-23) A1(20-23) A1(20-23) A0(20-23) A0(20-23) A1(20-23) A1(20-23) A0(20-23) A0(20-23) A1(20-23) A1(20-23)
+                    const __m512i lhs_mat_23_2_sp2 = _mm512_shuffle_epi32(lhs_mat_23_2, 245);  //A2(20-23) A2(20-23) A3(20-23) A3(20-23) A2(20-23) A2(20-23) A3(20-23) A3(20-23) A2(20-23) A2(20-23) A3(20-23) A3(20-23) A2(20-23) A2(20-23) A3(20-23) A3(20-23)
+
+                    const __m512i lhs_mat_01_3_sp2 = _mm512_shuffle_epi32(lhs_mat_01_3, 245);  //A0(28-31) A0(28-31) A1(28-31) A1(28-31) A0(28-31) A0(28-31) A1(28-31) A1(28-31) A0(28-31) A0(28-31) A1(28-31) A1(28-31) A0(28-31) A0(28-31) A1(28-31) A1(28-31)
+                    const __m512i lhs_mat_23_3_sp2 = _mm512_shuffle_epi32(lhs_mat_23_3, 245);  //A2(28-31) A2(28-31) A3(28-31) A3(28-31) A2(28-31) A2(28-31) A3(28-31) A3(28-31) A2(28-31) A2(28-31) A3(28-31) A3(28-31) A2(28-31) A2(28-31) A3(28-31) A3(28-31)
+
+                    // The values arranged in shuffle patterns are operated with dot product operation within 32 bit lane i.e corresponding bytes and multiplied and added into 32 bit integers within 32 bit lane
+                    // Resembles MMLAs into 2x2 matrices in ARM Version
+                    __m512i iacc_mat_00_sp1 =
+                        _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_01_3_sp1, rhs_mat_014589CD_3_sp1), mul_sum_i8_pairs_int32x16(lhs_mat_01_2_sp1, rhs_mat_014589CD_2_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_01_1_sp1, rhs_mat_014589CD_1_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_01_0_sp1, rhs_mat_014589CD_0_sp1));
+                    __m512i iacc_mat_01_sp1 =
+                        _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_01_3_sp1, rhs_mat_2367ABEF_3_sp1), mul_sum_i8_pairs_int32x16(lhs_mat_01_2_sp1, rhs_mat_2367ABEF_2_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_01_1_sp1, rhs_mat_2367ABEF_1_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_01_0_sp1, rhs_mat_2367ABEF_0_sp1));
+                    __m512i iacc_mat_10_sp1 =
+                        _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_23_3_sp1, rhs_mat_014589CD_3_sp1), mul_sum_i8_pairs_int32x16(lhs_mat_23_2_sp1, rhs_mat_014589CD_2_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_23_1_sp1, rhs_mat_014589CD_1_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_23_0_sp1, rhs_mat_014589CD_0_sp1));
+                    __m512i iacc_mat_11_sp1 =
+                        _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_23_3_sp1, rhs_mat_2367ABEF_3_sp1), mul_sum_i8_pairs_int32x16(lhs_mat_23_2_sp1, rhs_mat_2367ABEF_2_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_23_1_sp1, rhs_mat_2367ABEF_1_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_23_0_sp1, rhs_mat_2367ABEF_0_sp1));
+                    __m512i iacc_mat_00_sp2 =
+                        _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_01_3_sp2, rhs_mat_014589CD_3_sp2), mul_sum_i8_pairs_int32x16(lhs_mat_01_2_sp2, rhs_mat_014589CD_2_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_01_1_sp2, rhs_mat_014589CD_1_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_01_0_sp2, rhs_mat_014589CD_0_sp2));
+                    __m512i iacc_mat_01_sp2 =
+                        _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_01_3_sp2, rhs_mat_2367ABEF_3_sp2), mul_sum_i8_pairs_int32x16(lhs_mat_01_2_sp2, rhs_mat_2367ABEF_2_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_01_1_sp2, rhs_mat_2367ABEF_1_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_01_0_sp2, rhs_mat_2367ABEF_0_sp2));
+                    __m512i iacc_mat_10_sp2 =
+                        _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_23_3_sp2, rhs_mat_014589CD_3_sp2), mul_sum_i8_pairs_int32x16(lhs_mat_23_2_sp2, rhs_mat_014589CD_2_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_23_1_sp2, rhs_mat_014589CD_1_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_23_0_sp2, rhs_mat_014589CD_0_sp2));
+                    __m512i iacc_mat_11_sp2 =
+                        _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_23_3_sp2, rhs_mat_2367ABEF_3_sp2), mul_sum_i8_pairs_int32x16(lhs_mat_23_2_sp2, rhs_mat_2367ABEF_2_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_23_1_sp2, rhs_mat_2367ABEF_1_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_23_0_sp2, rhs_mat_2367ABEF_0_sp2));
+
+                    // Output of both shuffle patterns are added in order to sum dot product outputs of all 32 values in block
+                    __m512i iacc_mat_00 = _mm512_add_epi32(iacc_mat_00_sp1, iacc_mat_00_sp2);
+                    __m512i iacc_mat_01 = _mm512_add_epi32(iacc_mat_01_sp1, iacc_mat_01_sp2);
+                    __m512i iacc_mat_10 = _mm512_add_epi32(iacc_mat_10_sp1, iacc_mat_10_sp2);
+                    __m512i iacc_mat_11 = _mm512_add_epi32(iacc_mat_11_sp1, iacc_mat_11_sp2);
+
+
+                    // Straighten out to make 4 row vectors
+                    __m512i iacc_row_0 = _mm512_mask_blend_epi32(0xCCCC, iacc_mat_00, _mm512_shuffle_epi32(iacc_mat_01, 78));
+                    __m512i iacc_row_1 = _mm512_mask_blend_epi32(0xCCCC, _mm512_shuffle_epi32(iacc_mat_00, 78), iacc_mat_01);
+                    __m512i iacc_row_2 = _mm512_mask_blend_epi32(0xCCCC, iacc_mat_10, _mm512_shuffle_epi32(iacc_mat_11, 78));
+                    __m512i iacc_row_3 = _mm512_mask_blend_epi32(0xCCCC, _mm512_shuffle_epi32(iacc_mat_10, 78), iacc_mat_11);
+
+                    // Load the scale(d) values for all the 4 Q8_0 blocks and repeat it across lanes
+                    const __m128i row_scale_f16 = _mm_shuffle_epi32(_mm_maskload_epi32((int const*)(a_ptrs[rp][b].d), loadMask), 68);
+                    const __m512 row_scale_f32 = GGML_F32Cx16_REPEAT_LOAD(row_scale_f16);
+
+                    // Multiply with appropiate scales and accumulate
+                    acc_rows[rp * 4]     = _mm512_fmadd_ps(_mm512_cvtepi32_ps(iacc_row_0), _mm512_mul_ps(col_scale_f32, _mm512_shuffle_ps(row_scale_f32, row_scale_f32, 0)),   acc_rows[rp * 4]);
+                    acc_rows[rp * 4 + 1] = _mm512_fmadd_ps(_mm512_cvtepi32_ps(iacc_row_1), _mm512_mul_ps(col_scale_f32, _mm512_shuffle_ps(row_scale_f32, row_scale_f32, 85)),  acc_rows[rp * 4 + 1]);
+                    acc_rows[rp * 4 + 2] = _mm512_fmadd_ps(_mm512_cvtepi32_ps(iacc_row_2), _mm512_mul_ps(col_scale_f32, _mm512_shuffle_ps(row_scale_f32, row_scale_f32, 170)), acc_rows[rp * 4 + 2]);
+                    acc_rows[rp * 4 + 3] = _mm512_fmadd_ps(_mm512_cvtepi32_ps(iacc_row_3), _mm512_mul_ps(col_scale_f32, _mm512_shuffle_ps(row_scale_f32, row_scale_f32, 255)), acc_rows[rp * 4 + 3]);
+                }
+            }
+
+            // Store the accumulated values
+            for (int i = 0; i < 16; i++) {
+                _mm512_storeu_ps((float *)(s + ((y * 4 + i) * bs + x * 8)), acc_rows[i]);
+            }
+        }
+    }
+    // Take a block_q8_0x4 structures at each pass of the loop and perform dot product operation
+    for (; y < nr / 4; y ++) {
+
+        const block_q8_0x4 * a_ptr = a_ptr_start + (y * nb);
+
+        // Take group of two block_q4_0x8 structures at each pass of the loop and perform dot product operation
+        for (int64_t x = 0; x < anc / 8; x += 2) {
+
+            const block_q4_0x8 * b_ptr_0 = b_ptr_start + ((x)     * b_nb);
+            const block_q4_0x8 * b_ptr_1 = b_ptr_start + ((x + 1) * b_nb);
+
+            // Master FP accumulators
+            __m512 acc_rows[4];
+            for (int i = 0; i < 4; i++) {
+                acc_rows[i] = _mm512_setzero_ps();
+            }
+
+            for (int64_t b = 0; b < nb; b++) {
+                // Load the sixteen block_q4_0 quantized values interleaved with each other in chunks of eight - B0,B1 ....BE,BF
+                const __m256i rhs_raw_mat_0123_0 = _mm256_loadu_si256((const __m256i *)(b_ptr_0[b].qs));
+                const __m256i rhs_raw_mat_4567_0 = _mm256_loadu_si256((const __m256i *)(b_ptr_0[b].qs + 32));
+                const __m256i rhs_raw_mat_0123_1 = _mm256_loadu_si256((const __m256i *)(b_ptr_0[b].qs + 64));
+                const __m256i rhs_raw_mat_4567_1 = _mm256_loadu_si256((const __m256i *)(b_ptr_0[b].qs + 96));
+
+                const __m256i rhs_raw_mat_89AB_0 = _mm256_loadu_si256((const __m256i *)(b_ptr_1[b].qs));
+                const __m256i rhs_raw_mat_CDEF_0 = _mm256_loadu_si256((const __m256i *)(b_ptr_1[b].qs + 32));
+                const __m256i rhs_raw_mat_89AB_1 = _mm256_loadu_si256((const __m256i *)(b_ptr_1[b].qs + 64));
+                const __m256i rhs_raw_mat_CDEF_1 = _mm256_loadu_si256((const __m256i *)(b_ptr_1[b].qs + 96));
+
+                // Save the values in the following vectors in the formats B0B1B4B5, B2B3B6B7 for further processing and storing of valuess
+                const __m256i rhs_raw_mat_0145_0 = _mm256_blend_epi32(rhs_raw_mat_0123_0, _mm256_permutevar8x32_epi32(rhs_raw_mat_4567_0, requiredOrder), 240);
+                const __m256i rhs_raw_mat_2367_0 = _mm256_blend_epi32(_mm256_permutevar8x32_epi32(rhs_raw_mat_0123_0, requiredOrder), rhs_raw_mat_4567_0, 240);
+                const __m256i rhs_raw_mat_0145_1 = _mm256_blend_epi32(rhs_raw_mat_0123_1, _mm256_permutevar8x32_epi32(rhs_raw_mat_4567_1, requiredOrder), 240);
+                const __m256i rhs_raw_mat_2367_1 = _mm256_blend_epi32(_mm256_permutevar8x32_epi32(rhs_raw_mat_0123_1, requiredOrder), rhs_raw_mat_4567_1, 240);
+
+                const __m256i rhs_raw_mat_89CD_0 = _mm256_blend_epi32(rhs_raw_mat_89AB_0, _mm256_permutevar8x32_epi32(rhs_raw_mat_CDEF_0, requiredOrder), 240);
+                const __m256i rhs_raw_mat_ABEF_0 = _mm256_blend_epi32(_mm256_permutevar8x32_epi32(rhs_raw_mat_89AB_0, requiredOrder), rhs_raw_mat_CDEF_0, 240);
+                const __m256i rhs_raw_mat_89CD_1 = _mm256_blend_epi32(rhs_raw_mat_89AB_1, _mm256_permutevar8x32_epi32(rhs_raw_mat_CDEF_1, requiredOrder), 240);
+                const __m256i rhs_raw_mat_ABEF_1 = _mm256_blend_epi32(_mm256_permutevar8x32_epi32(rhs_raw_mat_89AB_1, requiredOrder), rhs_raw_mat_CDEF_1, 240);
+
+                const __m512i rhs_raw_mat_014589CD_0 = _mm512_inserti32x8(_mm512_castsi256_si512(rhs_raw_mat_0145_0), rhs_raw_mat_89CD_0, 1);
+                const __m512i rhs_raw_mat_2367ABEF_0 = _mm512_inserti32x8(_mm512_castsi256_si512(rhs_raw_mat_2367_0), rhs_raw_mat_ABEF_0, 1);
+                const __m512i rhs_raw_mat_014589CD_1 = _mm512_inserti32x8(_mm512_castsi256_si512(rhs_raw_mat_0145_1), rhs_raw_mat_89CD_1, 1);
+                const __m512i rhs_raw_mat_2367ABEF_1 = _mm512_inserti32x8(_mm512_castsi256_si512(rhs_raw_mat_2367_1), rhs_raw_mat_ABEF_1, 1);
+
+                // 4-bit -> 8-bit - Sign is maintained
+                const __m512i rhs_mat_014589CD_0 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(rhs_raw_mat_014589CD_0, m4bexpanded)); //B0(0-7) B1(0-7) B4(0-7) B5(0-7) B8(0-7) B9(0-7) BC(0-7) BD(0-7)
+                const __m512i rhs_mat_2367ABEF_0 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(rhs_raw_mat_2367ABEF_0, m4bexpanded)); //B2(0-7) B3(0-7) B6(0-7) B7(0-7) BA(0-7) BB(0-7) BE(0-7) BF(0-7)
+
+                const __m512i rhs_mat_014589CD_1 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(rhs_raw_mat_014589CD_1, m4bexpanded)); //B0(8-15) B1(8-15) B4(8-15) B5(8-15) B8(8-15) B9(8-15) BC(8-15) BD(8-15)
+                const __m512i rhs_mat_2367ABEF_1 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(rhs_raw_mat_2367ABEF_1, m4bexpanded)); //B2(8-15) B3(8-15) B6(8-15) B7(8-15) BA(8-15) BB(8-15) BE(8-15) BF(8-15)
+
+                const __m512i rhs_mat_014589CD_2 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(_mm512_srli_epi16(rhs_raw_mat_014589CD_0, 4), m4bexpanded)); //B0(16-23) B1(16-23) B4(16-23) B5(16-23) B8(16-23) B9(16-23) BC(16-23) BD(16-23)
+                const __m512i rhs_mat_2367ABEF_2 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(_mm512_srli_epi16(rhs_raw_mat_2367ABEF_0, 4), m4bexpanded)); //B2(16-23) B3(16-23) B6(16-23) B7(16-23) BA(16-23) BB(16-23) BE(16-23) BF(16-23)
+
+                const __m512i rhs_mat_014589CD_3 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(_mm512_srli_epi16(rhs_raw_mat_014589CD_1, 4), m4bexpanded)); //B0(24-31) B1(24-31) B4(24-31) B5(24-31) B8(24-31) B9(24-31) BC(24-31) BD(24-31)
+                const __m512i rhs_mat_2367ABEF_3 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(_mm512_srli_epi16(rhs_raw_mat_2367ABEF_1, 4), m4bexpanded)); //B2(24-31) B3(24-31) B6(24-31) B7(24-31) BA(24-31) BB(24-31) BE(24-31) BF(24-31)
+
+                // Shuffle pattern one - right side input
+                const __m512i rhs_mat_014589CD_0_sp1 = _mm512_shuffle_epi32(rhs_mat_014589CD_0, 136); //B0(0-3) B1(0-3) B0(0-3) B1(0-3) B4(0-3) B5(0-3) B4(0-3) B5(0-3) B8(0-3) B9(0-3) B8(0-3) B9(0-3) BC(0-3) BD(0-3) BC(0-3) BD(0-3)
+                const __m512i rhs_mat_2367ABEF_0_sp1 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_0, 136); //B2(0-3) B3(0-3) B2(0-3) B3(0-3) B6(0-3) B7(0-3) B6(0-3) B7(0-3) BA(0-3) BB(0-3) BA(0-3) BB(0-3) BE(0-3) BF(0-3) BE(0-3) BF(0-3)
+
+                const __m512i rhs_mat_014589CD_1_sp1 = _mm512_shuffle_epi32(rhs_mat_014589CD_1, 136); //B0(8-11) B1(8-11) B0(8-11) B1(8-11) B4(8-11) B5(8-11) B4(8-11) B5(8-11) B8(8-11) B9(8-11) B8(8-11) B9(8-11) BC(8-11) BD(8-11) BC(8-11) BD(8-11)
+                const __m512i rhs_mat_2367ABEF_1_sp1 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_1, 136); //B2(8-11) B3(8-11) B2(8-11) B3(8-11) B6(8-11) B7(8-11) B6(8-11) B7(8-11) BA(8-11) BB(8-11) BA(8-11) BB(8-11) BE(8-11) BF(8-11) BE(8-11) BF(8-11)
+
+                const __m512i rhs_mat_014589CD_2_sp1 = _mm512_shuffle_epi32(rhs_mat_014589CD_2, 136); //B0(16-19) B1(16-19) B0(16-19) B1(16-19) B4(16-19) B5(16-19) B4(16-19) B5(16-19) B8(16-19) B9(16-19) B8(16-19) B9(16-19) BC(16-19) BD(16-19) BC(16-19) BD(16-19)
+                const __m512i rhs_mat_2367ABEF_2_sp1 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_2, 136); //B2(16-19) B3(16-19) B2(16-19) B3(16-19) B6(16-19) B7(16-19) B6(16-19) B7(16-19) BA(16-19) BB(16-19) BA(16-19) BB(16-19) BE(16-19) BF(16-19) BE(16-19) BF(16-19)
+
+                const __m512i rhs_mat_014589CD_3_sp1 = _mm512_shuffle_epi32(rhs_mat_014589CD_3, 136); //B0(24-27) B1(24-27) B0(24-27) B1(24-27) B4(24-27) B5(24-27) B4(24-27) B5(24-27) B8(24-27) B9(24-27) B8(24-27) B9(24-27) BC(24-27) BD(24-27) BC(24-27) BD(24-27)
+                const __m512i rhs_mat_2367ABEF_3_sp1 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_3, 136); //B2(24-27) B3(24-27) B2(24-27) B3(24-27) B6(24-27) B7(24-27) B6(24-27) B7(24-27) BA(24-27) BB(24-27) BA(24-27) BB(24-27) BE(24-27) BF(24-27) BE(24-27) BF(24-27)
+
+                // Shuffle pattern two - right side input
+
+                const __m512i rhs_mat_014589CD_0_sp2 = _mm512_shuffle_epi32(rhs_mat_014589CD_0, 221); //B0(4-7) B1(4-7) B0(4-7) B1(4-7) B4(4-7) B5(4-7) B4(4-7) B5(4-7) B8(4-7) B9(4-7) B8(4-7) B9(4-7) BC(4-7) BD(4-7) BC(4-7) BD(4-7)
+                const __m512i rhs_mat_2367ABEF_0_sp2 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_0, 221); //B2(4-7) B3(4-7) B2(4-7) B3(4-7) B6(4-7) B7(4-7) B6(4-7) B7(4-7) BA(4-7) BB(4-7) BA(4-7) BB(4-7) BE(4-7) BF(4-7) BE(4-7) BF(4-7)
+
+                const __m512i rhs_mat_014589CD_1_sp2 = _mm512_shuffle_epi32(rhs_mat_014589CD_1, 221); //B0(12-15) B1(12-15) B0(12-15) B1(12-15) B4(12-15) B5(12-15) B4(12-15) B5(12-15) B8(12-15) B9(12-15) B8(12-15) B9(12-15) BC(12-15) BD(12-15) BC(12-15) BD(12-15)
+                const __m512i rhs_mat_2367ABEF_1_sp2 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_1, 221); //B2(12-15) B3(12-15) B2(12-15) B3(12-15) B6(12-15) B7(12-15) B6(12-15) B7(12-15) BA(12-15) BB(12-15) BA(12-15) BB(12-15) BE(12-15) BF(12-15) BE(12-15) BF(12-15)
+
+                const __m512i rhs_mat_014589CD_2_sp2 = _mm512_shuffle_epi32(rhs_mat_014589CD_2, 221); //B0(20-23) B1(20-23) B0(20-23) B1(20-23) B4(20-23) B5(20-23) B4(20-23) B5(20-23) B8(20-23) B9(20-23) B8(20-23) B9(20-23) BC(20-23) BD(20-23) BC(20-23) BD(20-23)
+                const __m512i rhs_mat_2367ABEF_2_sp2 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_2, 221); //B2(20-23) B3(20-23) B2(20-23) B3(20-23) B6(20-23) B7(20-23) B6(20-23) B7(20-23) BA(20-23) BB(20-23) BA(20-23) BB(20-23) BE(20-23) BF(20-23) BE(20-23) BF(20-23)
+
+                const __m512i rhs_mat_014589CD_3_sp2 = _mm512_shuffle_epi32(rhs_mat_014589CD_3, 221); //B0(28-31) B1(28-31) B0(28-31) B1(28-31) B4(28-31) B5(28-31) B4(28-31) B5(28-31) B8(28-31) B9(28-31) B8(28-31) B9(28-31) BC(28-31) BD(28-31) BC(28-31) BD(28-31)
+                const __m512i rhs_mat_2367ABEF_3_sp2 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_3, 221); //B2(28-31) B3(28-31) B2(28-31) B3(28-31) B6(28-31) B7(28-31) B6(28-31) B7(28-31) BA(28-31) BB(28-31) BA(28-31) BB(28-31) BE(28-31) BF(28-31) BE(28-31) BF(28-31)
+
+
+                // Scale values - Load the weight scale values of two block_q4_0x8
+                const __m512 col_scale_f32 = GGML_F32Cx8x2_LOAD(b_ptr_0[b].d, b_ptr_1[b].d);
+
+                // Load the four block_q4_0 quantized values interleaved with each other in chunks of eight - A0,A1,A2,A3
+                // Loaded as set of 128 bit vectors and repeated and stored into a 256 bit vector before again repeating into 512 bit vector
+                __m256i lhs_mat_ymm_0123_0 = _mm256_loadu_si256((const __m256i *)((a_ptr[b].qs)));
+                __m256i lhs_mat_ymm_01_0 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_0, lhs_mat_ymm_0123_0, 0);
+                __m256i lhs_mat_ymm_23_0 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_0, lhs_mat_ymm_0123_0, 17);
+                __m256i lhs_mat_ymm_0123_1 = _mm256_loadu_si256((const __m256i *)((a_ptr[b].qs + 32)));
+                __m256i lhs_mat_ymm_01_1 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_1, lhs_mat_ymm_0123_1, 0);
+                __m256i lhs_mat_ymm_23_1 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_1, lhs_mat_ymm_0123_1, 17);
+                __m256i lhs_mat_ymm_0123_2 = _mm256_loadu_si256((const __m256i *)((a_ptr[b].qs + 64)));
+                __m256i lhs_mat_ymm_01_2 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_2, lhs_mat_ymm_0123_2, 0);
+                __m256i lhs_mat_ymm_23_2 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_2, lhs_mat_ymm_0123_2, 17);
+                __m256i lhs_mat_ymm_0123_3 = _mm256_loadu_si256((const __m256i *)((a_ptr[b].qs + 96)));
+                __m256i lhs_mat_ymm_01_3 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_3, lhs_mat_ymm_0123_3, 0);
+                __m256i lhs_mat_ymm_23_3 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_3, lhs_mat_ymm_0123_3, 17);
+
+                __m512i lhs_mat_01_0 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_01_0), lhs_mat_ymm_01_0, 1);
+                __m512i lhs_mat_23_0 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_23_0), lhs_mat_ymm_23_0, 1);
+                __m512i lhs_mat_01_1 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_01_1), lhs_mat_ymm_01_1, 1);
+                __m512i lhs_mat_23_1 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_23_1), lhs_mat_ymm_23_1, 1);
+                __m512i lhs_mat_01_2 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_01_2), lhs_mat_ymm_01_2, 1);
+                __m512i lhs_mat_23_2 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_23_2), lhs_mat_ymm_23_2, 1);
+                __m512i lhs_mat_01_3 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_01_3), lhs_mat_ymm_01_3, 1);
+                __m512i lhs_mat_23_3 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_23_3), lhs_mat_ymm_23_3, 1);
+
+                // Shuffle pattern one - left side input
+
+                const __m512i lhs_mat_01_0_sp1 = _mm512_shuffle_epi32(lhs_mat_01_0, 160);  //A0(0-3) A0(0-3) A1(0-3) A1(0-3) A0(0-3) A0(0-3) A1(0-3) A1(0-3) A0(0-3) A0(0-3) A1(0-3) A1(0-3) A0(0-3) A0(0-3) A1(0-3) A1(0-3)
+                const __m512i lhs_mat_23_0_sp1 = _mm512_shuffle_epi32(lhs_mat_23_0, 160);  //A2(0-3) A2(0-3) A3(0-3) A3(0-3) A2(0-3) A2(0-3) A3(0-3) A3(0-3) A2(0-3) A2(0-3) A3(0-3) A3(0-3) A2(0-3) A2(0-3) A3(0-3) A3(0-3)
+
+                const __m512i lhs_mat_01_1_sp1 = _mm512_shuffle_epi32(lhs_mat_01_1, 160);  //A0(8-11) A0(8-11) A1(8-11) A1(8-11) A0(8-11) A0(8-11) A1(8-11) A1(8-11) A0(8-11) A0(8-11) A1(8-11) A1(8-11) A0(8-11) A0(8-11) A1(8-11) A1(8-11)
+                const __m512i lhs_mat_23_1_sp1 = _mm512_shuffle_epi32(lhs_mat_23_1, 160);  //A2(8-11) A2(8-11) A3(8-11) A3(8-11) A2(8-11) A2(8-11) A3(8-11) A3(8-11) A2(8-11) A2(8-11) A3(8-11) A3(8-11) A2(8-11) A2(8-11) A3(8-11) A3(8-11)
+
+                const __m512i lhs_mat_01_2_sp1 = _mm512_shuffle_epi32(lhs_mat_01_2, 160);  //A0(16-19) A0(16-19) A1(16-19) A1(16-19) A0(16-19) A0(16-19) A1(16-19) A1(16-19) A0(16-19) A0(16-19) A1(16-19) A1(16-19) A0(16-19) A0(16-19) A1(16-19) A1(16-19)
+                const __m512i lhs_mat_23_2_sp1 = _mm512_shuffle_epi32(lhs_mat_23_2, 160);  //A2(16-19) A2(16-19) A3(16-19) A3(16-19) A2(16-19) A2(16-19) A3(16-19) A3(16-19) A2(16-19) A2(16-19) A3(16-19) A3(16-19) A2(16-19) A2(16-19) A3(16-19) A3(16-19)
+
+                const __m512i lhs_mat_01_3_sp1 = _mm512_shuffle_epi32(lhs_mat_01_3, 160);  //A0(24-27) A0(24-27) A1(24-27) A1(24-27) A0(24-27) A0(24-27) A1(24-27) A1(24-27) A0(24-27) A0(24-27) A1(24-27) A1(24-27) A0(24-27) A0(24-27) A1(24-27) A1(24-27)
+                const __m512i lhs_mat_23_3_sp1 = _mm512_shuffle_epi32(lhs_mat_23_3, 160);  //A2(24-27) A2(24-27) A3(24-27) A3(24-27) A2(24-27) A2(24-27) A3(24-27) A3(24-27) A2(24-27) A2(24-27) A3(24-27) A3(24-27) A2(24-27) A2(24-27) A3(24-27) A3(24-27)
+
+                // Shuffle pattern two - left side input
+
+                const __m512i lhs_mat_01_0_sp2 = _mm512_shuffle_epi32(lhs_mat_01_0, 245);  //A0(4-7) A0(4-7) A1(4-7) A1(4-7) A0(4-7) A0(4-7) A1(4-7) A1(4-7) A0(4-7) A0(4-7) A1(4-7) A1(4-7) A0(4-7) A0(4-7) A1(4-7) A1(4-7)
+                const __m512i lhs_mat_23_0_sp2 = _mm512_shuffle_epi32(lhs_mat_23_0, 245);  //A2(4-7) A2(4-7) A3(4-7) A3(4-7) A2(4-7) A2(4-7) A3(4-7) A3(4-7) A2(4-7) A2(4-7) A3(4-7) A3(4-7) A2(4-7) A2(4-7) A3(4-7) A3(4-7)
+
+                const __m512i lhs_mat_01_1_sp2 = _mm512_shuffle_epi32(lhs_mat_01_1, 245);  //A0(12-15) A0(12-15) A1(12-15) A1(12-15) A0(12-15) A0(12-15) A1(12-15) A1(12-15) A0(12-15) A0(12-15) A1(12-15) A1(12-15) A0(12-15) A0(12-15) A1(12-15) A1(12-15)
+                const __m512i lhs_mat_23_1_sp2 = _mm512_shuffle_epi32(lhs_mat_23_1, 245);  //A2(12-15) A2(12-15) A3(12-15) A3(12-15) A2(12-15) A2(12-15) A3(12-15) A3(12-15) A2(12-15) A2(12-15) A3(12-15) A3(12-15) A2(12-15) A2(12-15) A3(12-15) A3(12-15)
+
+                const __m512i lhs_mat_01_2_sp2 = _mm512_shuffle_epi32(lhs_mat_01_2, 245);  //A0(20-23) A0(20-23) A1(20-23) A1(20-23) A0(20-23) A0(20-23) A1(20-23) A1(20-23) A0(20-23) A0(20-23) A1(20-23) A1(20-23) A0(20-23) A0(20-23) A1(20-23) A1(20-23)
+                const __m512i lhs_mat_23_2_sp2 = _mm512_shuffle_epi32(lhs_mat_23_2, 245);  //A2(20-23) A2(20-23) A3(20-23) A3(20-23) A2(20-23) A2(20-23) A3(20-23) A3(20-23) A2(20-23) A2(20-23) A3(20-23) A3(20-23) A2(20-23) A2(20-23) A3(20-23) A3(20-23)
+
+                const __m512i lhs_mat_01_3_sp2 = _mm512_shuffle_epi32(lhs_mat_01_3, 245);  //A0(28-31) A0(28-31) A1(28-31) A1(28-31) A0(28-31) A0(28-31) A1(28-31) A1(28-31) A0(28-31) A0(28-31) A1(28-31) A1(28-31) A0(28-31) A0(28-31) A1(28-31) A1(28-31)
+                const __m512i lhs_mat_23_3_sp2 = _mm512_shuffle_epi32(lhs_mat_23_3, 245);  //A2(28-31) A2(28-31) A3(28-31) A3(28-31) A2(28-31) A2(28-31) A3(28-31) A3(28-31) A2(28-31) A2(28-31) A3(28-31) A3(28-31) A2(28-31) A2(28-31) A3(28-31) A3(28-31)
+
+                // The values arranged in shuffle patterns are operated with dot product operation within 32 bit lane i.e corresponding bytes and multiplied and added into 32 bit integers within 32 bit lane
+                // Resembles MMLAs into 2x2 matrices in ARM Version
+                __m512i iacc_mat_00_sp1 =
+                    _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_01_3_sp1, rhs_mat_014589CD_3_sp1), mul_sum_i8_pairs_int32x16(lhs_mat_01_2_sp1, rhs_mat_014589CD_2_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_01_1_sp1, rhs_mat_014589CD_1_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_01_0_sp1, rhs_mat_014589CD_0_sp1));
+                __m512i iacc_mat_01_sp1 =
+                    _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_01_3_sp1, rhs_mat_2367ABEF_3_sp1), mul_sum_i8_pairs_int32x16(lhs_mat_01_2_sp1, rhs_mat_2367ABEF_2_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_01_1_sp1, rhs_mat_2367ABEF_1_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_01_0_sp1, rhs_mat_2367ABEF_0_sp1));
+                __m512i iacc_mat_10_sp1 =
+                    _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_23_3_sp1, rhs_mat_014589CD_3_sp1), mul_sum_i8_pairs_int32x16(lhs_mat_23_2_sp1, rhs_mat_014589CD_2_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_23_1_sp1, rhs_mat_014589CD_1_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_23_0_sp1, rhs_mat_014589CD_0_sp1));
+                __m512i iacc_mat_11_sp1 =
+                    _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_23_3_sp1, rhs_mat_2367ABEF_3_sp1), mul_sum_i8_pairs_int32x16(lhs_mat_23_2_sp1, rhs_mat_2367ABEF_2_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_23_1_sp1, rhs_mat_2367ABEF_1_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_23_0_sp1, rhs_mat_2367ABEF_0_sp1));
+                __m512i iacc_mat_00_sp2 =
+                    _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_01_3_sp2, rhs_mat_014589CD_3_sp2), mul_sum_i8_pairs_int32x16(lhs_mat_01_2_sp2, rhs_mat_014589CD_2_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_01_1_sp2, rhs_mat_014589CD_1_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_01_0_sp2, rhs_mat_014589CD_0_sp2));
+                __m512i iacc_mat_01_sp2 =
+                    _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_01_3_sp2, rhs_mat_2367ABEF_3_sp2), mul_sum_i8_pairs_int32x16(lhs_mat_01_2_sp2, rhs_mat_2367ABEF_2_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_01_1_sp2, rhs_mat_2367ABEF_1_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_01_0_sp2, rhs_mat_2367ABEF_0_sp2));
+                __m512i iacc_mat_10_sp2 =
+                    _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_23_3_sp2, rhs_mat_014589CD_3_sp2), mul_sum_i8_pairs_int32x16(lhs_mat_23_2_sp2, rhs_mat_014589CD_2_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_23_1_sp2, rhs_mat_014589CD_1_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_23_0_sp2, rhs_mat_014589CD_0_sp2));
+                __m512i iacc_mat_11_sp2 =
+                    _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_23_3_sp2, rhs_mat_2367ABEF_3_sp2), mul_sum_i8_pairs_int32x16(lhs_mat_23_2_sp2, rhs_mat_2367ABEF_2_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_23_1_sp2, rhs_mat_2367ABEF_1_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_23_0_sp2, rhs_mat_2367ABEF_0_sp2));
+
+                // Output of both shuffle patterns are added in order to sum dot product outputs of all 32 values in block
+                __m512i iacc_mat_00 = _mm512_add_epi32(iacc_mat_00_sp1, iacc_mat_00_sp2);
+                __m512i iacc_mat_01 = _mm512_add_epi32(iacc_mat_01_sp1, iacc_mat_01_sp2);
+                __m512i iacc_mat_10 = _mm512_add_epi32(iacc_mat_10_sp1, iacc_mat_10_sp2);
+                __m512i iacc_mat_11 = _mm512_add_epi32(iacc_mat_11_sp1, iacc_mat_11_sp2);
+
+
+                // Straighten out to make 4 row vectors
+                __m512i iacc_row_0 = _mm512_mask_blend_epi32(0xCCCC, iacc_mat_00, _mm512_shuffle_epi32(iacc_mat_01, 78));
+                __m512i iacc_row_1 = _mm512_mask_blend_epi32(0xCCCC, _mm512_shuffle_epi32(iacc_mat_00, 78), iacc_mat_01);
+                __m512i iacc_row_2 = _mm512_mask_blend_epi32(0xCCCC, iacc_mat_10, _mm512_shuffle_epi32(iacc_mat_11, 78));
+                __m512i iacc_row_3 = _mm512_mask_blend_epi32(0xCCCC, _mm512_shuffle_epi32(iacc_mat_10, 78), iacc_mat_11);
+
+                // Load the scale(d) values for all the 4 Q8_0 blocks and repeat it across lanes
+                const __m128i row_scale_f16 = _mm_shuffle_epi32(_mm_maskload_epi32((int const*)(a_ptr[b].d), loadMask), 68);
+                const __m512 row_scale_f32 = GGML_F32Cx16_REPEAT_LOAD(row_scale_f16);
+
+                // Multiply with appropiate scales and accumulate
+                acc_rows[0] = _mm512_fmadd_ps(_mm512_cvtepi32_ps(iacc_row_0), _mm512_mul_ps(col_scale_f32, _mm512_shuffle_ps(row_scale_f32, row_scale_f32, 0)),   acc_rows[0]);
+                acc_rows[1] = _mm512_fmadd_ps(_mm512_cvtepi32_ps(iacc_row_1), _mm512_mul_ps(col_scale_f32, _mm512_shuffle_ps(row_scale_f32, row_scale_f32, 85)),  acc_rows[1]);
+                acc_rows[2] = _mm512_fmadd_ps(_mm512_cvtepi32_ps(iacc_row_2), _mm512_mul_ps(col_scale_f32, _mm512_shuffle_ps(row_scale_f32, row_scale_f32, 170)), acc_rows[2]);
+                acc_rows[3] = _mm512_fmadd_ps(_mm512_cvtepi32_ps(iacc_row_3), _mm512_mul_ps(col_scale_f32, _mm512_shuffle_ps(row_scale_f32, row_scale_f32, 255)), acc_rows[3]);
+            }
+
+            // Store the accumulated values
+            for (int i = 0; i < 4; i++) {
+                _mm512_storeu_ps((float *)(s + ((y * 4 + i) * bs + x * 8)), acc_rows[i]);
+            }
+        }
+    }
+    if (anc != nc) {
+        xstart = anc/8;
+        y = 0;
+    }
+#endif // __AVX512F__

    // Take group of four block_q8_0x4 structures at each pass of the loop and perform dot product operation
-    int anr = nr - nr %16; // Used to align nr with boundary of 16

    for (; y < anr / 4; y += 4) {
        const block_q8_0x4 * a_ptrs[4];
@@ -2435,7 +2904,7 @@ void ggml_gemm_q4_0_8x8_q8_0(int n, float * restrict s, size_t bs, const void *
        }

        // Take group of eight block_q4_0x8 structures at each pass of the loop and perform dot product operation
-        for (int64_t x = 0; x < nc / 8; x++) {
+        for (int64_t x = xstart; x < nc / 8; x++) {

            const block_q4_0x8 * b_ptr = b_ptr_start + (x * b_nb);

@@ -2547,21 +3016,21 @@ void ggml_gemm_q4_0_8x8_q8_0(int n, float * restrict s, size_t bs, const void *
                    // The values arranged in shuffle patterns are operated with dot product operation within 32 bit lane i.e corresponding bytes and multiplied and added into 32 bit integers within 32 bit lane
                    // Resembles MMLAs into 2x2 matrices in ARM Version
                    __m256i iacc_mat_00_sp1 =
-                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_01_3_sp1, rhs_mat_0145_3_sp1), mul_sum_i8_pairs_int(lhs_mat_01_2_sp1, rhs_mat_0145_2_sp1)), mul_sum_i8_pairs_int(lhs_mat_01_1_sp1, rhs_mat_0145_1_sp1)), mul_sum_i8_pairs_int(lhs_mat_01_0_sp1, rhs_mat_0145_0_sp1));
+                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_01_3_sp1, rhs_mat_0145_3_sp1), mul_sum_i8_pairs_int32x8(lhs_mat_01_2_sp1, rhs_mat_0145_2_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_01_1_sp1, rhs_mat_0145_1_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_01_0_sp1, rhs_mat_0145_0_sp1));
                    __m256i iacc_mat_01_sp1 =
-                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_01_3_sp1, rhs_mat_2367_3_sp1), mul_sum_i8_pairs_int(lhs_mat_01_2_sp1, rhs_mat_2367_2_sp1)), mul_sum_i8_pairs_int(lhs_mat_01_1_sp1, rhs_mat_2367_1_sp1)), mul_sum_i8_pairs_int(lhs_mat_01_0_sp1, rhs_mat_2367_0_sp1));
+                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_01_3_sp1, rhs_mat_2367_3_sp1), mul_sum_i8_pairs_int32x8(lhs_mat_01_2_sp1, rhs_mat_2367_2_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_01_1_sp1, rhs_mat_2367_1_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_01_0_sp1, rhs_mat_2367_0_sp1));
                    __m256i iacc_mat_10_sp1 =
-                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_23_3_sp1, rhs_mat_0145_3_sp1), mul_sum_i8_pairs_int(lhs_mat_23_2_sp1, rhs_mat_0145_2_sp1)), mul_sum_i8_pairs_int(lhs_mat_23_1_sp1, rhs_mat_0145_1_sp1)), mul_sum_i8_pairs_int(lhs_mat_23_0_sp1, rhs_mat_0145_0_sp1));
+                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_23_3_sp1, rhs_mat_0145_3_sp1), mul_sum_i8_pairs_int32x8(lhs_mat_23_2_sp1, rhs_mat_0145_2_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_23_1_sp1, rhs_mat_0145_1_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_23_0_sp1, rhs_mat_0145_0_sp1));
                    __m256i iacc_mat_11_sp1 =
-                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_23_3_sp1, rhs_mat_2367_3_sp1), mul_sum_i8_pairs_int(lhs_mat_23_2_sp1, rhs_mat_2367_2_sp1)), mul_sum_i8_pairs_int(lhs_mat_23_1_sp1, rhs_mat_2367_1_sp1)), mul_sum_i8_pairs_int(lhs_mat_23_0_sp1, rhs_mat_2367_0_sp1));
+                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_23_3_sp1, rhs_mat_2367_3_sp1), mul_sum_i8_pairs_int32x8(lhs_mat_23_2_sp1, rhs_mat_2367_2_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_23_1_sp1, rhs_mat_2367_1_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_23_0_sp1, rhs_mat_2367_0_sp1));
                    __m256i iacc_mat_00_sp2 =
-                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_01_3_sp2, rhs_mat_0145_3_sp2), mul_sum_i8_pairs_int(lhs_mat_01_2_sp2, rhs_mat_0145_2_sp2)), mul_sum_i8_pairs_int(lhs_mat_01_1_sp2, rhs_mat_0145_1_sp2)), mul_sum_i8_pairs_int(lhs_mat_01_0_sp2, rhs_mat_0145_0_sp2));
+                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_01_3_sp2, rhs_mat_0145_3_sp2), mul_sum_i8_pairs_int32x8(lhs_mat_01_2_sp2, rhs_mat_0145_2_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_01_1_sp2, rhs_mat_0145_1_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_01_0_sp2, rhs_mat_0145_0_sp2));
                    __m256i iacc_mat_01_sp2 =
-                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_01_3_sp2, rhs_mat_2367_3_sp2), mul_sum_i8_pairs_int(lhs_mat_01_2_sp2, rhs_mat_2367_2_sp2)), mul_sum_i8_pairs_int(lhs_mat_01_1_sp2, rhs_mat_2367_1_sp2)), mul_sum_i8_pairs_int(lhs_mat_01_0_sp2, rhs_mat_2367_0_sp2));
+                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_01_3_sp2, rhs_mat_2367_3_sp2), mul_sum_i8_pairs_int32x8(lhs_mat_01_2_sp2, rhs_mat_2367_2_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_01_1_sp2, rhs_mat_2367_1_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_01_0_sp2, rhs_mat_2367_0_sp2));
                    __m256i iacc_mat_10_sp2 =
-                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_23_3_sp2, rhs_mat_0145_3_sp2), mul_sum_i8_pairs_int(lhs_mat_23_2_sp2, rhs_mat_0145_2_sp2)), mul_sum_i8_pairs_int(lhs_mat_23_1_sp2, rhs_mat_0145_1_sp2)), mul_sum_i8_pairs_int(lhs_mat_23_0_sp2, rhs_mat_0145_0_sp2));
+                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_23_3_sp2, rhs_mat_0145_3_sp2), mul_sum_i8_pairs_int32x8(lhs_mat_23_2_sp2, rhs_mat_0145_2_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_23_1_sp2, rhs_mat_0145_1_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_23_0_sp2, rhs_mat_0145_0_sp2));
                    __m256i iacc_mat_11_sp2 =
-                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_23_3_sp2, rhs_mat_2367_3_sp2), mul_sum_i8_pairs_int(lhs_mat_23_2_sp2, rhs_mat_2367_2_sp2)), mul_sum_i8_pairs_int(lhs_mat_23_1_sp2, rhs_mat_2367_1_sp2)), mul_sum_i8_pairs_int(lhs_mat_23_0_sp2, rhs_mat_2367_0_sp2));
+                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_23_3_sp2, rhs_mat_2367_3_sp2), mul_sum_i8_pairs_int32x8(lhs_mat_23_2_sp2, rhs_mat_2367_2_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_23_1_sp2, rhs_mat_2367_1_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_23_0_sp2, rhs_mat_2367_0_sp2));

                    // Output of both shuffle patterns are added in order to sum dot product outputs of all 32 values in block
                    __m256i iacc_mat_00 = _mm256_add_epi32(iacc_mat_00_sp1, iacc_mat_00_sp2);
@@ -2599,7 +3068,7 @@ void ggml_gemm_q4_0_8x8_q8_0(int n, float * restrict s, size_t bs, const void *
        const block_q8_0x4 * a_ptr = a_ptr_start + (y * nb);

        // Load the eight block_q4_0 quantized values interleaved with each other in chunks of eight - B0,B1 ....B6,B7
-        for (int64_t x = 0; x < nc / 8; x++) {
+        for (int64_t x = xstart; x < nc / 8; x++) {

            const block_q4_0x8 * b_ptr = b_ptr_start + (x * b_nb);

@@ -2711,21 +3180,21 @@ void ggml_gemm_q4_0_8x8_q8_0(int n, float * restrict s, size_t bs, const void *
                // The values arranged in shuffle patterns are operated with dot product operation within 32 bit lane i.e corresponding bytes and multiplied and added into 32 bit integers within 32 bit lane
                // Resembles MMLAs into 2x2 matrices in ARM Version
                __m256i iacc_mat_00_sp1 =
-                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_01_3_sp1, rhs_mat_0145_3_sp1), mul_sum_i8_pairs_int(lhs_mat_01_2_sp1, rhs_mat_0145_2_sp1)), mul_sum_i8_pairs_int(lhs_mat_01_1_sp1, rhs_mat_0145_1_sp1)), mul_sum_i8_pairs_int(lhs_mat_01_0_sp1, rhs_mat_0145_0_sp1));
+                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_01_3_sp1, rhs_mat_0145_3_sp1), mul_sum_i8_pairs_int32x8(lhs_mat_01_2_sp1, rhs_mat_0145_2_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_01_1_sp1, rhs_mat_0145_1_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_01_0_sp1, rhs_mat_0145_0_sp1));
                __m256i iacc_mat_01_sp1 =
-                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_01_3_sp1, rhs_mat_2367_3_sp1), mul_sum_i8_pairs_int(lhs_mat_01_2_sp1, rhs_mat_2367_2_sp1)), mul_sum_i8_pairs_int(lhs_mat_01_1_sp1, rhs_mat_2367_1_sp1)), mul_sum_i8_pairs_int(lhs_mat_01_0_sp1, rhs_mat_2367_0_sp1));
+                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_01_3_sp1, rhs_mat_2367_3_sp1), mul_sum_i8_pairs_int32x8(lhs_mat_01_2_sp1, rhs_mat_2367_2_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_01_1_sp1, rhs_mat_2367_1_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_01_0_sp1, rhs_mat_2367_0_sp1));
                __m256i iacc_mat_10_sp1 =
-                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_23_3_sp1, rhs_mat_0145_3_sp1), mul_sum_i8_pairs_int(lhs_mat_23_2_sp1, rhs_mat_0145_2_sp1)), mul_sum_i8_pairs_int(lhs_mat_23_1_sp1, rhs_mat_0145_1_sp1)), mul_sum_i8_pairs_int(lhs_mat_23_0_sp1, rhs_mat_0145_0_sp1));
+                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_23_3_sp1, rhs_mat_0145_3_sp1), mul_sum_i8_pairs_int32x8(lhs_mat_23_2_sp1, rhs_mat_0145_2_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_23_1_sp1, rhs_mat_0145_1_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_23_0_sp1, rhs_mat_0145_0_sp1));
                __m256i iacc_mat_11_sp1 =
-                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_23_3_sp1, rhs_mat_2367_3_sp1), mul_sum_i8_pairs_int(lhs_mat_23_2_sp1, rhs_mat_2367_2_sp1)), mul_sum_i8_pairs_int(lhs_mat_23_1_sp1, rhs_mat_2367_1_sp1)), mul_sum_i8_pairs_int(lhs_mat_23_0_sp1, rhs_mat_2367_0_sp1));
+                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_23_3_sp1, rhs_mat_2367_3_sp1), mul_sum_i8_pairs_int32x8(lhs_mat_23_2_sp1, rhs_mat_2367_2_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_23_1_sp1, rhs_mat_2367_1_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_23_0_sp1, rhs_mat_2367_0_sp1));
                __m256i iacc_mat_00_sp2 =
-                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_01_3_sp2, rhs_mat_0145_3_sp2), mul_sum_i8_pairs_int(lhs_mat_01_2_sp2, rhs_mat_0145_2_sp2)), mul_sum_i8_pairs_int(lhs_mat_01_1_sp2, rhs_mat_0145_1_sp2)), mul_sum_i8_pairs_int(lhs_mat_01_0_sp2, rhs_mat_0145_0_sp2));
+                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_01_3_sp2, rhs_mat_0145_3_sp2), mul_sum_i8_pairs_int32x8(lhs_mat_01_2_sp2, rhs_mat_0145_2_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_01_1_sp2, rhs_mat_0145_1_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_01_0_sp2, rhs_mat_0145_0_sp2));
                __m256i iacc_mat_01_sp2 =
-                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_01_3_sp2, rhs_mat_2367_3_sp2), mul_sum_i8_pairs_int(lhs_mat_01_2_sp2, rhs_mat_2367_2_sp2)), mul_sum_i8_pairs_int(lhs_mat_01_1_sp2, rhs_mat_2367_1_sp2)), mul_sum_i8_pairs_int(lhs_mat_01_0_sp2, rhs_mat_2367_0_sp2));
+                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_01_3_sp2, rhs_mat_2367_3_sp2), mul_sum_i8_pairs_int32x8(lhs_mat_01_2_sp2, rhs_mat_2367_2_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_01_1_sp2, rhs_mat_2367_1_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_01_0_sp2, rhs_mat_2367_0_sp2));
                __m256i iacc_mat_10_sp2 =
-                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_23_3_sp2, rhs_mat_0145_3_sp2), mul_sum_i8_pairs_int(lhs_mat_23_2_sp2, rhs_mat_0145_2_sp2)), mul_sum_i8_pairs_int(lhs_mat_23_1_sp2, rhs_mat_0145_1_sp2)), mul_sum_i8_pairs_int(lhs_mat_23_0_sp2, rhs_mat_0145_0_sp2));
+                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_23_3_sp2, rhs_mat_0145_3_sp2), mul_sum_i8_pairs_int32x8(lhs_mat_23_2_sp2, rhs_mat_0145_2_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_23_1_sp2, rhs_mat_0145_1_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_23_0_sp2, rhs_mat_0145_0_sp2));
                __m256i iacc_mat_11_sp2 =
-                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_23_3_sp2, rhs_mat_2367_3_sp2), mul_sum_i8_pairs_int(lhs_mat_23_2_sp2, rhs_mat_2367_2_sp2)), mul_sum_i8_pairs_int(lhs_mat_23_1_sp2, rhs_mat_2367_1_sp2)), mul_sum_i8_pairs_int(lhs_mat_23_0_sp2, rhs_mat_2367_0_sp2));
+                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_23_3_sp2, rhs_mat_2367_3_sp2), mul_sum_i8_pairs_int32x8(lhs_mat_23_2_sp2, rhs_mat_2367_2_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_23_1_sp2, rhs_mat_2367_1_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_23_0_sp2, rhs_mat_2367_0_sp2));

                // Output of both shuffle patterns are added in order to sum dot product outputs of all 32 values in block
                __m256i iacc_mat_00 = _mm256_add_epi32(iacc_mat_00_sp1, iacc_mat_00_sp2);
@@ -227,6 +227,7 @@ struct ggml_backend_cann_context {
     * @brief Destructor for cleaning up resources.
     */
    ~ggml_backend_cann_context() {
+        ggml_cann_set_device(device);
        if (copy_event != nullptr) {
            ACL_CHECK(aclrtDestroyEvent(copy_event));
        }
@@ -2899,6 +2899,9 @@ GGML_CALL static bool ggml_backend_cuda_supports_op(ggml_backend_t backend, cons
                if (src0_type == GGML_TYPE_F32 && src1_type == GGML_TYPE_Q8_0) {
                    return true;
                }
+                if (src0_type == GGML_TYPE_Q8_0 && src1_type == GGML_TYPE_F32) {
+                    return true;
+                }
                if (src0_type == GGML_TYPE_F32 && src1_type == GGML_TYPE_Q4_0) {
                    return true;
                }
@@ -81,6 +81,17 @@ static __device__ void cpy_blck_f32_q8_0(const char * cxi, char * cdsti) {
    }
 }

+static __device__ void cpy_blck_q8_0_f32(const char * cxi, char * cdsti) {
+    const block_q8_0 * xi = (const block_q8_0 *) cxi;
+    float * dsti = (float *) cdsti;
+
+    const float d = (float)xi->d;
+
+    for (int j = 0; j < QK8_0; j++) {
+       dsti[j] = xi->qs[j] * d;
+    }
+}
+
 static __device__ void cpy_blck_f32_q4_0(const char * cxi, char * cdsti) {
    const float * xi = (const float *) cxi;
    block_q4_0 * dsti = (block_q4_0 *) cdsti;
@@ -288,6 +299,32 @@ static __global__ void cpy_f32_q(const char * cx, char * cdst, const int ne,
    cpy_blck(cx + x_offset, cdst + dst_offset);
 }

+template <cpy_kernel_t cpy_blck, int qk>
+static __global__ void cpy_q_f32(const char * cx, char * cdst, const int ne,
+                                 const int ne00, const int ne01, const int ne02, const int nb00, const int nb01, const int nb02,
+                                 const int nb03, const int ne10, const int ne11, const int ne12, const int nb10, const int nb11,
+                                 const int nb12, const int nb13) {
+    const int i = (blockDim.x*blockIdx.x + threadIdx.x)*qk;
+
+    if (i >= ne) {
+        return;
+    }
+
+    const int i03 = i/(ne00 * ne01 * ne02);
+    const int i02 = (i - i03*ne00*ne01*ne02 )/ (ne00*ne01);
+    const int i01 = (i - i03*ne00*ne01*ne02  -  i02*ne01*ne00) / ne00;
+    const int i00 = i - i03*ne00*ne01*ne02 - i02*ne01*ne00 - i01*ne00;
+    const int x_offset = (i00/qk)*nb00 + i01*nb01 + i02*nb02 + i03 * nb03;
+
+    const int i13 = i/(ne10 * ne11 * ne12);
+    const int i12 = (i - i13*ne10*ne11*ne12) / (ne10*ne11);
+    const int i11 = (i - i13*ne10*ne11*ne12 - i12*ne10*ne11) / ne10;
+    const int i10 = i - i13*ne10*ne11*ne12 - i12*ne10*ne11 - i11*ne10;
+    const int dst_offset = i10*nb10 + i11*nb11 + i12*nb12 + i13*nb13;
+
+    cpy_blck(cx + x_offset, cdst + dst_offset);
+}
+
 static void ggml_cpy_f16_f32_cuda(
    const char * cx, char * cdst, const int ne,
    const int ne00, const int ne01, const int ne02, const int nb00, const int nb01, const int nb02,
@@ -329,6 +366,16 @@ static void ggml_cpy_f32_q8_0_cuda(
        (cx, cdst, ne, ne00, ne01, ne02, nb00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb13);
 }

+static void ggml_cpy_q8_0_f32_cuda(
+    const char * cx, char * cdst, const int ne,
+    const int ne00, const int ne01, const int ne02, const int nb00, const int nb01, const int nb02,
+    const int nb03, const int ne10, const int ne11, const int ne12, const int nb10, const int nb11, const int nb12, const int nb13, cudaStream_t stream) {
+
+    const int num_blocks = ne;
+    cpy_q_f32<cpy_blck_q8_0_f32, QK8_0><<<num_blocks, 1, 0, stream>>>
+        (cx, cdst, ne, ne00, ne01, ne02, nb00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb13);
+}
+
 static void ggml_cpy_f32_q4_0_cuda(
    const char * cx, char * cdst, const int ne,
    const int ne00, const int ne01, const int ne02, const int nb00, const int nb01, const int nb02,
@@ -437,6 +484,8 @@ void ggml_cuda_cpy(ggml_backend_cuda_context & ctx, const ggml_tensor * src0, gg
        ggml_cpy_f32_f16_cuda (src0_ddc, src1_ddc, ne, ne00, ne01, ne02, nb00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb13, main_stream);
    } else if (src0->type == GGML_TYPE_F32 && src1->type == GGML_TYPE_Q8_0) {
        ggml_cpy_f32_q8_0_cuda(src0_ddc, src1_ddc, ne, ne00, ne01, ne02, nb00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb13, main_stream);
+    } else if (src0->type == GGML_TYPE_Q8_0 && src1->type == GGML_TYPE_F32) {
+        ggml_cpy_q8_0_f32_cuda(src0_ddc, src1_ddc, ne, ne00, ne01, ne02, nb00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb13, main_stream);
    } else if (src0->type == GGML_TYPE_F32 && src1->type == GGML_TYPE_Q4_0) {
        ggml_cpy_f32_q4_0_cuda(src0_ddc, src1_ddc, ne, ne00, ne01, ne02, nb00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb13, main_stream);
    } else if (src0->type == GGML_TYPE_F32 && src1->type == GGML_TYPE_Q4_1) {
@@ -471,6 +520,8 @@ void* ggml_cuda_cpy_fn(const ggml_tensor * src0, ggml_tensor * src1) {
        return (void*) cpy_f32_f16<cpy_1_f32_f16>;
    } else if (src0->type == GGML_TYPE_F32 && src1->type == GGML_TYPE_Q8_0) {
        return (void*) cpy_f32_q<cpy_blck_f32_q8_0, QK8_0>;
+    } else if (src0->type == GGML_TYPE_Q8_0 && src1->type == GGML_TYPE_F32) {
+        return (void*) cpy_q_f32<cpy_blck_q8_0_f32, QK8_0>;
    } else if (src0->type == GGML_TYPE_F32 && src1->type == GGML_TYPE_Q4_0) {
        return (void*) cpy_f32_q<cpy_blck_f32_q4_0, QK4_0>;
    } else if (src0->type == GGML_TYPE_F32 && src1->type == GGML_TYPE_Q4_1) {
@@ -63,6 +63,25 @@ int ggml_sve_cnt_b = 0;
 #pragma warning(disable: 4702)
 #endif

+// Note: once we move threading into a separate C++ file
+// will use std::hardware_destructive_interference_size instead of hardcoding it here
+// and we'll use C++ attribute syntax.
+#define GGML_CACHE_LINE  64
+
+#if defined(__clang__) || defined(__GNUC__)
+#define GGML_CACHE_ALIGN __attribute__((aligned(GGML_CACHE_LINE)))
+#endif
+
+#if defined(__has_feature)
+#if __has_feature(thread_sanitizer)
+#define GGML_TSAN_ENABLED 1
+#endif
+#else  // __has_feature
+#if defined(__SANITIZE_THREAD__)
+#define GGML_TSAN_ENABLED 1
+#endif
+#endif // __has_feature
+
 #if defined(_WIN32)

 #define WIN32_LEAN_AND_MEAN
@@ -72,6 +91,8 @@ int ggml_sve_cnt_b = 0;
 #include <windows.h>

 #if !defined(__clang__)
+#define GGML_CACHE_ALIGN __declspec(align(GGML_CACHE_LINE))
+
 typedef volatile LONG atomic_int;
 typedef atomic_int atomic_bool;
 typedef atomic_int atomic_flag;
@@ -114,6 +135,9 @@ static atomic_bool atomic_flag_test_and_set(atomic_flag * ptr) {
 static void atomic_flag_clear(atomic_flag * ptr) {
    InterlockedExchange(ptr, 0);
 }
+static void atomic_thread_fence(memory_order mo) {
+    MemoryBarrier();
+}
 #else // clang
 #include <stdatomic.h>
 #endif
@@ -289,7 +313,6 @@ void ggml_abort(const char * file, int line, const char * fmt, ...) {
 #define GGML_DEBUG 0
 #define GGML_GELU_FP16
 #define GGML_GELU_QUICK_FP16
-#define GGML_N_TASKS_MAX (-1)

 #define GGML_SOFT_MAX_UNROLL 4
 #define GGML_VEC_DOT_UNROLL  2
@@ -2007,8 +2030,8 @@ struct ggml_threadpool {

    // synchronization primitives
    atomic_int n_graph;       // incremented when there is work to be done (i.e each graph)
-    atomic_int n_barrier;
-    atomic_int n_barrier_passed;
+    atomic_int GGML_CACHE_ALIGN n_barrier;
+    atomic_int GGML_CACHE_ALIGN n_barrier_passed;
    atomic_int current_chunk; // currently processing chunk during Mat_Mul, shared between all the threads.

    // these are atomic as an annotation for thread-sanitizer
@@ -3196,20 +3219,27 @@ static void ggml_barrier(struct ggml_threadpool * tp) {
    // enter barrier (full seq-cst fence)
    int n_barrier = atomic_fetch_add_explicit(&tp->n_barrier, 1, memory_order_seq_cst);

-    int last = 0;
    if (n_barrier == (n_threads - 1)) {
        // last thread
        atomic_store_explicit(&tp->n_barrier, 0, memory_order_relaxed);
-        last = 1;
-    } else {
-        // wait for other threads
-        while (atomic_load_explicit(&tp->n_barrier_passed, memory_order_relaxed) == n_passed) {
-            ggml_thread_cpu_relax();
-        }
+
+        // exit barrier (fill seq-cst fence)
+        atomic_fetch_add_explicit(&tp->n_barrier_passed, 1, memory_order_seq_cst);
+        return;
+    }
+
+    // wait for other threads
+    while (atomic_load_explicit(&tp->n_barrier_passed, memory_order_relaxed) == n_passed) {
+        ggml_thread_cpu_relax();
    }

    // exit barrier (full seq-cst fence)
-    atomic_fetch_add_explicit(&tp->n_barrier_passed, last, memory_order_seq_cst);
+    // TSAN doesn't support standalone fence yet, we use a dummy read-modify-write instead
+    #ifdef GGML_TSAN_ENABLED
+    atomic_fetch_add_explicit(&tp->n_barrier_passed, 0, memory_order_seq_cst);
+    #else
+    atomic_thread_fence(memory_order_seq_cst);
+    #endif
 #endif
 }

@@ -20240,10 +20270,13 @@ static inline bool ggml_graph_compute_thread_ready(struct ggml_compute_state * s

 // sync thread state after polling
 static inline void ggml_graph_compute_thread_sync(struct ggml_compute_state * state) {
-    struct ggml_threadpool * threadpool = state->threadpool;
-    // this should just be atomic_thread_fence(seq_cst) but it confuses thread-sanitizer
-    // so instead we just use a dummy read-modify-write
-    atomic_fetch_add_explicit(&threadpool->n_graph, 0, memory_order_seq_cst);
+    // TSAN doesn't support standalone fence yet, we use a dummy read-modify-write instead
+    #ifdef GGML_TSAN_ENABLED
+    atomic_fetch_add_explicit(&state->threadpool->n_graph, 0, memory_order_seq_cst);
+    #else
+    atomic_thread_fence(memory_order_seq_cst);
+    #endif
+    UNUSED(state);
 }

 static inline bool ggml_graph_compute_poll_for_work(struct ggml_compute_state * state) {
@@ -235,6 +235,7 @@ class MODEL_ARCH(IntEnum):
    NEMOTRON     = auto()
    EXAONE       = auto()
    GRANITE      = auto()
+    GRANITE_MOE  = auto()


 class MODEL_TENSOR(IntEnum):
@@ -392,6 +393,7 @@ MODEL_ARCH_NAMES: dict[MODEL_ARCH, str] = {
    MODEL_ARCH.NEMOTRON:       "nemotron",
    MODEL_ARCH.EXAONE:         "exaone",
    MODEL_ARCH.GRANITE:        "granite",
+    MODEL_ARCH.GRANITE_MOE:    "granitemoe",
 }

 TENSOR_NAMES: dict[MODEL_TENSOR, str] = {
@@ -1232,6 +1234,7 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
    MODEL_ARCH.GRANITE: [
        MODEL_TENSOR.TOKEN_EMBD,
        MODEL_TENSOR.OUTPUT_NORM,
+        MODEL_TENSOR.OUTPUT,
        MODEL_TENSOR.ATTN_NORM,
        MODEL_TENSOR.ATTN_Q,
        MODEL_TENSOR.ATTN_K,
@@ -1242,6 +1245,21 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
        MODEL_TENSOR.FFN_DOWN,
        MODEL_TENSOR.FFN_UP,
    ],
+    MODEL_ARCH.GRANITE_MOE: [
+        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_INP,
+        MODEL_TENSOR.FFN_GATE_EXP,
+        MODEL_TENSOR.FFN_DOWN_EXP,
+        MODEL_TENSOR.FFN_UP_EXP,
+    ],
    # TODO
 }

@@ -251,11 +251,12 @@ class TensorNameMap:
        ),

        MODEL_TENSOR.FFN_GATE_INP: (
-            "layers.{bid}.feed_forward.gate",             # mixtral
-            "model.layers.{bid}.block_sparse_moe.gate",   # mixtral
-            "model.layers.{bid}.mlp.gate",                # qwen2moe olmoe
-            "transformer.decoder_layer.{bid}.router",     # Grok
-            "transformer.blocks.{bid}.ffn.router.layer",  # dbrx
+            "layers.{bid}.feed_forward.gate",                   # mixtral
+            "model.layers.{bid}.block_sparse_moe.gate",         # mixtral
+            "model.layers.{bid}.mlp.gate",                      # qwen2moe olmoe
+            "transformer.decoder_layer.{bid}.router",           # Grok
+            "transformer.blocks.{bid}.ffn.router.layer",        # dbrx
+            "model.layers.{bid}.block_sparse_moe.router.layer", # granitemoe
        ),

        MODEL_TENSOR.FFN_GATE_INP_SHEXP: (
@@ -364,10 +365,11 @@ class TensorNameMap:
        ),

        MODEL_TENSOR.FFN_DOWN_EXP: (
-            "layers.{bid}.feed_forward.experts.w2",          # mixtral (merged)
-            "transformer.decoder_layer.{bid}.moe.linear_1",  # Grok (merged)
-            "transformer.blocks.{bid}.ffn.experts.mlp.w2",   # dbrx
-            "model.layers.{bid}.mlp.experts.down_proj",      # qwen2moe olmoe (merged)
+            "layers.{bid}.feed_forward.experts.w2",              # mixtral (merged)
+            "transformer.decoder_layer.{bid}.moe.linear_1",      # Grok (merged)
+            "transformer.blocks.{bid}.ffn.experts.mlp.w2",       # dbrx
+            "model.layers.{bid}.mlp.experts.down_proj",          # qwen2moe olmoe (merged)
+            "model.layers.{bid}.block_sparse_moe.output_linear", # granitemoe
        ),

        MODEL_TENSOR.FFN_DOWN_SHEXP: (
@@ -1066,6 +1066,7 @@ extern "C" {
    LLAMA_API struct llama_sampler * llama_sampler_init_dist       (uint32_t seed);

    /// @details Sorts candidate tokens by their logits in descending order and calculate probabilities based on logits.
+    /// NOTE: Avoid using on the full vocabulary as the sorting can become slow. For example, apply top-k or top-p sampling first.
    LLAMA_API struct llama_sampler * llama_sampler_init_softmax    (void);

    /// @details Top-K sampling described in academic paper "The Curious Case of Neural Text Degeneration" https://arxiv.org/abs/1904.09751
@@ -1 +1 @@
-e7b23907cb2816e9951fe9b524d7127ab777297a
+336c10a4c3c8ec99af484b25a0cddd397a09cdb2
@@ -28,6 +28,8 @@ void llama_log_callback_default(ggml_log_level level, const char * text, void *
 #define LLAMA_LOG_INFO(...)  llama_log_internal(GGML_LOG_LEVEL_INFO , __VA_ARGS__)
 #define LLAMA_LOG_WARN(...)  llama_log_internal(GGML_LOG_LEVEL_WARN , __VA_ARGS__)
 #define LLAMA_LOG_ERROR(...) llama_log_internal(GGML_LOG_LEVEL_ERROR, __VA_ARGS__)
+#define LLAMA_LOG_DEBUG(...) llama_log_internal(GGML_LOG_LEVEL_DEBUG, __VA_ARGS__)
+#define LLAMA_LOG_CONT(...)  llama_log_internal(GGML_LOG_LEVEL_CONT , __VA_ARGS__)

 //
 // helpers
@@ -3,13 +3,14 @@
 #include "llama-vocab.h"
 #include "llama-grammar.h"

-#include <cassert>
 #include <algorithm>
-#include <cstring>
-#include <ctime>
+#include <cassert>
 #include <cfloat>
 #include <chrono>
 #include <cmath>
+#include <cstdlib>
+#include <cstring>
+#include <ctime>
 #include <numeric>
 #include <random>
 #include <unordered_map>
@@ -1570,11 +1570,7 @@ llama_token_attr llama_token_get_attr_impl(const struct llama_vocab & vocab, lla
 }

 bool llama_token_is_eog_impl(const struct llama_vocab & vocab, llama_token token) {
-    return token != -1 && (
-        token == llama_token_eos_impl(vocab) ||
-        token == llama_token_eot_impl(vocab) ||
-        token == llama_token_eom_impl(vocab)
-    );
+    return token != -1 && vocab.special_eog_ids.count(token) > 0;
 }

 bool llama_token_is_control_impl(const struct llama_vocab & vocab, llama_token token) {
@@ -6,6 +6,7 @@
 #include <vector>
 #include <unordered_map>
 #include <map>
+#include <set>

 struct llama_vocab {
    using id    = llama_token;
@@ -49,12 +50,15 @@ struct llama_vocab {
    id special_eot_id    = -1; // TODO: move above after "eos_id", and here add "file separator" token
    id special_eom_id    = -1;

+    // set of all tokens that cause "end of generation"
+    std::set<id> special_eog_ids;
+
    // tokenizer flags
-    bool tokenizer_add_space_prefix = false;
-    bool tokenizer_add_bos          = false;
-    bool tokenizer_add_eos          = false;
-    bool tokenizer_ignore_merges    = false;
-    bool tokenizer_clean_spaces     = false;  // clean_up_tokenization_spaces
+    bool tokenizer_add_space_prefix           = false;
+    bool tokenizer_add_bos                    = false;
+    bool tokenizer_add_eos                    = false;
+    bool tokenizer_ignore_merges              = false;
+    bool tokenizer_clean_spaces               = false;  // clean_up_tokenization_spaces
    bool tokenizer_remove_extra_whitespaces   = false;
    bool tokenizer_escape_whitespaces         = true;
    bool tokenizer_treat_whitespace_as_suffix = false;
@@ -215,6 +215,7 @@ enum llm_arch {
    LLM_ARCH_EXAONE,
    LLM_ARCH_RWKV6,
    LLM_ARCH_GRANITE,
+    LLM_ARCH_GRANITE_MOE,
    LLM_ARCH_UNKNOWN,
 };

@@ -266,6 +267,7 @@ static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
    { LLM_ARCH_EXAONE,          "exaone"       },
    { LLM_ARCH_RWKV6,           "rwkv6"        },
    { LLM_ARCH_GRANITE,         "granite"      },
+    { LLM_ARCH_GRANITE_MOE,     "granitemoe"   },
    { LLM_ARCH_UNKNOWN,         "(unknown)"    },
 };

@@ -1467,6 +1469,7 @@ static const std::map<llm_arch, std::map<llm_tensor, std::string>> LLM_TENSOR_NA
        {
            { 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" },
@@ -1478,6 +1481,24 @@ static const std::map<llm_arch, std::map<llm_tensor, std::string>> LLM_TENSOR_NA
            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
        },
    },
+    {
+        LLM_ARCH_GRANITE_MOE,
+        {
+            { 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_INP,    "blk.%d.ffn_gate_inp" },
+            { LLM_TENSOR_FFN_GATE_EXPS,   "blk.%d.ffn_gate_exps" },
+            { LLM_TENSOR_FFN_DOWN_EXPS,   "blk.%d.ffn_down_exps" },
+            { LLM_TENSOR_FFN_UP_EXPS,     "blk.%d.ffn_up_exps" },
+        },
+    },
    {
        LLM_ARCH_UNKNOWN,
        {
@@ -2396,7 +2417,7 @@ struct llama_hparams {
    float f_max_alibi_bias = 0.0f;
    float f_logit_scale    = 0.0f;

-    // Additional scale factors (Granite)
+    // Additional scale factors (Granite/Granite MoE)
    float f_residual_scale  = 0.0f;
    float f_embedding_scale = 0.0f;
    float f_attention_scale = 0.0f;
@@ -6048,6 +6069,7 @@ static void llm_load_hparams(
                }
            } break;
        case LLM_ARCH_GRANITE:
+        case LLM_ARCH_GRANITE_MOE:
            {
                ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
                ml.get_key(LLM_KV_LOGIT_SCALE, hparams.f_logit_scale);
@@ -6056,6 +6078,7 @@ static void llm_load_hparams(
                ml.get_key(LLM_KV_ATTENTION_SCALE, hparams.f_attention_scale);

                switch (hparams.n_layer) {
+                    case 32: model.type = e_model::MODEL_3B; break;
                    case 40: model.type = e_model::MODEL_3B; break;
                    // Add additional layer/vocab/etc checks here for other model sizes
                    default: model.type = e_model::MODEL_UNKNOWN;
@@ -6509,21 +6532,21 @@ static void llm_load_vocab(
        //       for now, we apply this workaround to find the EOT token based on its text
        if (vocab.special_eot_id == -1) {
            for (const auto & t : vocab.token_to_id) {
-                if (
+                if (false
                        // TODO: gemma "<end_of_turn>" is exported as a normal token, so the following check does not work
                        //       need to fix convert script
                        //vocab.id_to_token[t.second].type == LLAMA_TOKEN_TYPE_CONTROL &&
-                        (t.first == "<|eot_id|>" ||
-                         t.first == "<|im_end|>" ||
-                         t.first == "<|end|>" ||
-                         t.first == "<end_of_turn>" ||
-                         t.first == "<|endoftext|>"
-                        )
+                        || t.first == "<|eot_id|>"
+                        || t.first == "<|im_end|>"
+                        || t.first == "<|end|>"
+                        || t.first == "<end_of_turn>"
+                        || t.first == "<|endoftext|>"
+                        || t.first == "<EOT>"
                   ) {
                    vocab.special_eot_id = t.second;
                    if ((vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
                        LLAMA_LOG_WARN("%s: control-looking token: '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
-                            __func__, t.first.c_str());
+                                __func__, t.first.c_str());
                        vocab.id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_CONTROL;
                    }
                    break;
@@ -6546,6 +6569,44 @@ static void llm_load_vocab(
                }
            }
        }
+
+        // maintain a list of tokens that cause end-of-generation
+        // this is currently determined based on the token text, which is obviously not ideal
+        // ref: https://github.com/ggerganov/llama.cpp/issues/9606
+        vocab.special_eog_ids.clear();
+        for (const auto & t : vocab.token_to_id) {
+            if (false
+                    || t.first == "<|eot_id|>"
+                    || t.first == "<|im_end|>"
+                    || t.first == "<|end|>"
+                    || t.first == "<end_of_turn>"
+                    || t.first == "<|endoftext|>"
+                    || t.first == "<|eom_id|>"
+                    || t.first == "<EOT>"
+               ) {
+                vocab.special_eog_ids.insert(t.second);
+                if ((vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
+                    LLAMA_LOG_WARN("%s: control-looking token: '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
+                            __func__, t.first.c_str());
+                    vocab.id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_CONTROL;
+                }
+            }
+        }
+
+        if (vocab.special_eos_id != -1 && vocab.special_eog_ids.count(vocab.special_eos_id) == 0) {
+            vocab.special_eog_ids.insert(vocab.special_eos_id);
+            LLAMA_LOG_WARN("%s: special_eos_id is not in special_eog_ids - the tokenizer config may be incorrect\n", __func__);
+        }
+
+        if (vocab.special_eot_id != -1 && vocab.special_eog_ids.count(vocab.special_eot_id) == 0) {
+            vocab.special_eog_ids.insert(vocab.special_eot_id);
+            LLAMA_LOG_WARN("%s: special_eot_id is not in special_eog_ids - the tokenizer config may be incorrect\n", __func__);
+        }
+
+        if (vocab.special_eom_id != -1 && vocab.special_eog_ids.count(vocab.special_eom_id) == 0) {
+            vocab.special_eog_ids.insert(vocab.special_eom_id);
+            LLAMA_LOG_WARN("%s: special_eom_id is not in special_eog_ids - the tokenizer config may be incorrect\n", __func__);
+        }
    }

    // build special tokens cache
@@ -6749,6 +6810,11 @@ static void llm_load_print_meta(llama_model_loader & ml, llama_model & model) {
    if (vocab.special_suffix_id != -1) { LLAMA_LOG_INFO( "%s: SUF token        = %d '%s'\n", __func__, vocab.special_suffix_id, vocab.id_to_token[vocab.special_suffix_id].text.c_str() ); }
    if (vocab.special_middle_id != -1) { LLAMA_LOG_INFO( "%s: MID token        = %d '%s'\n", __func__, vocab.special_middle_id, vocab.id_to_token[vocab.special_middle_id].text.c_str() ); }
    if (vocab.special_eot_id    != -1) { LLAMA_LOG_INFO( "%s: EOT token        = %d '%s'\n", __func__, vocab.special_eot_id,    vocab.id_to_token[vocab.special_eot_id].text.c_str() );    }
+    if (vocab.special_eom_id    != -1) { LLAMA_LOG_INFO( "%s: EOM token        = %d '%s'\n", __func__, vocab.special_eom_id,    vocab.id_to_token[vocab.special_eom_id].text.c_str() );    }
+
+    for (const auto & id : vocab.special_eog_ids) {
+        LLAMA_LOG_INFO( "%s: EOG token        = %d '%s'\n", __func__, id, vocab.id_to_token[id].text.c_str() );
+    }

    LLAMA_LOG_INFO("%s: max token length = %d\n", __func__, vocab.max_token_len);

@@ -6767,7 +6833,7 @@ static void llm_load_print_meta(llama_model_loader & ml, llama_model & model) {
        LLAMA_LOG_INFO("%s: n_ff_shexp       = %d\n",     __func__, hparams.n_ff_shexp);
    }

-    if (model.arch == LLM_ARCH_GRANITE) {
+    if (model.arch == LLM_ARCH_GRANITE || model.arch == LLM_ARCH_GRANITE_MOE) {
        LLAMA_LOG_INFO("%s: f_embedding_scale = %f\n", __func__, hparams.f_embedding_scale);
        LLAMA_LOG_INFO("%s: f_residual_scale  = %f\n", __func__, hparams.f_residual_scale);
        LLAMA_LOG_INFO("%s: f_attention_scale = %f\n", __func__, hparams.f_attention_scale);
@@ -6941,6 +7007,7 @@ static bool llm_load_tensors(
            case LLM_ARCH_REFACT:
            case LLM_ARCH_MINICPM:
            case LLM_ARCH_GRANITE:
+            case LLM_ARCH_GRANITE_MOE:
                {
                    model.tok_embd = ml.create_tensor(ctx_input, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab});

@@ -9930,17 +9997,36 @@ struct llm_build_context {
            const int64_t n_head_kv = hparams.n_head_kv(il);
            const int64_t n_embd_k_gqa = hparams.n_embd_k_gqa(il);
            struct ggml_tensor * rope_factors = build_rope_factors(il);
-            struct ggml_tensor * tmp =
-                // we rotate only the first n_rot dimensions
-                ggml_rope_ext_inplace(ctx0,
-                        ggml_view_3d(ctx0, kv_self.k_l[il],
-                            n_embd_head_k, n_head_kv, n_ctx,
-                            ggml_row_size(kv_self.k_l[il]->type, n_embd_head_k),
-                            ggml_row_size(kv_self.k_l[il]->type, n_embd_k_gqa),
-                            0),
+            struct ggml_tensor * k =
+                ggml_view_3d(ctx0, kv_self.k_l[il],
+                    n_embd_head_k, n_head_kv, n_ctx,
+                    ggml_row_size(kv_self.k_l[il]->type, n_embd_head_k),
+                    ggml_row_size(kv_self.k_l[il]->type, n_embd_k_gqa),
+                    0);
+
+            struct ggml_tensor * tmp;
+            if (ggml_is_quantized(k->type)) {
+                // dequantize to f32 -> RoPE -> quantize back
+                tmp = ggml_cast(ctx0, k, GGML_TYPE_F32);
+                cb(tmp, "K_f32", il);
+                for (auto * backend : lctx.backends) {
+                    // Figure out which backend KV cache belongs to
+                    if (ggml_backend_supports_buft(backend, lctx.model.buft_layer[il].buft)) {
+                        ggml_backend_sched_set_tensor_backend(lctx.sched, tmp, backend);
+                        break;
+                    }
+                }
+                tmp = ggml_rope_ext_inplace(ctx0, tmp,
                        lctx.inp_K_shift, rope_factors, n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
                        ext_factor, attn_factor, beta_fast, beta_slow);
-
+                cb(tmp, "K_shifted_f32", il);
+                tmp = ggml_cpy(ctx0, tmp, k);
+            } else {
+                // we rotate only the first n_rot dimensions
+                tmp = ggml_rope_ext_inplace(ctx0, k,
+                        lctx.inp_K_shift, rope_factors, n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                        ext_factor, attn_factor, beta_fast, beta_slow);
+            }
            cb(tmp, "K_shifted", il);
            ggml_build_forward_expand(gf, tmp);
        }
@@ -15868,6 +15954,7 @@ static struct ggml_cgraph * llama_build_graph(
    switch (model.arch) {
        case LLM_ARCH_LLAMA:
        case LLM_ARCH_GRANITE:
+        case LLM_ARCH_GRANITE_MOE:
            {
                result = llm.build_llama();
            } break;
@@ -18652,9 +18739,9 @@ struct llama_model * llama_load_model_from_file(
            unsigned percentage = (unsigned) (100 * progress);
            while (percentage > *cur_percentage_p) {
                *cur_percentage_p = percentage;
-                LLAMA_LOG(".");
+                LLAMA_LOG_CONT(".");
                if (percentage >= 100) {
-                    LLAMA_LOG("\n");
+                    LLAMA_LOG_CONT("\n");
                }
            }
            return true;
@@ -19169,6 +19256,7 @@ enum llama_rope_type llama_rope_type(const struct llama_model * model) {
        case LLM_ARCH_DEEPSEEK2:
        case LLM_ARCH_CHATGLM:
        case LLM_ARCH_GRANITE:
+        case LLM_ARCH_GRANITE_MOE:
            return LLAMA_ROPE_TYPE_NORM;

        // the pairs of head values are offset by n_rot/2
@@ -1,6 +1,5 @@
 #include "ggml.h"
 #include "llama.h"
-#include "llama-sampling.h"

 #ifdef NDEBUG
 #undef NDEBUG
@@ -249,6 +248,45 @@ static void test_sampler_queue(const size_t n_vocab, const std::string & sampler
           samplers_sequence.c_str(), n_vocab, top_k, top_p, min_p);
 }

+static void bench(llama_sampler * cnstr, const char * cnstr_name, const std::vector<llama_token_data> & data, int n_iter) {
+    std::vector<llama_token_data> cur(data.size());
+    std::copy(data.begin(), data.end(), cur.begin());
+    llama_token_data_array cur_p = { cur.data(), cur.size(), -1, false };
+    llama_sampler_apply(cnstr, &cur_p);
+    llama_sampler_reset(cnstr);
+    const int64_t t_start = ggml_time_us();
+    for (int i = 0; i < n_iter; i++) {
+        std::copy(data.begin(), data.end(), cur.begin());
+        llama_token_data_array cur_p = { cur.data(), cur.size(), -1, false };
+        llama_sampler_apply(cnstr, &cur_p);
+        llama_sampler_reset(cnstr);
+    }
+    const int64_t t_end = ggml_time_us();
+    llama_sampler_free(cnstr);
+    printf("%-42s: %8.3f us/iter\n", cnstr_name, (t_end - t_start) / (float)n_iter);
+}
+
+#define BENCH(__cnstr, __data, __n_iter) bench((__cnstr), #__cnstr, (__data), (__n_iter))
+
+static void test_perf() {
+    const int n_vocab = 1 << 17;
+
+    std::vector<llama_token_data> data;
+
+    data.reserve(n_vocab);
+    for (int i = 0; i < n_vocab; i++) {
+        const float logit = 2.0f*((float)(rand())/RAND_MAX - 0.5f);
+        data.emplace_back(llama_token_data{i, logit, 0.0f});
+    }
+
+    BENCH(llama_sampler_init_top_k    (40),      data, 32);
+    BENCH(llama_sampler_init_top_p    (0.8f, 1), data, 32);
+    BENCH(llama_sampler_init_min_p    (0.2f, 1), data, 32);
+    BENCH(llama_sampler_init_tail_free(0.5f, 1), data, 32);
+    BENCH(llama_sampler_init_typical  (0.5f, 1), data, 32);
+    BENCH(llama_sampler_init_softmax  (),        data, 32);
+}
+
 int main(void) {
    ggml_time_init();

@@ -316,5 +354,7 @@ int main(void) {

    printf("OK\n");

+    test_perf();
+
    return 0;
 }
Author	SHA1	Message	Date
Gabe Goodhart	3d6bf6919f	llama : add IBM Granite MoE architecture (#9438 ) * feat(gguf-py): Add granitemoe architecture This includes the addition of new tensor names for the new moe layers. These may not be correct at this point due to the need for the hack in gguf_writer.py to double-check the length of the shape for these layers. Branch: GraniteMoE Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * feat(convert_hf_to_gguf): Add GraniteMoeModel GraniteMoe has the same configuration deltas as Granite Branch: GraniteMoE Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix(granitemoe convert): Split the double-sized input layer into gate and up After a lot of staring and squinting, it's clear that the standard mixtral expert implementation is equivalent to the vectorized parallel experts in granite. The difference is that in granite, the w1 and w3 are concatenated into a single tensor "input_linear." Rather than reimplementing all of the math on the llama.cpp side, the much simpler route is to just split this tensor during conversion and follow the standard mixtral route. Branch: GraniteMoE Co-Authored-By: alex.brooks@ibm.com Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * feat(granitemoe): Implement granitemoe GraniteMoE follows the mixtral architecture (once the input_linear layers are split into gate_exps/up_exps). The main delta is the addition of the same four multipliers used in Granite. Branch: GraniteMoE Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * Typo fix in docstring Co-Authored-By: ggerganov@gmail.com Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix(conversion): Simplify tensor name mapping in conversion Branch: GraniteMoE Co-Authored-By: git@compilade.net Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix(convert): Remove unused tensor name mappings Branch: GraniteMoE Co-Authored-By: git@compilade.net Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix(convert): Sanity check on merged FFN tensor sizes Branch: GraniteMoE Co-Authored-By: git@compilade.net Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix: Allow "output" layer in granite moe architecture (convert and cpp) Branch: GraniteMoE Co-Authored-By: git@compilade.net Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix(granite): Add missing 'output' tensor for Granite This is a fix for the previous `granite` architecture PR. Recent snapshots have included this (`lm_head.weights`) as part of the architecture Branch: GraniteMoE Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> --------- Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>	2024-09-25 10:06:52 +03:00
Dou Xinpeng	904837e0cb	cann: fix crash when llama-bench is running on multiple cann devices (#9627 )	2024-09-25 11:30:38 +08:00
Eric Zhang	70392f1f81	ggml : add AVX512DQ requirement for AVX512 builds (#9622 )	2024-09-24 11:03:21 +03:00
Georgi Gerganov	bb5f819975	sync : ggml	2024-09-24 11:01:18 +03:00
Georgi Gerganov	c038931615	examples : adapt to ggml.h changes (ggml/0) ggml-ci	2024-09-24 11:00:52 +03:00
Georgi Gerganov	31ac5834fe	llama : keep track of all EOG tokens in the vocab (#9609 ) ggml-ci	2024-09-24 10:16:06 +03:00
Georgi Gerganov	cea1486ecf	log : add CONT level for continuing previous log entry (#9610 )	2024-09-24 10:15:35 +03:00
StrangeBytesDev	0aa15011e3	server : add newline after chat example (#9616 )	2024-09-24 09:04:39 +03:00
Georgi Gerganov	b0f27361f3	sampling : avoid expensive softmax during greedy sampling (#9605 ) * sampling : avoid expensive softmax during greedy sampling ggml-ci * speculative : fix default RNG seed + set sparams.n_probs * Update tests/test-sampling.cpp Co-authored-by: slaren <slarengh@gmail.com> * sampling : add clarifying comment [no ci] --------- Co-authored-by: slaren <slarengh@gmail.com>	2024-09-24 09:03:17 +03:00
Max Krasnyansky	c087b6f11d	threads: fix msvc build without openmp (#9615 ) We're missing atomic_thread_fence() in MSVC builds when openmp is disabled.	2024-09-23 21:18:48 -07:00
Ivan	116efee0ee	cuda: add q8_0->f32 cpy operation (#9571 ) llama: enable K-shift for quantized KV cache It will fail on unsupported backends or quant types.	2024-09-24 02:14:24 +02:00
Xuan Son Nguyen	0b3bf966f4	server : add --no-context-shift option (#9607 ) * server : add --no-context-shift option * small fix * Update examples/server/tests/features/embeddings.feature Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> * tests : minor fix * revert usage of GGML_ASSERT * update server documentation --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>	2024-09-23 22:23:54 +02:00
Max Krasnyansky	f0c7b5edf8	threads: improve ggml_barrier scaling with large number of threads (#9598 ) Make sure n_barrier and n_barrier_passed do not share the cache line to avoid cache line bouncing. This optimization shows performance improvements even for n_threads <= 8 cases. Resurect TSAN (Thread Sanitizer) check so that we can avoid doing expensive read-modify-write in the normal case and just use thread-fence as originally intended. --- Here is the original description and suggestions from Willy Tarreau : There's currently some false sharing between n_barrier and n_barrier_passed that is amplified in ggml_barrier() by the fact that all threads need to increment n_barrier when entering, while all previous threads continue to read n_barrier_passed, waiting for the last one to release them all. The side effect is that all these readers are slowing down all new threads by making the cache line bounce back and forth between readers and writers. Just placing them in two distinct cache lines is sufficient to boost the performance by 21% on a 80-core ARM server compared to the no-openmp version, and by 3% compared to the openmp version. Note that the variables could have been spread apart in the structure as well, but it doesn't seem that the size of this threadpool struct is critical so here we're simply aligning them. Finally, the same issue was present when leaving the barrier since all threads had to update the n_barrier_passed counter, though only one would add a non-zero value. This alone is responsible for half of the cost due to undesired serialization. It might be possible that using a small array of n_barrier counters could make things even faster on many-core systems, but it would likely complicate the logic needed to detect the last thread. Co-authored-by: Willy Tarreau <w@1wt.eu>	2024-09-23 11:42:43 -07:00
Riceball LEE	1d48e98e4f	readme : add programmable prompt engine language CLI (#9599 )	2024-09-23 18:58:17 +03:00
Georgi Gerganov	f3979df762	flake.lock: Update (#9586 ) Flake lock file updates: • Updated input 'nixpkgs': 'github:NixOS/nixpkgs/4f807e8940284ad7925ebd0a0993d2a1791acb2f?narHash=sha256-IiA3jfbR7K/B5%2B9byVi9BZGWTD4VSbWe8VLpp9B/iYk%3D' (2024-09-11) → 'github:NixOS/nixpkgs/c04d5652cfa9742b1d519688f65d1bbccea9eb7e?narHash=sha256-PmUr/2GQGvFTIJ6/Tvsins7Q43KTMvMFhvG6oaYK%2BWk%3D' (2024-09-19) Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>	2024-09-23 08:43:40 -07:00
Srihari-mcw	1e7b9299c6	ggml : AVX512 gemm for Q4_0_8_8 (#9532 ) * AVX512 version of ggml_gemm_q4_0_8x8_q8_0 * Remove zero vector parameter passing * Rename functions and rearrange order of macros * Edit commments * style : minor adjustments * Update x to start from 0 --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>	2024-09-23 17:06:38 +03:00