Hexagon: Process M-tail rows on HMX instead of HVX (#22724)

* hex-mm: process m-tail rows on HMX instead of HVX

* hmx-mm: unroll and optimize padded activation loop

---------

Co-authored-by: Max Krasnyansky <maxk@qti.qualcomm.com>

.github/ISSUE_TEMPLATE/010-bug-compilation.yml

@@ -12,6 +12,8 @@ body:
         after recreating the CMake build directory and with `-DGGML_CCACHE=OFF`.
         If the compilation succeeds with ccache disabled you should be able to permanently fix the issue
         by clearing `~/.cache/ccache` (on Linux).
+
+        Please fill out this template yourself, copypasting language model outputs is [strictly prohibited](https://github.com/ggml-org/llama.cpp/blob/master/CONTRIBUTING.md#ai-usage-policy).
   - type: textarea
     id: commit
     attributes:

.github/ISSUE_TEMPLATE/011-bug-results.yml

@@ -1,5 +1,5 @@
 name: Bug (model use)
-description: Something goes wrong when using a model (in general, not specific to a single llama.cpp module).
+description: Something goes wrong when running a model (crashes, garbled outputs, etc.).
 title: "Eval bug: "
 labels: ["bug-unconfirmed", "model evaluation"]
 body:
@@ -12,6 +12,8 @@ body:
         If you encountered the issue while using an external UI (e.g. ollama),
         please reproduce your issue using one of the examples/binaries in this repository.
         The `llama-completion` binary can be used for simple and reproducible model inference.
+
+        Please fill out this template yourself, copypasting language model outputs is [strictly prohibited](https://github.com/ggml-org/llama.cpp/blob/master/CONTRIBUTING.md#ai-usage-policy).
   - type: textarea
     id: version
     attributes:

.github/ISSUE_TEMPLATE/019-bug-misc.yml

@@ -10,6 +10,8 @@ body:
         This issue template is intended for miscellaneous bugs that don't fit into any other category.
         If you encountered the issue while using an external UI (e.g. ollama),
         please reproduce your issue using one of the examples/binaries in this repository.
+
+        Please fill out this template yourself, copypasting language model outputs is [strictly prohibited](https://github.com/ggml-org/llama.cpp/blob/master/CONTRIBUTING.md#ai-usage-policy).
   - type: textarea
     id: version
     attributes:

.github/ISSUE_TEMPLATE/020-enhancement.yml

@@ -8,6 +8,8 @@ body:
       value: |
         [Please post your idea first in Discussion if there is not yet a consensus for this enhancement request. This will help to keep this issue tracker focused on enhancements that the community has agreed needs to be implemented.](https://github.com/ggml-org/llama.cpp/discussions/categories/ideas)
 
+        Please fill out this template yourself, copypasting language model outputs is [strictly prohibited](https://github.com/ggml-org/llama.cpp/blob/master/CONTRIBUTING.md#ai-usage-policy).
+
   - type: checkboxes
     id: prerequisites
     attributes:

.github/ISSUE_TEMPLATE/030-research.yml

@@ -8,6 +8,8 @@ body:
       value: |
         Don't forget to check for any [duplicate research issue tickets](https://github.com/ggml-org/llama.cpp/issues?q=is%3Aopen+is%3Aissue+label%3A%22research+%F0%9F%94%AC%22)
 
+        Please fill out this template yourself, copypasting language model outputs is [strictly prohibited](https://github.com/ggml-org/llama.cpp/blob/master/CONTRIBUTING.md#ai-usage-policy).
+
   - type: checkboxes
     id: research-stage
     attributes:

.github/ISSUE_TEMPLATE/040-refactor.yml

@@ -9,6 +9,8 @@ body:
         Don't forget to [check for existing refactor issue tickets](https://github.com/ggml-org/llama.cpp/issues?q=is%3Aopen+is%3Aissue+label%3Arefactoring) in case it's already covered.
         Also you may want to check [Pull request refactor label as well](https://github.com/ggml-org/llama.cpp/pulls?q=is%3Aopen+is%3Apr+label%3Arefactoring) for duplicates too.
 
+        Please fill out this template yourself, copypasting language model outputs is [strictly prohibited](https://github.com/ggml-org/llama.cpp/blob/master/CONTRIBUTING.md#ai-usage-policy).
+
   - type: textarea
     id: background-description
     attributes:

.pi/gg/SYSTEM.md

@@ -4,6 +4,7 @@ General:
 - By very precise and concise when writing code, comments, explanations, etc.
 - PR and commit titles format: `<module> : <title>`. Lookup recents for examples
 - Don't try to build or run the code unless you are explicitly asked to do so
+- Use the `gh` CLI tool when querying PRs, issues, or other GitHub resources
 
 Coding:
 - When in doubt, always refer to the CONTRIBUTING.md file of the project

common/arg.cpp

@@ -248,6 +248,8 @@ std::vector<std::string> common_arg::get_env() const {
 
 // Helper function to parse tensor buffer override strings
 static void parse_tensor_buffer_overrides(const std::string & value, std::vector<llama_model_tensor_buft_override> & overrides) {
+    ggml_backend_load_all();
+
     std::map<std::string, ggml_backend_buffer_type_t> buft_list;
     for (size_t i = 0; i < ggml_backend_dev_count(); ++i) {
         auto * dev = ggml_backend_dev_get(i);
@@ -425,6 +427,10 @@ static bool parse_bool_value(const std::string & value) {
     }
 }
 
+[[noreturn]] static void arg_removed(const std::string & msg) {
+    throw std::invalid_argument("the argument has been removed. " + msg);
+}
+
 //
 // CLI argument parsing functions
 //
@@ -803,6 +809,7 @@ static std::vector<ggml_backend_dev_t> parse_device_list(const std::string & val
     if (dev_names.size() == 1 && dev_names[0] == "none") {
         devices.push_back(nullptr);
     } else {
+        ggml_backend_load_all();
         for (const auto & device : dev_names) {
             auto * dev = ggml_backend_dev_by_name(device.c_str());
             if (!dev || ggml_backend_dev_type(dev) == GGML_BACKEND_DEVICE_TYPE_CPU) {
@@ -820,6 +827,7 @@ static void add_rpc_devices(const std::string & servers) {
     if (rpc_servers.empty()) {
         throw std::invalid_argument("no RPC servers specified");
     }
+    ggml_backend_load_all();
     ggml_backend_reg_t rpc_reg = ggml_backend_reg_by_name("RPC");
     if (!rpc_reg) {
         throw std::invalid_argument("failed to find RPC backend");
@@ -1016,9 +1024,6 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
 
     params.use_color = tty_can_use_colors();
 
-    // load dynamic backends
-    ggml_backend_load_all();
-
     common_params_context ctx_arg(params);
     ctx_arg.print_usage = print_usage;
     ctx_arg.ex          = ex;
@@ -2275,6 +2280,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
         {"--list-devices"},
         "print list of available devices and exit",
         [](common_params &) {
+            ggml_backend_load_all();
             std::vector<ggml_backend_dev_t> devices;
             for (size_t i = 0; i < ggml_backend_dev_count(); ++i) {
                 auto * dev = ggml_backend_dev_get(i);
@@ -2864,7 +2870,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
         {"--tools"}, "TOOL1,TOOL2,...",
         "experimental: whether to enable built-in tools for AI agents - do not enable in untrusted environments (default: no tools)\n"
         "specify \"all\" to enable all tools\n"
-        "available tools: read_file, file_glob_search, grep_search, exec_shell_command, write_file, edit_file, apply_diff",
+        "available tools: read_file, file_glob_search, grep_search, exec_shell_command, write_file, edit_file, apply_diff, get_datetime",
         [](common_params & params, const std::string & value) {
             params.server_tools = parse_csv_row(value);
         }
@@ -3380,7 +3386,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
     ).set_spec().set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_CLI}));
     add_opt(common_arg(
         {"--spec-draft-poll", "--poll-draft"}, "<0|1>",
-        "Use polling to wait for draft model work (default: same as --poll])",
+        "Use polling to wait for draft model work (default: same as --poll)",
         [](common_params & params, int value) {
             params.speculative.draft.cpuparams.poll = value;
         }
@@ -3715,35 +3721,35 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
         {"--draft", "--draft-n", "--draft-max"}, "N",
         "the argument has been removed. use --spec-draft-n-max or --spec-ngram-mod-n-max",
         [](common_params & /*params*/, int /*value*/) {
-            throw std::invalid_argument("the argument has been removed. use --spec-draft-n-max or --spec-ngram-mod-n-max");
+            arg_removed("use --spec-draft-n-max or --spec-ngram-mod-n-max");
         }
     ).set_spec().set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_LOOKUP, LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_CLI}).set_env("LLAMA_ARG_DRAFT_MAX"));
     add_opt(common_arg(
         {"--draft-min", "--draft-n-min"}, "N",
         "the argument has been removed. use --spec-draft-n-min or --spec-ngram-mod-n-min",
         [](common_params & /*params*/, int /*value*/) {
-            throw std::invalid_argument("the argument has been removed. use --spec-draft-n-min or --spec-ngram-mod-n-min");
+            arg_removed("use --spec-draft-n-min or --spec-ngram-mod-n-min");
         }
     ).set_spec().set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_LOOKUP, LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_CLI}).set_env("LLAMA_ARG_DRAFT_MIN"));
     add_opt(common_arg(
         {"--spec-ngram-size-n"}, "N",
         "the argument has been removed. use the respective --spec-ngram-*-size-n or --spec-ngram-mod-n-match",
         [](common_params & /*params*/, int /*value*/) {
-            throw std::invalid_argument("the argument has been removed. use the respective --spec-ngram-*-size-n");
+            arg_removed("use the respective --spec-ngram-*-size-n");
         }
     ).set_spec().set_examples({LLAMA_EXAMPLE_SERVER}));
     add_opt(common_arg(
         {"--spec-ngram-size-m"}, "N",
         "the argument has been removed. use the respective --spec-ngram-*-size-m",
         [](common_params & /*params*/, int /*value*/) {
-            throw std::invalid_argument("the argument has been removed. use the respective --spec-ngram-*-size-m");
+            arg_removed("use the respective --spec-ngram-*-size-m");
         }
     ).set_spec().set_examples({LLAMA_EXAMPLE_SERVER}));
     add_opt(common_arg(
         {"--spec-ngram-min-hits"}, "N",
         "the argument has been removed. use the respective --spec-ngram-*-min-hits",
         [](common_params & /*params*/, int /*value*/) {
-            throw std::invalid_argument("the argument has been removed. use the respective --spec-ngram-*-min-hits");
+            arg_removed("use the respective --spec-ngram-*-min-hits");
         }
     ).set_spec().set_examples({LLAMA_EXAMPLE_SERVER}));
 
@@ -3794,7 +3800,10 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
     ).set_examples({ LLAMA_EXAMPLE_DIFFUSION }));
     add_opt(common_arg(
         {"--diffusion-algorithm"}, "N",
-        string_format("diffusion algorithm: 0=ORIGIN, 1=ENTROPY_BASED, 2=MARGIN_BASED, 3=RANDOM, 4=LOW_CONFIDENCE (default: %d)", params.diffusion.algorithm),
+        string_format(
+            "diffusion algorithm: 0=DIFFUSION_ALGORITHM_ORIGIN, 1=DIFFUSION_ALGORITHM_ENTROPY_BASED, "
+            "2=DIFFUSION_ALGORITHM_MARGIN_BASED, 3=DIFFUSION_ALGORITHM_RANDOM, "
+            "4=DIFFUSION_ALGORITHM_CONFIDENCE_BASED (default: %d)", params.diffusion.algorithm),
         [](common_params & params, int value) { params.diffusion.algorithm = value; }
     ).set_examples({ LLAMA_EXAMPLE_DIFFUSION }));
     add_opt(common_arg(

common/chat-auto-parser-generator.cpp

@@ -136,10 +136,10 @@ common_peg_parser analyze_reasoning::build_parser(parser_build_context & ctx) co
         if (!end.empty()) {
             if (!start.empty()) {
                 // Standard tag-based: optional(<think>reasoning</think>)
-                return p.optional(start + p.reasoning(p.until(end)) + end + p.space());
+                return p.optional(p.optspace(start) + p.reasoning(p.until(trim_whitespace(end))) + p.optspace(end));
             }
             // Delimiter-style (empty start)
-            return p.optional(p.reasoning(p.until(end)) + end + p.space());
+            return p.optional(p.reasoning(p.until(trim_whitespace(end))) + p.optspace(end));
         }
     }
 
@@ -186,7 +186,6 @@ common_peg_parser analyze_tools::build_parser(parser_build_context & ctx) const
 common_peg_parser analyze_tools::build_tool_parser_json_native(parser_build_context & ctx) const {
     auto &       p           = ctx.p;
     const auto & inputs      = ctx.inputs;
-    bool         force_tools = inputs.tool_choice == COMMON_CHAT_TOOL_CHOICE_REQUIRED;
 
     // Build effective field names with dot notation if function_field is set
     std::string name_field = format.name_field;
@@ -225,8 +224,7 @@ common_peg_parser analyze_tools::build_tool_parser_json_native(parser_build_cont
         tool_start = format.per_call_start;
     }
 
-    return ctx.reasoning_parser + (force_tools ? p.eps() : p.optional(p.content(p.until(tool_start)))) + tools_parser +
-           p.end();
+    return ctx.reasoning_parser + p.optional(p.content(p.until(tool_start))) + tools_parser + p.end();
 }
 
 common_peg_parser analyze_tools::build_func_parser(common_chat_peg_builder & p, const std::string & name,
@@ -270,7 +268,6 @@ common_peg_parser analyze_tools::build_func_parser(common_chat_peg_builder & p,
 common_peg_parser analyze_tools::build_tool_parser_tag_json(parser_build_context & ctx) const {
     auto &       p           = ctx.p;
     const auto & inputs      = ctx.inputs;
-    bool         force_tools = inputs.tool_choice == COMMON_CHAT_TOOL_CHOICE_REQUIRED;
 
     common_peg_parser tool_choice = p.choice();
 
@@ -336,14 +333,12 @@ common_peg_parser analyze_tools::build_tool_parser_tag_json(parser_build_context
 
     std::string trigger_marker       = !format.section_start.empty() ? format.section_start : format.per_call_start;
     auto        content_before_tools = trigger_marker.empty() ? p.eps() : p.until(trigger_marker);
-    return ctx.reasoning_parser + (force_tools ? p.eps() : p.optional(p.content(content_before_tools))) + tool_calls +
-           p.end();
+    return ctx.reasoning_parser + p.optional(p.content(content_before_tools)) + tool_calls + p.end();
 }
 
 common_peg_parser analyze_tools::build_tool_parser_tag_tagged(parser_build_context & ctx) const {
     auto &       p           = ctx.p;
     const auto & inputs      = ctx.inputs;
-    bool         force_tools = inputs.tool_choice == COMMON_CHAT_TOOL_CHOICE_REQUIRED;
 
     auto until_suffix = p.rule("until-suffix", p.until(arguments.value_suffix));
 
@@ -471,8 +466,7 @@ common_peg_parser analyze_tools::build_tool_parser_tag_tagged(parser_build_conte
 
     std::string trigger_marker       = !format.section_start.empty() ? format.section_start : format.per_call_start;
     auto        content_before_tools = trigger_marker.empty() ? p.eps() : p.until(trigger_marker);
-    return ctx.reasoning_parser + (force_tools ? p.eps() : p.optional(p.content(content_before_tools))) + tool_calls +
-           p.end();
+    return ctx.reasoning_parser + p.optional(p.content(content_before_tools)) + tool_calls + p.end();
 }
 
 }  // namespace autoparser

common/chat-diff-analyzer.cpp

@@ -342,7 +342,7 @@ void analyze_reasoning::compare_thinking_enabled() {
     if (left_trimmed.empty() && !diff.right.empty()) {
         if (!right_trimmed.empty() && string_ends_with(comparison->output_B, right_trimmed)) {
             if (start.empty()) {
-                start = trim_leading_whitespace(diff.right);
+                start = diff.right;
                 mode  = reasoning_mode::TAG_BASED;
             }
         }
@@ -353,7 +353,7 @@ void analyze_reasoning::compare_thinking_enabled() {
                 if (seg.size() >= 2 && seg[seg.size() - 1].value == left_trimmed && seg[seg.size() - 2].type == segment_type::MARKER) {
                     start = seg[seg.size() - 2].value;
                 }
-                end = trim_trailing_whitespace(diff.left);
+                end = diff.left;
                 mode = reasoning_mode::TAG_BASED;
             }
         }
@@ -445,14 +445,14 @@ void analyze_reasoning::compare_reasoning_scope() {
         auto result = parser_wrapped.parse_anywhere_and_extract(comparison->output_B);
         if (result.result.success()) {
             start = result.tags["pre"];
-            end = trim_trailing_whitespace(result.tags["post"]);
+            end = result.tags["post"];
         } else {
             auto parser_delimiter = build_tagged_peg_parser([&](common_peg_parser_builder &p) {
                 return p.literal(reasoning_content) + p.space() + p.optional(p.tag("post", (p.marker() + p.space())));
             });
             result = parser_delimiter.parse_anywhere_and_extract(comparison->output_B);
             if (result.result.success()) {
-                end = trim_trailing_whitespace(result.tags["post"]);
+                end = result.tags["post"];
             } else {
                 LOG_DBG(ANSI_ORANGE "%s: Unable to extract reasoning markers, falling back to reasoning = NONE\n" ANSI_RESET, __func__);
                 mode = reasoning_mode::NONE;

common/chat-peg-parser.cpp

@@ -816,6 +816,32 @@ common_peg_parser common_chat_peg_builder::prefix(const std::string & s, const s
     return literal(s.substr(0, s.rfind(delimiter)));
 }
 
+common_peg_parser common_chat_peg_builder::optspace(const std::string & tag) {
+    auto parser = eps();
+    size_t end_of_prefix_space = tag.size();
+    size_t start_of_suffix_space = tag.size();
+    for (size_t i = 0; i < tag.size(); i++) {
+        if (!std::isspace(tag[i])) {
+            end_of_prefix_space = i;
+            break;
+        }
+    }
+    for (size_t i = tag.size(); i > 0; i--) {
+        if (!std::isspace(tag[i - 1])) {
+            start_of_suffix_space = i;
+            break;
+        }
+    }
+    for (size_t i = 0; i < end_of_prefix_space; i++) {
+        parser += optional(literal(std::string(1, tag[i])));
+    }
+    parser += literal(tag.substr(end_of_prefix_space, start_of_suffix_space - end_of_prefix_space));
+    for (size_t i = start_of_suffix_space; i < tag.size(); i++) {
+        parser += optional(literal(std::string(1, tag[i])));
+    }
+    return parser;
+}
+
 common_peg_parser common_chat_peg_builder::standard_json_tools(
                                                        const std::string &              section_start,
                                                        const std::string &              section_end,

common/chat-peg-parser.h

@@ -96,6 +96,9 @@ class common_chat_peg_builder : public common_peg_parser_builder {
     // Return a parser that parses the prefix of a string, up to a given delimiter.
     common_peg_parser prefix(const std::string & s, const std::string & delimiter = {});
 
+    // Return a parser that parses all elements of tag, but leading and trailing spaces are optional
+    common_peg_parser optspace(const std::string & tag);
+
     // Legacy-compatible helper for building standard JSON tool calls
     // Used by tests and manual parsers
     // name_key/args_key: JSON key names for function name and arguments

common/chat.cpp

@@ -2116,22 +2116,38 @@ std::optional<common_chat_params> common_chat_try_specialized_template(
     return std::nullopt;
 }
 
+static std::string common_chat_templates_generation_prompt(const common_chat_template & tmpl, const autoparser::generation_params & inputs) {
+    autoparser::generation_params params = inputs;
+    params.add_generation_prompt = false;
+    std::string no_gen_prompt    = common_chat_template_direct_apply_impl(tmpl, params);
+    params.add_generation_prompt = true;
+    std::string gen_prompt       = common_chat_template_direct_apply_impl(tmpl, params);
+
+    size_t prefix_len = 0;
+    size_t min_size = std::min(no_gen_prompt.size(), gen_prompt.size());
+    while (prefix_len < min_size && no_gen_prompt[prefix_len] == gen_prompt[prefix_len]) {
+        prefix_len++;
+    }
+    return gen_prompt.substr(prefix_len);
+}
+
 static common_chat_params common_chat_templates_apply_jinja(const struct common_chat_templates *        tmpls,
                                                             const struct common_chat_templates_inputs & inputs) {
     autoparser::generation_params params;
     params.tools = common_chat_tools_to_json_oaicompat(inputs.tools);
     const auto & tmpl =
         params.tools.is_array() && tmpls->template_tool_use ? *tmpls->template_tool_use : *tmpls->template_default;
-    const auto & src        = tmpl.source();
-    const auto & caps       = tmpl.original_caps();
-    params.messages         = render_message_to_json(inputs.messages, tmpl.original_caps());
-    params.tool_choice      = inputs.tool_choice;
-    params.reasoning_format = inputs.reasoning_format;
-    params.enable_thinking  = inputs.enable_thinking;
-    params.grammar          = inputs.grammar;
-    params.now              = inputs.now;
-    params.add_bos          = tmpls->add_bos;
-    params.add_eos          = tmpls->add_eos;
+    const auto & src             = tmpl.source();
+    const auto & caps            = tmpl.original_caps();
+    params.messages              = render_message_to_json(inputs.messages, tmpl.original_caps());
+    params.tool_choice           = inputs.tool_choice;
+    params.reasoning_format      = inputs.reasoning_format;
+    params.enable_thinking       = inputs.enable_thinking;
+    params.grammar               = inputs.grammar;
+    params.now                   = inputs.now;
+    params.add_generation_prompt = inputs.add_generation_prompt;
+    params.add_bos               = tmpls->add_bos;
+    params.add_eos               = tmpls->add_eos;
 
     if (src.find("<|channel|>") == std::string::npos) {
         // map developer to system for all models except for GPT-OSS
@@ -2153,14 +2169,7 @@ static common_chat_params common_chat_templates_apply_jinja(const struct common_
         workaround::func_args_not_string(params.messages);
     }
 
-    params.add_generation_prompt = false;
-    std::string no_gen_prompt    = common_chat_template_direct_apply_impl(tmpl, params);
-    params.add_generation_prompt = true;
-    std::string gen_prompt       = common_chat_template_direct_apply_impl(tmpl, params);
-    auto        diff             = calculate_diff_split(no_gen_prompt, gen_prompt);
-    params.generation_prompt     = diff.right + diff.suffix;
-
-    params.add_generation_prompt = inputs.add_generation_prompt;
+    params.generation_prompt = common_chat_templates_generation_prompt(tmpl, params);
 
     params.extra_context = common_chat_extra_context();
     for (auto el : inputs.chat_template_kwargs) {
@@ -2212,8 +2221,8 @@ static common_chat_params common_chat_templates_apply_jinja(const struct common_
         auto auto_params = autoparser::peg_generator::generate_parser(tmpl, params, autoparser);
         auto_params.supports_thinking = autoparser.reasoning.mode != autoparser::reasoning_mode::NONE;
         if (auto_params.supports_thinking) {
-            auto_params.thinking_start_tag = autoparser.reasoning.start;
-            auto_params.thinking_end_tag   = autoparser.reasoning.end;
+            auto_params.thinking_start_tag = trim_whitespace(autoparser.reasoning.start);
+            auto_params.thinking_end_tag   = trim_whitespace(autoparser.reasoning.end);
         }
         auto_params.generation_prompt = params.generation_prompt;
         common_peg_arena arena;

common/reasoning-budget.cpp

@@ -158,6 +158,8 @@ static void common_reasoning_budget_apply(struct llama_sampler * smpl, llama_tok
     for (size_t i = 0; i < cur_p->size; i++) {
         if (cur_p->data[i].id != forced) {
             cur_p->data[i].logit = -INFINITY;
+        } else {
+            cur_p->data[i].logit = +INFINITY; // force the token
         }
     }
 }

common/speculative.cpp

@@ -252,14 +252,14 @@ struct common_speculative_state_draft : public common_speculative_state {
 
     size_t create_checkpoint(int n_tokens_prompt) {
         int slot_id = 0;
-        const size_t checkpoint_size = llama_state_seq_get_size_ext(ctx_dft, slot_id, LLAMA_STATE_SEQ_FLAGS_PARTIAL_ONLY);
+        const size_t checkpoint_size = llama_state_seq_get_size_ext(ctx_dft, slot_id, LLAMA_STATE_SEQ_FLAGS_PARTIAL_ONLY | LLAMA_STATE_SEQ_FLAGS_ON_DEVICE);
 
         ckpt.pos_min  = llama_memory_seq_pos_min(llama_get_memory(ctx_dft), slot_id);
         ckpt.pos_max  = llama_memory_seq_pos_max(llama_get_memory(ctx_dft), slot_id);
         ckpt.n_tokens = n_tokens_prompt;
         ckpt.data.resize(checkpoint_size);
 
-        const size_t n = llama_state_seq_get_data_ext(ctx_dft, ckpt.data.data(), checkpoint_size, slot_id, LLAMA_STATE_SEQ_FLAGS_PARTIAL_ONLY);
+        const size_t n = llama_state_seq_get_data_ext(ctx_dft, ckpt.data.data(), checkpoint_size, slot_id, LLAMA_STATE_SEQ_FLAGS_PARTIAL_ONLY | LLAMA_STATE_SEQ_FLAGS_ON_DEVICE);
         if (n != checkpoint_size) {
             GGML_ABORT("checkpoint size mismatch: expected %zu, got %zu\n", checkpoint_size, n);
         }
@@ -272,7 +272,7 @@ struct common_speculative_state_draft : public common_speculative_state {
     size_t restore_checkpoint() {
         int slot_id = 0;
         LOG_DBG("%s: pos_min = %d, pos_max = %d\n", __func__, ckpt.pos_min, ckpt.pos_max);
-        const size_t n = llama_state_seq_set_data_ext(ctx_dft, ckpt.data.data(), ckpt.size(), slot_id, LLAMA_STATE_SEQ_FLAGS_PARTIAL_ONLY);
+        const size_t n = llama_state_seq_set_data_ext(ctx_dft, ckpt.data.data(), ckpt.size(), slot_id, LLAMA_STATE_SEQ_FLAGS_PARTIAL_ONLY | LLAMA_STATE_SEQ_FLAGS_ON_DEVICE);
         if (n != ckpt.size()) {
             GGML_ABORT("%s: failed to restore context checkpoint (pos_min=%d, pos_max=%d, size=%zu",
                         __func__, ckpt.pos_min, ckpt.pos_max, ckpt.size());

convert_hf_to_gguf.py

@@ -2889,6 +2889,20 @@ class LlamaModel(TextModel):
                 .swapaxes(1, 2)
                 .reshape(weights.shape))
 
+    def _repack_nvfp4(self, name: str, weight: Tensor, scale: Tensor, scale2: Tensor, input_scale: Tensor):
+        # Mirror the BF16 Q/K RoPE permutation site in modify_tensors; the NVFP4 path bypasses it.
+        if self.undo_permute:
+            n_head = self.find_hparam(["n_heads", "num_attention_heads"], optional=True)
+            n_kv_head = self.find_hparam(["n_kv_heads", "num_key_value_heads"], optional=True)
+            if n_head is not None:
+                if name.endswith("q_proj.weight"):
+                    weight = LlamaModel.permute(weight, n_head, n_head)
+                    scale  = LlamaModel.permute(scale, n_head, n_head)
+                elif name.endswith("k_proj.weight"):
+                    weight = LlamaModel.permute(weight, n_head, n_kv_head)
+                    scale  = LlamaModel.permute(scale, n_head, n_kv_head)
+        super()._repack_nvfp4(name, weight, scale, scale2, input_scale)
+
     _experts: list[dict[str, Tensor]] | None = None
 
     def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
@@ -12702,11 +12716,12 @@ class MistralModel(LlamaModel):
     def set_mistral_config(gguf_writer: gguf.GGUFWriter, hparams: dict):
         if "yarn" in hparams:
             yarn_params = hparams["yarn"]
+            mscale_all_dim = 1.0 if not yarn_params["apply_scale"] else 0.0
             gguf_writer.add_rope_scaling_type(gguf.RopeScalingType.YARN)
             gguf_writer.add_rope_scaling_factor(yarn_params["factor"])
             gguf_writer.add_rope_scaling_yarn_beta_fast(yarn_params["beta"])
             gguf_writer.add_rope_scaling_yarn_beta_slow(yarn_params["alpha"])
-            gguf_writer.add_rope_scaling_yarn_log_mul(1.0) # mscale_all_dim
+            gguf_writer.add_rope_scaling_yarn_log_mul(mscale_all_dim)
             gguf_writer.add_rope_scaling_orig_ctx_len(yarn_params["original_max_position_embeddings"])
 
         if "llama_4_scaling" in hparams:
@@ -13232,17 +13247,18 @@ class LazyTorchTensor(gguf.LazyBase):
     }
 
     # only used when byteswapping data. Only correct size is needed
+    # TODO: uncomment uint64, uint32, and uint16, ref: https://github.com/pytorch/pytorch/issues/58734
     _dtype_byteswap_map: dict[torch.dtype, type] = {
         torch.float64: np.float64,
         torch.float32: np.float32,
         torch.bfloat16: np.float16,
         torch.float16: np.float16,
         torch.int64: np.int64,
-        torch.uint64: np.uint64,
+        # torch.uint64: np.uint64,
         torch.int32: np.int32,
-        torch.uint32: np.uint32,
+        # torch.uint32: np.uint32,
         torch.int16: np.int16,
-        torch.uint16: np.uint16,
+        # torch.uint16: np.uint16,
         torch.int8: np.int8,
         torch.uint8: np.uint8,
         torch.bool: np.uint8,

docs/speculative.md

@@ -33,18 +33,18 @@ An example to use this approach can be the rewriting of source code by a LLM.
 This implementation looks for the last n-gram in history that matches the current n-gram and creates a draft using the m tokens following the matched n-gram. It is the simplest self-speculative approach with minimal overhead.
 
 ```
-llama-server [...] --spec-type ngram-simple --draft-max 64
+llama-server [...] --spec-type ngram-simple --spec-draft-n-max 64
 ```
 
 #### n-gram Map Key (`ngram-map-k`)
 
-This implementation looks for the current n-gram of size n (called the _key_) in the token history. If the key n-gram is followed by the same m tokens (called the _mgram_) multiple times, it creates a draft using these m tokens. This approach requires a minimum number of occurrences (argument `--spec-ngram-min-hits`, default is 1) before generating drafts.
+This implementation looks for the current n-gram of size n (called the _key_) in the token history. If the key n-gram is followed by the same m tokens (called the _mgram_) multiple times, it creates a draft using these m tokens. This approach requires a minimum number of occurrences (argument `--spec-ngram-map-k-min-hits`, default is 1) before generating drafts.
 
 The number of accepted tokens is stored for each used n-gram.
 
 **Example:**
 ```
-llama-server [...] --spec-type ngram-map-k --draft-max 64
+llama-server [...] --spec-type ngram-map-k --spec-draft-n-max 64
 ```
 
 #### n-gram Map Key-4-Values (`ngram-map-k4v`)
@@ -55,7 +55,7 @@ The number of accepted tokens is stored for each used n-gram.
 
 **Example:** Server options to be used if there are a lot of longer repetitions.
 ```
-llama-server [...] --spec-type ngram-map-k4v --spec-ngram-size-n 8 --spec-ngram-size-m 8 --spec-ngram-min-hits 2 --draft-max 64
+llama-server [...] --spec-type ngram-map-k4v --spec-ngram-map-k4v-size-n 8 --spec-ngram-map-k4v-size-m 8 --spec-ngram-map-k4v-min-hits 2 --spec-draft-n-max 64
 ```
 
 ### n-gram Mod (`ngram-mod`)
@@ -80,9 +80,9 @@ Currently, a single hash pool is shared across all server slots, so different re
 # notes:
 # - small `n` are not recommended
 # - MoEs require long drafts
-# - dense models: can reduce `--draft-min` and `--draft-max`
+# - dense models: can reduce `--spec-ngram-mod-n-min` and `--spec-ngram-mod-n-max`
 
-llama-server ... --spec-type ngram-mod --spec-ngram-size-n 24 --draft-min 48 --draft-max 64
+llama-server ... --spec-type ngram-mod --spec-ngram-mod-n-match 24 --spec-ngram-mod-n-min 48 --spec-ngram-mod-n-max 64
 ```
 
 Applications:
@@ -105,21 +105,90 @@ Example Video:
 
 If a draft model is combined with a draftless decoding the draftless decoding has higher precedence.
 
+### General Speculative Parameters
+
 ```
---draft, --draft-n, --draft-max N       number of tokens to draft for speculative decoding (default: 16)
-                                        (env: LLAMA_ARG_DRAFT_MAX)
---draft-min, --draft-n-min N            minimum number of draft tokens to use for speculative decoding
-                                        (default: 0)
-                                        (env: LLAMA_ARG_DRAFT_MIN)
-[...]
 --spec-type [none|ngram-cache|ngram-simple|ngram-map-k|ngram-map-k4v|ngram-mod]
                                         type of speculative decoding to use when no draft model is provided
                                         (default: none)
---spec-ngram-size-n N                   ngram size N for ngram-simple/ngram-map speculative decoding, length
-                                        of lookup n-gram (default: 12)
---spec-ngram-size-m N                   ngram size M for ngram-simple/ngram-map speculative decoding, length
-                                        of draft m-gram (default: 48)
---spec-ngram-min-hits N                 minimum hits for ngram-map speculative decoding (default: 1)
+                                        (env: LLAMA_ARG_SPEC_TYPE)
+--spec-default                          use default speculative decoding
+```
+
+### Draft Model Parameters
+
+```
+--spec-draft-model, -md, --model-draft  FNAME
+                                        draft model for speculative decoding (default: unused)
+                                        (env: LLAMA_ARG_SPEC_DRAFT_MODEL)
+--spec-draft-hf, -hfd, -hfrd, --hf-repo-draft  <user>/<model>[:quant]
+                                        HuggingFace repository for the draft model
+--spec-draft-n-max                      N
+                                        number of tokens to draft for speculative decoding (default: 16)
+                                        (env: LLAMA_ARG_SPEC_DRAFT_N_MAX)
+--spec-draft-n-min                      N
+                                        minimum number of draft tokens to use for speculative decoding (default: 0)
+                                        (env: LLAMA_ARG_SPEC_DRAFT_N_MIN)
+--spec-draft-p-split, --draft-p-split   P
+                                        speculative decoding split probability (default: 0.10)
+                                        (env: LLAMA_ARG_SPEC_DRAFT_P_SPLIT)
+--spec-draft-p-min, --draft-p-min       P
+                                        minimum speculative decoding probability (greedy) (default: 0.75)
+                                        (env: LLAMA_ARG_SPEC_DRAFT_P_MIN)
+--spec-draft-ctx-size, -cd, --ctx-size-draft  N
+                                        size of the prompt context for the draft model (default: 0, 0 = loaded from model)
+                                        (env: LLAMA_ARG_SPEC_DRAFT_CTX_SIZE)
+--spec-draft-ngl, -ngld, --gpu-layers-draft, --n-gpu-layers-draft  N
+                                        max. number of draft model layers to store in VRAM, either an exact number, 'auto', or 'all' (default: auto)
+                                        (env: LLAMA_ARG_N_GPU_LAYERS_DRAFT)
+--spec-draft-device, -devd, --device-draft  <dev1,dev2,..>
+                                        comma-separated list of devices to use for offloading the draft model
+--spec-draft-replace, --spec-replace    TARGET  DRAFT
+                                        translate the string in TARGET into DRAFT if the draft model and main model are not compatible
+```
+
+### n-gram Mod Parameters
+
+```
+--spec-ngram-mod-n-match                N
+                                        ngram-mod lookup length (default: 24)
+--spec-ngram-mod-n-min                  N
+                                        minimum number of ngram tokens to use for ngram-based speculative decoding (default: 48)
+--spec-ngram-mod-n-max                  N
+                                        maximum number of ngram tokens to use for ngram-based speculative decoding (default: 64)
+```
+
+### n-gram Simple Parameters
+
+```
+--spec-ngram-simple-size-n              N
+                                        ngram size N for ngram-simple speculative decoding, length of lookup n-gram (default: 12)
+--spec-ngram-simple-size-m              N
+                                        ngram size M for ngram-simple speculative decoding, length of draft m-gram (default: 48)
+--spec-ngram-simple-min-hits            N
+                                        minimum hits for ngram-simple speculative decoding (default: 1)
+```
+
+### n-gram Map Key Parameters
+
+```
+--spec-ngram-map-k-size-n               N
+                                        ngram size N for ngram-map-k speculative decoding, length of lookup n-gram (default: 12)
+--spec-ngram-map-k-size-m               N
+                                        ngram size M for ngram-map-k speculative decoding, length of draft m-gram (default: 48)
+--spec-ngram-map-k-min-hits             N
+                                        minimum hits for ngram-map-k speculative decoding (default: 1)
+```
+
+### n-gram Map Key-4-Values Parameters
+
+```
+--spec-ngram-map-k4v-size-n             N
+                                        ngram size N for ngram-map-k4v speculative decoding, length of lookup n-gram (default: 12)
+--spec-ngram-map-k4v-size-m             N
+                                        ngram size M for ngram-map-k4v speculative decoding, length of draft m-gram (default: 48)
+--spec-ngram-map-k4v-min-hits           N
+                                        minimum hits for ngram-map-k4v speculative decoding (default: 1)
 ```
 
 ### `--spec-type TYPE`
@@ -140,21 +209,40 @@ Specifies a type of speculative decoding without draft model.
 ./llama-server [...] --spec-type ngram-simple
 ```
 
-### `--spec-ngram-size-n N`
+### `--spec-ngram-*-size-n N`
 
 Sets the size N of the lookup n-gram for n-gram map based speculative decoding.
 The n-gram size N determines how many tokens in a row to look back when searching for matching patterns.
 
-### `--spec-ngram-size-m M`
+Each n-gram implementation has its own parameter:
+
+- `--spec-ngram-simple-size-n` for `ngram-simple`
+- `--spec-ngram-map-k-size-n` for `ngram-map-k`
+- `--spec-ngram-map-k4v-size-n` for `ngram-map-k4v`
+- `--spec-ngram-mod-n-match` for `ngram-mod`
+
+### `--spec-ngram-*-size-m M`
 
 Sets the size M of the draft m-gram for n-gram map based speculative decoding.
 The m-gram size determines how many tokens to draft when a match is found.
 Larger values can provide more speedup but may reduce acceptance rate.
 
-### `--spec-ngram-min-hits H`
+Each n-gram implementation has its own parameter:
+
+- `--spec-ngram-simple-size-m` for `ngram-simple`
+- `--spec-ngram-map-k-size-m` for `ngram-map-k`
+- `--spec-ngram-map-k4v-size-m` for `ngram-map-k4v`
+
+### `--spec-ngram-*-min-hits H`
 
 This option defines how often a key has to appear in the token history to be used as a draft (default is 1).
 
+Each n-gram implementation has its own parameter:
+
+- `--spec-ngram-simple-min-hits` for `ngram-simple`
+- `--spec-ngram-map-k-min-hits` for `ngram-map-k`
+- `--spec-ngram-map-k4v-min-hits` for `ngram-map-k4v`
+
 ## Statistics
 Each speculative decoding implementation prints statistics.
 
@@ -180,4 +268,3 @@ statistics ngram_map_k: #calls(b,g,a) = 6 1690 26, #gen drafts = 26, #acc drafts
 - `#gen tokens`: number of tokens generated by this implementation (including rejected tokens)
 - `#acc tokens`: number of tokens accepted by the main model
 - `dur(b,g,a): durations of begin (new prompt), generation and accumulation (process acceptance).
-

examples/diffusion/CMakeLists.txt

@@ -1,5 +1,10 @@
+set(TARGET llama-diffusion)
+add_library(${TARGET} STATIC diffusion.cpp diffusion.h)
+target_link_libraries(${TARGET} PUBLIC llama llama-common ${CMAKE_THREAD_LIBS_INIT})
+target_compile_features(${TARGET} PUBLIC cxx_std_17)
+
 set(TARGET llama-diffusion-cli)
 add_executable(${TARGET} diffusion-cli.cpp)
 install(TARGETS ${TARGET} RUNTIME)
-target_link_libraries(${TARGET} PRIVATE llama llama-common ${CMAKE_THREAD_LIBS_INIT})
+target_link_libraries(${TARGET} PRIVATE llama-diffusion llama llama-common ${CMAKE_THREAD_LIBS_INIT})
 target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/diffusion/README.md

@@ -12,11 +12,11 @@ The diffusion CLI supports various parameters to control the generation process:
 ### Core Diffusion Parameters
 - `--diffusion-steps`: Number of diffusion steps (default: 256)
 - `--diffusion-algorithm`: Algorithm for token selection
-  - `0`: ORIGIN - Token will be generated in a purely random order from https://arxiv.org/abs/2107.03006.
-  - `1`: ENTROPY_BASED - Entropy-based selection
-  - `2`: MARGIN_BASED - Margin-based selection
-  - `3`: RANDOM - Random selection
-  - `4`: CONFIDENCE_BASED - Confidence-based selection (default)
+  - `0`: DIFFUSION_ALGORITHM_ORIGIN - Token will be generated in a purely random order from https://arxiv.org/abs/2107.03006.
+  - `1`: DIFFUSION_ALGORITHM_ENTROPY_BASED - Entropy-based selection
+  - `2`: DIFFUSION_ALGORITHM_MARGIN_BASED - Margin-based selection
+  - `3`: DIFFUSION_ALGORITHM_RANDOM - Random selection
+  - `4`: DIFFUSION_ALGORITHM_CONFIDENCE_BASED - Confidence-based selection (default)
   - More documentation here https://github.com/DreamLM/Dream
 - `--diffusion-visual`: Enable live visualization during generation
 

examples/diffusion/diffusion-cli.cpp

@@ -1,127 +1,23 @@
 #include "arg.h"
 #include "chat.h"
 #include "common.h"
+#include "diffusion.h"
 #include "llama.h"
 #include "log.h"
 
 #include <limits.h>
 
-#include <algorithm>
 #include <clocale>
-#include <cmath>
 #include <cstring>
-#include <limits>
-#include <random>
 #include <string>
 #include <vector>
 
-enum diffusion_algorithm { ORIGIN = 0, ENTROPY_BASED = 1, MARGIN_BASED = 2, RANDOM = 3, CONFIDENCE_BASED = 4 };
-
-// Unified transfer scheduling methods
-enum transfer_schedule {
-    TIMESTEP_BASED = 0,  // Dream-style: (1.0 - s/t) * remaining
-    BLOCK_BASED    = 1,  // LLaDA-style: process in blocks with get_num_transfer_tokens
-};
-
-typedef bool (*diffusion_step_callback_t)(int32_t             step,
-                                          int32_t             total_steps,
-                                          const llama_token * tokens,
-                                          int32_t             n_tokens,
-                                          void *              user_data);
-
-struct diffusion_params {
-    int32_t                   steps                   = 0;
-    float                     temperature             = 0;
-    llama_token               mask_token_id           = LLAMA_TOKEN_NULL;
-    diffusion_step_callback_t step_callback           = nullptr;
-    void *                    step_callback_user_data = nullptr;
-    int32_t                   seed                    = 0;
-    bool                      visual_mode             = false;
-    bool                      shift_logits            = false;  // Shift logits by -1 after decode
-
-    float   top_p = 0.;
-    int32_t top_k = 0.;
-
-    diffusion_algorithm algorithm = CONFIDENCE_BASED;
-    transfer_schedule   schedule  = TIMESTEP_BASED;
-
-    float   cfg_scale        = 0.;     // Config scale for classifier-free guidance
-    float   eps              = 0.;     // Timestep scheduling
-    int32_t block_length     = 0;      // Block size (for block scheduling)
-    float   alg_temp         = 0;      // algorithm temperature (0.0 = deterministic)
-    bool    add_gumbel_noise = false;  // Add gumbel noise to the logits if temp > 0.0
-
-    int32_t max_length = 0;            // Maximum sequence length
-};
-
 struct callback_data {
     diffusion_params *  diff_params;
     const llama_vocab * vocab;
     int32_t             n_input;
 };
 
-static float calculate_confidence(const llama_token_data_array & cur_p,
-                                  diffusion_algorithm            algorithm,
-                                  std::mt19937 &                 rng) {
-    switch (algorithm) {
-        case CONFIDENCE_BASED:
-            return cur_p.data[cur_p.selected].p;  // Selected token probability
-
-        case ENTROPY_BASED:
-            {
-                float       entropy = 0.0f;
-                const float epsilon = 1e-10f;
-                for (size_t i = 0; i < cur_p.size; i++) {
-                    float prob = cur_p.data[i].p;
-                    entropy += prob * logf(prob + epsilon);
-                }
-                return -entropy;  // Higher entropy = lower confidence
-            }
-
-        case MARGIN_BASED:
-            return (cur_p.size > 1) ? cur_p.data[0].p - cur_p.data[1].p : cur_p.data[0].p;
-
-        case RANDOM:
-            {
-                std::uniform_real_distribution<float> uniform(0.0f, 1.0f);
-                return uniform(rng);  // Random confidence
-            }
-
-        case ORIGIN:
-            return cur_p.data[cur_p.selected].p;
-
-        default:
-            return 0.0f;
-    }
-}
-
-// Unified transfer count calculation function
-static int32_t calculate_transfer_count(int32_t                      step,
-                                        int32_t                      total_steps,
-                                        int32_t                      remaining_masked,
-                                        transfer_schedule            schedule,
-                                        float                        eps,
-                                        const std::vector<int32_t> & num_transfer_tokens = {}) {
-    switch (schedule) {
-        case TIMESTEP_BASED:
-            {
-                float t          = 1.0f - (float) step / total_steps * (1.0f - eps);
-                float s          = 1.0f - (float) (step + 1) / total_steps * (1.0f - eps);
-                float p_transfer = (step < total_steps - 1) ? (1.0f - s / t) : 1.0f;
-                return (int32_t) (remaining_masked * p_transfer);
-            }
-
-        case BLOCK_BASED:
-            if (!num_transfer_tokens.empty() && step < (int32_t) num_transfer_tokens.size()) {
-                return num_transfer_tokens[step];
-            }
-            return remaining_masked / (total_steps - step);  // Fallback
-
-        default:
-            return remaining_masked / (total_steps - step);
-    }
-}
-
 static bool diffusion_step_callback(int32_t             step,
                                     int32_t             total_steps,
                                     const llama_token * tokens,
@@ -176,341 +72,6 @@ static bool diffusion_step_callback(int32_t             step,
     return true;
 }
 
-static void add_gumbel_noise(float * logits, int32_t n_vocab, float temperature, std::mt19937 & rng) {
-    if (temperature == 0.0f) {
-        return;
-    }
-
-    std::uniform_real_distribution<double> uniform(0.0, 1.0);
-    for (int32_t i = 0; i < n_vocab; i++) {
-        double noise        = uniform(rng);
-        // Prevent log(0)
-        noise               = std::max(noise, 1e-20);
-        double gumbel_noise = std::pow(-std::log(noise), temperature);
-        logits[i]           = std::exp(logits[i]) / gumbel_noise;
-    }
-}
-
-static std::vector<int32_t> get_num_transfer_tokens(int32_t mask_count, int32_t steps) {
-    std::vector<int32_t> num_transfer_tokens(steps);
-
-    int32_t base      = mask_count / steps;
-    int32_t remainder = mask_count % steps;
-
-    for (int32_t i = 0; i < steps; i++) {
-        num_transfer_tokens[i] = base + (i < remainder ? 1 : 0);
-    }
-
-    return num_transfer_tokens;
-}
-
-static void diffusion_generate(llama_context *          ctx,
-                               const llama_token *      input_tokens,
-                               llama_token *            output_tokens,
-                               int32_t                  n_input,
-                               const diffusion_params & params,
-                               int32_t &                n_generated) {
-    n_generated = 0;
-    if (!ctx || !input_tokens || !output_tokens || n_input <= 0 || params.max_length <= n_input) {
-        return;
-    }
-
-    const llama_model * model = llama_get_model(ctx);
-
-    // Initialize with input and pad with mask tokens
-    std::copy(input_tokens, input_tokens + n_input, output_tokens);
-    std::fill(output_tokens + n_input, output_tokens + params.max_length, params.mask_token_id);
-
-    std::mt19937 rng(params.seed);
-
-    llama_set_causal_attn(ctx, false);
-
-    int32_t n_vocab = llama_vocab_n_tokens(llama_model_get_vocab(model));
-
-    std::vector<llama_token_data> candidates(n_vocab);
-    std::vector<llama_token_data> conf_candidates;
-    conf_candidates.reserve(params.max_length);
-    std::vector<int32_t> mask_positions;
-    mask_positions.reserve(params.max_length);
-
-    // Setup sampler chain
-    struct llama_sampler * sampler = llama_sampler_chain_init(llama_sampler_chain_default_params());
-    if (params.top_k > 0) {
-        llama_sampler_chain_add(sampler, llama_sampler_init_top_k(params.top_k));
-    }
-    if (params.top_p < 1.0f) {
-        llama_sampler_chain_add(sampler, llama_sampler_init_top_p(params.top_p, 1));
-    }
-    if (params.temperature > 0.0f) {
-        llama_sampler_chain_add(sampler, llama_sampler_init_temp(params.temperature));
-    }
-    llama_sampler_chain_add(sampler, llama_sampler_init_dist(params.seed));
-
-    struct llama_sampler * dist_sampler = llama_sampler_init_dist(params.seed);
-
-    llama_batch batch = llama_batch_init(params.max_length, 0, 1);
-    batch.n_tokens    = params.max_length;
-
-    // Pre-allocate buffers for CFG if needed
-    int32_t                  logits_size = n_vocab * params.max_length;
-    std::vector<float>       cond_logits_buffer;
-    std::vector<llama_token> un_x_buffer;
-    if (params.cfg_scale > 0.0f) {
-        cond_logits_buffer.resize(logits_size);
-        un_x_buffer.resize(params.max_length);
-    }
-
-    // For block-based processing
-    std::vector<int32_t> num_transfer_tokens;
-    int32_t              num_blocks      = 1;
-    int32_t              steps_per_block = params.steps;
-
-    if (params.schedule == BLOCK_BASED) {
-        GGML_ASSERT(params.max_length % params.block_length == 0);
-        num_blocks = params.max_length / params.block_length;
-        GGML_ASSERT(params.steps % num_blocks == 0);
-        steps_per_block = params.steps / num_blocks;
-    }
-
-    std::vector<float> confidence(params.max_length);
-
-    int64_t total_sampling_time = 0;
-    int64_t total_time          = 0;
-    int64_t time_start          = ggml_time_us();
-
-    for (int block_num = 0; block_num < num_blocks; block_num++) {
-        int32_t block_start = (params.schedule == BLOCK_BASED) ? n_input + block_num * params.block_length : 0;
-        int32_t block_end   = (params.schedule == BLOCK_BASED) ?
-                                  std::min(n_input + (block_num + 1) * params.block_length, params.max_length) :
-                                  params.max_length;
-
-        // Count masked tokens in current block for block-based processing
-        if (params.schedule == BLOCK_BASED) {
-            int32_t block_mask_count = 0;
-            for (int i = block_start; i < block_end; i++) {
-                if (output_tokens[i] == params.mask_token_id) {
-                    block_mask_count++;
-                }
-            }
-            num_transfer_tokens = get_num_transfer_tokens(block_mask_count, steps_per_block);
-        }
-
-        for (int32_t step = 0; step < steps_per_block; step++) {
-            int32_t global_step = block_num * steps_per_block + step;
-
-            if (params.step_callback) {
-                if (!params.step_callback(
-                        global_step, params.steps, output_tokens, params.max_length, params.step_callback_user_data)) {
-                    break;
-                }
-            }
-
-            // Setup batch
-            for (int32_t i = 0; i < params.max_length; i++) {
-                batch.token[i]     = output_tokens[i];
-                batch.pos[i]       = i;
-                batch.n_seq_id[i]  = 1;
-                batch.seq_id[i][0] = 0;
-                batch.logits[i]    = 1;
-            }
-
-            float * logits = nullptr;
-
-            if (params.cfg_scale > 0.0f) {
-                int ret = llama_decode(ctx, batch);
-                if (ret != 0) {
-                    LOG_ERR("Failed to generate conditional");
-                    break;
-                }
-                float * cond_logits_ptr = llama_get_logits(ctx);
-                std::memcpy(cond_logits_buffer.data(), cond_logits_ptr, logits_size * sizeof(float));
-
-                // Unconditional generation (mask input)
-                std::copy(output_tokens, output_tokens + params.max_length, un_x_buffer.begin());
-                for (int32_t i = 0; i < n_input; i++) {
-                    un_x_buffer[i] = params.mask_token_id;
-                }
-
-                for (int32_t i = 0; i < params.max_length; i++) {
-                    batch.token[i] = un_x_buffer[i];
-                }
-                ret = llama_decode(ctx, batch);
-                if (ret != 0) {
-                    LOG_ERR("Failed to generate unconditional");
-                    break;
-                }
-                float * uncond_logits = llama_get_logits(ctx);
-
-                // Apply CFG
-                for (int32_t i = 0; i < logits_size; i++) {
-                    cond_logits_buffer[i] =
-                        uncond_logits[i] + (params.cfg_scale + 1.0f) * (cond_logits_buffer[i] - uncond_logits[i]);
-                }
-                logits = cond_logits_buffer.data();
-            } else {
-                int ret = llama_decode(ctx, batch);
-                if (ret != 0) {
-                    LOG_ERR("%s: failed to decode at step %d, ret = %d\n", __func__, global_step, ret);
-                    break;
-                }
-                logits = llama_get_logits(ctx);
-            }
-
-            if (!logits) {
-                LOG_ERR("%s: failed to get logits at step %d\n", __func__, global_step);
-                break;
-            }
-
-            auto get_logits_for_pos = [&](int32_t pos) -> const float * {
-                if (params.shift_logits) {
-                    return pos == 0 ? logits : logits + (pos - 1) * n_vocab;
-                }
-                return logits + (pos) *n_vocab;
-            };
-
-            int64_t time_start_sampling = ggml_time_us();
-
-            mask_positions.clear();
-            for (int32_t i = 0; i < params.max_length; i++) {
-                if (output_tokens[i] == params.mask_token_id) {
-                    // For block-based, only consider current block
-                    if (params.schedule != BLOCK_BASED || (i >= block_start && i < block_end)) {
-                        mask_positions.push_back(i);
-                    }
-                }
-            }
-
-            if (mask_positions.empty()) {
-                break;
-            }
-
-            if (params.add_gumbel_noise && params.temperature > 0.0f) {
-                add_gumbel_noise(logits, n_vocab, params.temperature, rng);
-            }
-
-            if (params.algorithm == ORIGIN) {
-                int32_t transfer_count = calculate_transfer_count(
-                    step, steps_per_block, mask_positions.size(), params.schedule, params.eps, num_transfer_tokens);
-                float p_transfer = (float) transfer_count / mask_positions.size();
-
-                for (int32_t pos : mask_positions) {
-                    if (std::uniform_real_distribution<float>(0.0f, 1.0f)(rng) < p_transfer) {
-                        const float * pos_logits = get_logits_for_pos(pos);
-                        for (int32_t token_id = 0; token_id < n_vocab; token_id++) {
-                            candidates[token_id].id    = token_id;
-                            candidates[token_id].logit = pos_logits[token_id];
-                            candidates[token_id].p     = 0.0f;
-                        }
-
-                        llama_token_data_array cur_p = {
-                            candidates.data(),
-                            (size_t) n_vocab,
-                            -1,
-                            false,
-                        };
-
-                        llama_sampler_apply(sampler, &cur_p);
-                        output_tokens[pos] = cur_p.data[cur_p.selected].id;
-                    }
-                }
-            } else {
-                std::vector<std::pair<float, int32_t>> confidences;
-                std::vector<llama_token>               sampled_tokens(mask_positions.size());
-
-                for (size_t i = 0; i < mask_positions.size(); i++) {
-                    int32_t       pos        = mask_positions[i];
-                    const float * pos_logits = get_logits_for_pos(pos);
-
-                    for (int32_t token_id = 0; token_id < n_vocab; token_id++) {
-                        candidates[token_id].logit = pos_logits[token_id];
-                        candidates[token_id].p     = 0.0f;
-                        candidates[token_id].id    = token_id;
-                    }
-
-                    llama_token_data_array cur_p = {
-                        candidates.data(),
-                        candidates.size(),
-                        -1,
-                        false,
-                    };
-
-                    llama_sampler_apply(sampler, &cur_p);
-                    llama_token sampled_token = cur_p.data[cur_p.selected].id;
-
-                    float conf = calculate_confidence(cur_p, params.algorithm, rng);
-
-                    sampled_tokens[i] = sampled_token;
-                    confidences.emplace_back(conf, i);
-                }
-
-                int32_t transfer_count = calculate_transfer_count(
-                    step, steps_per_block, mask_positions.size(), params.schedule, params.eps, num_transfer_tokens);
-
-                if (transfer_count > 0) {
-                    if (params.alg_temp == 0.0f) {
-                        std::partial_sort(confidences.begin(),
-                                          confidences.begin() + std::min(transfer_count, (int32_t) confidences.size()),
-                                          confidences.end(),
-                                          [](const std::pair<float, int32_t> & a, const std::pair<float, int32_t> & b) {
-                                              if (a.first != b.first) {
-                                                  return a.first > b.first;
-                                              }
-                                              return a.second < b.second;
-                                          });
-
-                        for (int32_t i = 0; i < std::min(transfer_count, (int32_t) confidences.size()); i++) {
-                            int32_t mask_idx   = confidences[i].second;
-                            int32_t pos        = mask_positions[mask_idx];
-                            output_tokens[pos] = sampled_tokens[mask_idx];
-                        }
-                    } else {
-                        conf_candidates.clear();
-                        for (size_t i = 0; i < confidences.size(); i++) {
-                            float conf_logit = confidences[i].first / params.alg_temp;
-                            conf_candidates.emplace_back(llama_token_data{ (int32_t) i, conf_logit, 0.0f });
-                        }
-
-                        llama_token_data_array conf_array = {
-                            conf_candidates.data(),
-                            conf_candidates.size(),
-                            -1,
-                            false,
-                        };
-
-                        for (int32_t i = 0; i < std::min(transfer_count, (int32_t) confidences.size()); i++) {
-                            llama_sampler_apply(dist_sampler, &conf_array);
-                            int32_t selected_idx = conf_array.selected;
-                            int32_t mask_idx     = selected_idx;
-                            int32_t pos          = mask_positions[mask_idx];
-                            output_tokens[pos]   = sampled_tokens[mask_idx];
-
-                            conf_candidates[selected_idx].p = 0.0f;
-                            conf_array.selected             = -1;
-                        }
-                    }
-                }
-            }
-
-            int64_t time_end_sampling = ggml_time_us();
-            total_sampling_time += time_end_sampling - time_start_sampling;
-        }
-    }
-
-    int64_t time_end = ggml_time_us();
-    total_time += time_end - time_start;
-
-    LOG_INF("\ntotal time: %0.2fms, time per step: %0.2fms, sampling time per step: %0.2fms\n",
-            total_time / 1000.0,
-            total_time / 1000.0 / params.steps,
-            total_sampling_time / 1000.0 / params.steps);
-
-    llama_batch_free(batch);
-    llama_sampler_free(sampler);
-    llama_sampler_free(dist_sampler);
-
-    n_generated = params.max_length;
-}
-
 static std::string format_input_text(const std::string & prompt, const std::string & system_prompt, bool use_chat_template, llama_model * model) {
     if (!use_chat_template) {
         return prompt;
@@ -631,10 +192,10 @@ int main(int argc, char ** argv) {
     GGML_ASSERT((params.diffusion.eps == 0) ^ (params.diffusion.block_length == 0));
 
     if (params.diffusion.eps) {
-        diff_params.schedule = TIMESTEP_BASED;
+        diff_params.schedule = DIFFUSION_TRANSFER_SCHEDULE_TIMESTEP_BASED;
         diff_params.eps      = params.diffusion.eps;
     } else if (params.diffusion.block_length) {
-        diff_params.schedule     = BLOCK_BASED;
+        diff_params.schedule     = DIFFUSION_TRANSFER_SCHEDULE_BLOCK_BASED;
         diff_params.block_length = params.diffusion.block_length;
     }
 
@@ -653,8 +214,17 @@ int main(int argc, char ** argv) {
     callback_data cb_data               = { &diff_params, vocab, n_input };
     diff_params.step_callback_user_data = &cb_data;
 
-    const char * alg_names[]   = { "ORIGIN", "ENTROPY_BASED", "MARGIN_BASED", "RANDOM", "CONFIDENCE_BASED" };
-    const char * sched_names[] = { "TIMESTEP_BASED", "BLOCK_BASED" };
+    const char * alg_names[]   = {
+        "DIFFUSION_ALGORITHM_ORIGIN",
+        "DIFFUSION_ALGORITHM_ENTROPY_BASED",
+        "DIFFUSION_ALGORITHM_MARGIN_BASED",
+        "DIFFUSION_ALGORITHM_RANDOM",
+        "DIFFUSION_ALGORITHM_CONFIDENCE_BASED",
+    };
+    const char * sched_names[] = {
+        "DIFFUSION_TRANSFER_SCHEDULE_TIMESTEP_BASED",
+        "DIFFUSION_TRANSFER_SCHEDULE_BLOCK_BASED",
+    };
     const char * alg_name =
         (diff_params.algorithm >= 0 && diff_params.algorithm <= 4) ? alg_names[diff_params.algorithm] : "UNKNOWN";
     const char * sched_name =
@@ -666,11 +236,11 @@ int main(int argc, char ** argv) {
     LOG_INF("diffusion_params: - %-25s enum             = %d (%s)\n", "algorithm", diff_params.algorithm, alg_name);
     LOG_INF("diffusion_params: - %-25s enum             = %d (%s)\n", "schedule", diff_params.schedule, sched_name);
     LOG_INF("diffusion_params: - %-25s f32              = %.3f\n", "temperature", diff_params.temperature);
-    if (diff_params.schedule == TIMESTEP_BASED) {
+    if (diff_params.schedule == DIFFUSION_TRANSFER_SCHEDULE_TIMESTEP_BASED) {
         LOG_INF("diffusion_params: - %-25s f32              = %.6f\n", "eps", diff_params.eps);
         LOG_INF("diffusion_params: - %-25s f32              = %.3f\n", "alg_temp", diff_params.alg_temp);
     }
-    if (diff_params.schedule == BLOCK_BASED) {
+    if (diff_params.schedule == DIFFUSION_TRANSFER_SCHEDULE_BLOCK_BASED) {
         LOG_INF("diffusion_params: - %-25s u32              = %d\n", "block_length", diff_params.block_length);
         LOG_INF("diffusion_params: - %-25s f32              = %.3f\n", "cfg_scale", diff_params.cfg_scale);
     }

examples/diffusion/diffusion.cpp

@@ -0,0 +1,408 @@
+#include "diffusion.h"
+
+#include "log.h"
+
+#include <algorithm>
+#include <cstddef>
+#include <cmath>
+#include <cstring>
+#include <random>
+#include <utility>
+#include <vector>
+
+static float calculate_confidence(const llama_token_data_array & cur_p,
+                                  diffusion_algorithm            algorithm,
+                                  std::mt19937 &                 rng) {
+    switch (algorithm) {
+        case DIFFUSION_ALGORITHM_CONFIDENCE_BASED:
+            return cur_p.data[cur_p.selected].p;  // Selected token probability
+
+        case DIFFUSION_ALGORITHM_ENTROPY_BASED:
+            {
+                float       entropy = 0.0f;
+                const float epsilon = 1e-10f;
+                for (size_t i = 0; i < cur_p.size; i++) {
+                    float prob = cur_p.data[i].p;
+                    entropy += prob * logf(prob + epsilon);
+                }
+                return -entropy;  // Higher entropy = lower confidence
+            }
+
+        case DIFFUSION_ALGORITHM_MARGIN_BASED:
+            return (cur_p.size > 1) ? cur_p.data[0].p - cur_p.data[1].p : cur_p.data[0].p;
+
+        case DIFFUSION_ALGORITHM_RANDOM:
+            {
+                std::uniform_real_distribution<float> uniform(0.0f, 1.0f);
+                return uniform(rng);  // Random confidence
+            }
+
+        case DIFFUSION_ALGORITHM_ORIGIN:
+            return cur_p.data[cur_p.selected].p;
+
+        default:
+            return 0.0f;
+    }
+}
+
+// Unified transfer count calculation function
+static int32_t calculate_transfer_count(int32_t                      step,
+                                        int32_t                      total_steps,
+                                        int32_t                      remaining_masked,
+                                        diffusion_transfer_schedule  schedule,
+                                        float                        eps,
+                                        const std::vector<int32_t> & num_transfer_tokens = {}) {
+    switch (schedule) {
+        case DIFFUSION_TRANSFER_SCHEDULE_TIMESTEP_BASED:
+            {
+                float t          = 1.0f - (float) step / total_steps * (1.0f - eps);
+                float s          = 1.0f - (float) (step + 1) / total_steps * (1.0f - eps);
+                float p_transfer = (step < total_steps - 1) ? (1.0f - s / t) : 1.0f;
+                return (int32_t) (remaining_masked * p_transfer);
+            }
+
+        case DIFFUSION_TRANSFER_SCHEDULE_BLOCK_BASED:
+            if (!num_transfer_tokens.empty() && step < (int32_t) num_transfer_tokens.size()) {
+                return num_transfer_tokens[step];
+            }
+            return remaining_masked / (total_steps - step);  // Fallback
+
+        default:
+            return remaining_masked / (total_steps - step);
+    }
+}
+
+static void add_gumbel_noise(float * logits, int32_t n_vocab, float temperature, std::mt19937 & rng) {
+    if (temperature == 0.0f) {
+        return;
+    }
+
+    std::uniform_real_distribution<double> uniform(0.0, 1.0);
+    for (int32_t i = 0; i < n_vocab; i++) {
+        double noise        = uniform(rng);
+        // Prevent log(0)
+        noise               = std::max(noise, 1e-20);
+        double gumbel_noise = std::pow(-std::log(noise), temperature);
+        logits[i]           = std::exp(logits[i]) / gumbel_noise;
+    }
+}
+
+static std::vector<int32_t> get_num_transfer_tokens(int32_t mask_count, int32_t steps) {
+    std::vector<int32_t> num_transfer_tokens(steps);
+
+    int32_t base      = mask_count / steps;
+    int32_t remainder = mask_count % steps;
+
+    for (int32_t i = 0; i < steps; i++) {
+        num_transfer_tokens[i] = base + (i < remainder ? 1 : 0);
+    }
+
+    return num_transfer_tokens;
+}
+
+void diffusion_generate(llama_context *          ctx,
+                        const llama_token *      input_tokens,
+                        llama_token *            output_tokens,
+                        int32_t                  n_input,
+                        const diffusion_params & params,
+                        int32_t &                n_generated) {
+    n_generated = 0;
+    if (!ctx || !input_tokens || !output_tokens || n_input <= 0 || params.max_length <= n_input) {
+        return;
+    }
+
+    const llama_model * model = llama_get_model(ctx);
+
+    // Initialize with input and pad with mask tokens
+    std::copy(input_tokens, input_tokens + n_input, output_tokens);
+    std::fill(output_tokens + n_input, output_tokens + params.max_length, params.mask_token_id);
+
+    std::mt19937 rng(params.seed);
+
+    llama_set_causal_attn(ctx, false);
+
+    int32_t n_vocab = llama_vocab_n_tokens(llama_model_get_vocab(model));
+
+    std::vector<llama_token_data> candidates(n_vocab);
+    std::vector<llama_token_data> conf_candidates;
+    conf_candidates.reserve(params.max_length);
+    std::vector<int32_t> mask_positions;
+    mask_positions.reserve(params.max_length);
+
+    // Setup sampler chain
+    struct llama_sampler * sampler = llama_sampler_chain_init(llama_sampler_chain_default_params());
+    if (params.top_k > 0) {
+        llama_sampler_chain_add(sampler, llama_sampler_init_top_k(params.top_k));
+    }
+    if (params.top_p < 1.0f) {
+        llama_sampler_chain_add(sampler, llama_sampler_init_top_p(params.top_p, 1));
+    }
+    if (params.temperature > 0.0f) {
+        llama_sampler_chain_add(sampler, llama_sampler_init_temp(params.temperature));
+    }
+    llama_sampler_chain_add(sampler, llama_sampler_init_dist(params.seed));
+
+    struct llama_sampler * dist_sampler = llama_sampler_init_dist(params.seed);
+
+    llama_batch batch = llama_batch_init(params.max_length, 0, 1);
+    batch.n_tokens    = params.max_length;
+
+    // Pre-allocate buffers for CFG if needed
+    int32_t                  logits_size = n_vocab * params.max_length;
+    std::vector<float>       cond_logits_buffer;
+    std::vector<llama_token> un_x_buffer;
+    if (params.cfg_scale > 0.0f) {
+        cond_logits_buffer.resize(logits_size);
+        un_x_buffer.resize(params.max_length);
+    }
+
+    // For block-based processing
+    std::vector<int32_t> num_transfer_tokens;
+    int32_t              num_blocks      = 1;
+    int32_t              steps_per_block = params.steps;
+
+    if (params.schedule == DIFFUSION_TRANSFER_SCHEDULE_BLOCK_BASED) {
+        GGML_ASSERT(params.max_length % params.block_length == 0);
+        num_blocks = params.max_length / params.block_length;
+        GGML_ASSERT(params.steps % num_blocks == 0);
+        steps_per_block = params.steps / num_blocks;
+    }
+
+    std::vector<float> confidence(params.max_length);
+
+    int64_t total_sampling_time = 0;
+    int64_t total_time          = 0;
+    int64_t time_start          = ggml_time_us();
+
+    for (int block_num = 0; block_num < num_blocks; block_num++) {
+        int32_t block_start = (params.schedule == DIFFUSION_TRANSFER_SCHEDULE_BLOCK_BASED) ? n_input + block_num * params.block_length : 0;
+        int32_t block_end   = (params.schedule == DIFFUSION_TRANSFER_SCHEDULE_BLOCK_BASED) ?
+                                  std::min(n_input + (block_num + 1) * params.block_length, params.max_length) :
+                                  params.max_length;
+
+        // Count masked tokens in current block for block-based processing
+        if (params.schedule == DIFFUSION_TRANSFER_SCHEDULE_BLOCK_BASED) {
+            int32_t block_mask_count = 0;
+            for (int i = block_start; i < block_end; i++) {
+                if (output_tokens[i] == params.mask_token_id) {
+                    block_mask_count++;
+                }
+            }
+            num_transfer_tokens = get_num_transfer_tokens(block_mask_count, steps_per_block);
+        }
+
+        for (int32_t step = 0; step < steps_per_block; step++) {
+            int32_t global_step = block_num * steps_per_block + step;
+
+            if (params.step_callback) {
+                if (!params.step_callback(
+                        global_step, params.steps, output_tokens, params.max_length, params.step_callback_user_data)) {
+                    break;
+                }
+            }
+
+            // Setup batch
+            for (int32_t i = 0; i < params.max_length; i++) {
+                batch.token[i]     = output_tokens[i];
+                batch.pos[i]       = i;
+                batch.n_seq_id[i]  = 1;
+                batch.seq_id[i][0] = 0;
+                batch.logits[i]    = 1;
+            }
+
+            float * logits = nullptr;
+
+            if (params.cfg_scale > 0.0f) {
+                int ret = llama_decode(ctx, batch);
+                if (ret != 0) {
+                    LOG_ERR("Failed to generate conditional");
+                    break;
+                }
+                float * cond_logits_ptr = llama_get_logits(ctx);
+                std::memcpy(cond_logits_buffer.data(), cond_logits_ptr, logits_size * sizeof(float));
+
+                // Unconditional generation (mask input)
+                std::copy(output_tokens, output_tokens + params.max_length, un_x_buffer.begin());
+                for (int32_t i = 0; i < n_input; i++) {
+                    un_x_buffer[i] = params.mask_token_id;
+                }
+
+                for (int32_t i = 0; i < params.max_length; i++) {
+                    batch.token[i] = un_x_buffer[i];
+                }
+                ret = llama_decode(ctx, batch);
+                if (ret != 0) {
+                    LOG_ERR("Failed to generate unconditional");
+                    break;
+                }
+                float * uncond_logits = llama_get_logits(ctx);
+
+                // Apply CFG
+                for (int32_t i = 0; i < logits_size; i++) {
+                    cond_logits_buffer[i] =
+                        uncond_logits[i] + (params.cfg_scale + 1.0f) * (cond_logits_buffer[i] - uncond_logits[i]);
+                }
+                logits = cond_logits_buffer.data();
+            } else {
+                int ret = llama_decode(ctx, batch);
+                if (ret != 0) {
+                    LOG_ERR("%s: failed to decode at step %d, ret = %d\n", __func__, global_step, ret);
+                    break;
+                }
+                logits = llama_get_logits(ctx);
+            }
+
+            if (!logits) {
+                LOG_ERR("%s: failed to get logits at step %d\n", __func__, global_step);
+                break;
+            }
+
+            auto get_logits_for_pos = [&](int32_t pos) -> const float * {
+                if (params.shift_logits) {
+                    return pos == 0 ? logits : logits + (pos - 1) * n_vocab;
+                }
+                return logits + pos * n_vocab;
+            };
+
+            int64_t time_start_sampling = ggml_time_us();
+
+            mask_positions.clear();
+            for (int32_t i = 0; i < params.max_length; i++) {
+                if (output_tokens[i] == params.mask_token_id) {
+                    // For block-based, only consider current block
+                    if (params.schedule != DIFFUSION_TRANSFER_SCHEDULE_BLOCK_BASED || (i >= block_start && i < block_end)) {
+                        mask_positions.push_back(i);
+                    }
+                }
+            }
+
+            if (mask_positions.empty()) {
+                break;
+            }
+
+            if (params.add_gumbel_noise && params.temperature > 0.0f) {
+                add_gumbel_noise(logits, n_vocab, params.temperature, rng);
+            }
+
+            if (params.algorithm == DIFFUSION_ALGORITHM_ORIGIN) {
+                int32_t transfer_count = calculate_transfer_count(
+                    step, steps_per_block, mask_positions.size(), params.schedule, params.eps, num_transfer_tokens);
+                float p_transfer = (float) transfer_count / mask_positions.size();
+
+                for (int32_t pos : mask_positions) {
+                    if (std::uniform_real_distribution<float>(0.0f, 1.0f)(rng) < p_transfer) {
+                        const float * pos_logits = get_logits_for_pos(pos);
+                        for (int32_t token_id = 0; token_id < n_vocab; token_id++) {
+                            candidates[token_id].id    = token_id;
+                            candidates[token_id].logit = pos_logits[token_id];
+                            candidates[token_id].p     = 0.0f;
+                        }
+
+                        llama_token_data_array cur_p = {
+                            candidates.data(),
+                            (size_t) n_vocab,
+                            -1,
+                            false,
+                        };
+
+                        llama_sampler_apply(sampler, &cur_p);
+                        output_tokens[pos] = cur_p.data[cur_p.selected].id;
+                    }
+                }
+            } else {
+                std::vector<std::pair<float, int32_t>> confidences;
+                std::vector<llama_token>               sampled_tokens(mask_positions.size());
+
+                for (size_t i = 0; i < mask_positions.size(); i++) {
+                    int32_t       pos        = mask_positions[i];
+                    const float * pos_logits = get_logits_for_pos(pos);
+
+                    for (int32_t token_id = 0; token_id < n_vocab; token_id++) {
+                        candidates[token_id].logit = pos_logits[token_id];
+                        candidates[token_id].p     = 0.0f;
+                        candidates[token_id].id    = token_id;
+                    }
+
+                    llama_token_data_array cur_p = {
+                        candidates.data(),
+                        candidates.size(),
+                        -1,
+                        false,
+                    };
+
+                    llama_sampler_apply(sampler, &cur_p);
+                    llama_token sampled_token = cur_p.data[cur_p.selected].id;
+
+                    float conf = calculate_confidence(cur_p, params.algorithm, rng);
+
+                    sampled_tokens[i] = sampled_token;
+                    confidences.emplace_back(conf, i);
+                }
+
+                int32_t transfer_count = calculate_transfer_count(
+                    step, steps_per_block, mask_positions.size(), params.schedule, params.eps, num_transfer_tokens);
+
+                if (transfer_count > 0) {
+                    if (params.alg_temp == 0.0f) {
+                        std::partial_sort(confidences.begin(),
+                                          confidences.begin() + std::min(transfer_count, (int32_t) confidences.size()),
+                                          confidences.end(),
+                                          [](const std::pair<float, int32_t> & a, const std::pair<float, int32_t> & b) {
+                                              if (a.first != b.first) {
+                                                  return a.first > b.first;
+                                              }
+                                              return a.second < b.second;
+                                          });
+
+                        for (int32_t i = 0; i < std::min(transfer_count, (int32_t) confidences.size()); i++) {
+                            int32_t mask_idx   = confidences[i].second;
+                            int32_t pos        = mask_positions[mask_idx];
+                            output_tokens[pos] = sampled_tokens[mask_idx];
+                        }
+                    } else {
+                        conf_candidates.clear();
+                        for (size_t i = 0; i < confidences.size(); i++) {
+                            float conf_logit = confidences[i].first / params.alg_temp;
+                            conf_candidates.emplace_back(llama_token_data{ (int32_t) i, conf_logit, 0.0f });
+                        }
+
+                        llama_token_data_array conf_array = {
+                            conf_candidates.data(),
+                            conf_candidates.size(),
+                            -1,
+                            false,
+                        };
+
+                        for (int32_t i = 0; i < std::min(transfer_count, (int32_t) confidences.size()); i++) {
+                            llama_sampler_apply(dist_sampler, &conf_array);
+                            int32_t selected_idx = conf_array.selected;
+                            int32_t mask_idx     = selected_idx;
+                            int32_t pos          = mask_positions[mask_idx];
+                            output_tokens[pos]   = sampled_tokens[mask_idx];
+
+                            conf_candidates[selected_idx].p = 0.0f;
+                            conf_array.selected             = -1;
+                        }
+                    }
+                }
+            }
+
+            int64_t time_end_sampling = ggml_time_us();
+            total_sampling_time += time_end_sampling - time_start_sampling;
+        }
+    }
+
+    int64_t time_end = ggml_time_us();
+    total_time += time_end - time_start;
+
+    LOG_INF("\ntotal time: %0.2fms, time per step: %0.2fms, sampling time per step: %0.2fms\n",
+            total_time / 1000.0,
+            total_time / 1000.0 / params.steps,
+            total_sampling_time / 1000.0 / params.steps);
+
+    llama_batch_free(batch);
+    llama_sampler_free(sampler);
+    llama_sampler_free(dist_sampler);
+
+    n_generated = params.max_length;
+}

examples/diffusion/diffusion.h

@@ -0,0 +1,57 @@
+#pragma once
+
+#include "llama.h"
+
+#include <cstdint>
+
+enum diffusion_algorithm {
+    DIFFUSION_ALGORITHM_ORIGIN           = 0,
+    DIFFUSION_ALGORITHM_ENTROPY_BASED    = 1,
+    DIFFUSION_ALGORITHM_MARGIN_BASED     = 2,
+    DIFFUSION_ALGORITHM_RANDOM           = 3,
+    DIFFUSION_ALGORITHM_CONFIDENCE_BASED = 4,
+};
+
+// Unified transfer scheduling methods
+enum diffusion_transfer_schedule {
+    DIFFUSION_TRANSFER_SCHEDULE_TIMESTEP_BASED = 0,  // Dream-style: (1.0 - s/t) * remaining
+    DIFFUSION_TRANSFER_SCHEDULE_BLOCK_BASED    = 1,  // LLaDA-style: process in blocks with get_num_transfer_tokens
+};
+
+typedef bool (*diffusion_step_callback_t)(int32_t             step,
+                                          int32_t             total_steps,
+                                          const llama_token * tokens,
+                                          int32_t             n_tokens,
+                                          void *              user_data);
+
+struct diffusion_params {
+    int32_t                   steps                   = 0;
+    float                     temperature             = 0;
+    llama_token               mask_token_id           = LLAMA_TOKEN_NULL;
+    diffusion_step_callback_t step_callback           = nullptr;
+    void *                    step_callback_user_data = nullptr;
+    int32_t                   seed                    = 0;
+    bool                      visual_mode             = false;
+    bool                      shift_logits            = false;  // Shift logits by -1 after decode
+
+    float   top_p = 0.;
+    int32_t top_k = 0.;
+
+    diffusion_algorithm         algorithm = DIFFUSION_ALGORITHM_CONFIDENCE_BASED;
+    diffusion_transfer_schedule schedule  = DIFFUSION_TRANSFER_SCHEDULE_TIMESTEP_BASED;
+
+    float   cfg_scale        = 0.;     // Config scale for classifier-free guidance
+    float   eps              = 0.;     // Timestep scheduling
+    int32_t block_length     = 0;      // Block size (for block scheduling)
+    float   alg_temp         = 0;      // algorithm temperature (0.0 = deterministic)
+    bool    add_gumbel_noise = false;  // Add gumbel noise to the logits if temp > 0.0
+
+    int32_t max_length = 0;            // Maximum sequence length
+};
+
+void diffusion_generate(llama_context *          ctx,
+                        const llama_token *      input_tokens,
+                        llama_token *            output_tokens,
+                        int32_t                  n_input,
+                        const diffusion_params & params,
+                        int32_t &                n_generated);

examples/save-load-state/save-load-state.cpp

@@ -38,8 +38,12 @@ int main(int argc, char ** argv) {
     std::string result0;
     std::string result1;
     std::string result2;
+    std::string result3;
 
     // init
+
+    ggml_backend_load_all();
+
     auto llama_init = common_init_from_params(params);
 
     auto * model = llama_init->model();
@@ -213,11 +217,83 @@ int main(int argc, char ** argv) {
         n_past += 1;
     }
 
+    // test on-device state save/load
+    auto params_ctx4 = common_context_params_to_llama(params);
+    params_ctx4.n_seq_max = 2;
+    llama_context * ctx4 = llama_init_from_model(model, params_ctx4);
+
+    llama_sampler * smpl4 = llama_sampler_chain_init(sparams);
+
+    llama_sampler_chain_add(smpl4, llama_sampler_init_dist(params.sampling.seed));
+
+    printf("\nsingle seq run: %s", params.prompt.c_str());
+
+    // load state (rng, logits, embedding and kv_cache) from file
+    n_token_count_out = 0;
+
+    if (!llama_state_load_file(ctx4, state_file.data(), unused_sts.data(), unused_sts.size(), &n_token_count_out)) {
+        fprintf(stderr, "\n%s : failed to load state\n", __func__);
+        return 1;
+    }
+
+    fprintf(stderr, "%s : loaded state with %zu tokens\n", __func__, n_token_count_out);
+
+    // restore state (last tokens)
+    n_past = n_token_count_out;
+    if (!common_replay_last_token(ctx4, tokens.back(), n_past)) {
+        return 1;
+    }
+    ++n_past;
+
+    // save seq 0 and load into seq 1
+    {
+        // save kv of seq 0
+        std::vector<uint8_t> seq_store(llama_state_seq_get_size_ext(ctx4, 0, LLAMA_STATE_SEQ_FLAGS_ON_DEVICE));
+        const size_t ncopy = llama_state_seq_get_data_ext(ctx4, seq_store.data(), seq_store.size(), 0, LLAMA_STATE_SEQ_FLAGS_ON_DEVICE);
+        if (ncopy != seq_store.size()) {
+            fprintf(stderr, "\n%s : seq copy data length %zd does not match expected length %zd\n", __func__, ncopy, seq_store.size());
+            return 1;
+        }
+        fprintf(stderr, "%s : seq 0 copied, %zd bytes\n", __func__, ncopy);
+
+        // erase whole kv
+        llama_memory_clear(llama_get_memory(ctx4), true);
+        fprintf(stderr, "%s : kv cache cleared\n", __func__);
+
+        // restore kv into seq 0
+        const size_t nset = llama_state_seq_set_data_ext(ctx4, seq_store.data(), seq_store.size(), 1, LLAMA_STATE_SEQ_FLAGS_ON_DEVICE);
+        if (nset != seq_store.size()) {
+            fprintf(stderr, "\n%s : seq set data length %zd does not match expected length %zd\n", __func__, nset, seq_store.size());
+            return 1;
+        }
+        fprintf(stderr, "%s : seq 1 restored, %zd bytes\n", __func__, nset);
+    }
+
+    // forth run
+    for (auto i = 0; i < params.n_predict; i++) {
+        auto next_token     = llama_sampler_sample(smpl4, ctx4, -1);
+        auto next_token_str = common_token_to_piece(ctx4, next_token);
+
+        printf("%s", next_token_str.c_str());
+        result3 += next_token_str;
+
+        common_batch_clear(batch);
+        common_batch_add(batch, next_token, n_past, {1}, true);
+
+        if (llama_decode(ctx4, batch)) {
+            fprintf(stderr, "\n%s : failed to evaluate\n", __func__);
+            llama_batch_free(batch);
+            return 1;
+        }
+        n_past += 1;
+    }
+
     printf("\n");
 
     llama_sampler_free(smpl);
     llama_sampler_free(smpl2);
     llama_sampler_free(smpl3);
+    llama_sampler_free(smpl4);
 
     llama_batch_free(batch);
 
@@ -226,12 +302,18 @@ int main(int argc, char ** argv) {
 
     llama_free(ctx2);
     llama_free(ctx3);
+    llama_free(ctx4);
 
     if (result0 != result2) {
         fprintf(stderr, "\n%s : error : the seq restore generation is different\n", __func__);
         return 1;
     }
 
+    if (result0 != result3) {
+        fprintf(stderr, "\n%s : error : the seq restore generation is different\n", __func__);
+        return 1;
+    }
+
     fprintf(stderr, "\n%s : success\n", __func__);
 
     return 0;

ggml/CMakeLists.txt

@@ -4,8 +4,8 @@ project("ggml" C CXX ASM)
 
 ### GGML Version
 set(GGML_VERSION_MAJOR 0)
-set(GGML_VERSION_MINOR 10)
-set(GGML_VERSION_PATCH 1)
+set(GGML_VERSION_MINOR 11)
+set(GGML_VERSION_PATCH 0)
 set(GGML_VERSION_BASE "${GGML_VERSION_MAJOR}.${GGML_VERSION_MINOR}.${GGML_VERSION_PATCH}")
 
 list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/cmake/")

ggml/include/ggml.h

@@ -438,6 +438,12 @@ extern "C" {
         GGML_PREC_F32     = 10,
     };
 
+    // op hint
+    enum ggml_op_hint {
+        GGML_HINT_NONE             = 0,
+        GGML_HINT_SRC0_IS_HADAMARD = 1,
+    };
+
     // model file types
     enum ggml_ftype {
         GGML_FTYPE_UNKNOWN        = -1,
@@ -1419,6 +1425,11 @@ extern "C" {
             struct ggml_tensor * a,
             enum ggml_prec       prec);
 
+    // change the hint of a matrix multiplication
+    GGML_API void ggml_mul_mat_set_hint(
+            struct ggml_tensor * a,
+            enum ggml_op_hint    hint);
+
     // indirect matrix multiplication
     GGML_API struct ggml_tensor * ggml_mul_mat_id(
             struct ggml_context * ctx,

ggml/src/ggml-cpu/CMakeLists.txt

@@ -578,13 +578,13 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
 
         # Fetch KleidiAI sources:
         include(FetchContent)
-        set(KLEIDIAI_COMMIT_TAG "v1.22.0")
-        set(KLEIDIAI_DOWNLOAD_URL "https://github.com/ARM-software/kleidiai/archive/refs/tags/${KLEIDIAI_COMMIT_TAG}.tar.gz")
-        set(KLEIDIAI_ARCHIVE_MD5  "54049037570ab0ee0a0d126b2ba5ece1")
+        set(KLEIDIAI_COMMIT_TAG "v1.24.0")
+        set(KLEIDIAI_DOWNLOAD_URL "https://github.com/ARM-software/kleidiai/releases/download/${KLEIDIAI_COMMIT_TAG}/kleidiai-${KLEIDIAI_COMMIT_TAG}-src.tar.gz")
+        set(KLEIDIAI_RELEASE_ARCHIVE_MD5  "2f02ebe29573d45813e671eb304f2a00")
 
         set(KLEIDIAI_FETCH_ARGS
             URL ${KLEIDIAI_DOWNLOAD_URL}
-            URL_HASH MD5=${KLEIDIAI_ARCHIVE_MD5}
+            URL_HASH MD5=${KLEIDIAI_RELEASE_ARCHIVE_MD5}
         )
         if (CMAKE_VERSION VERSION_GREATER_EQUAL "3.24")
             list(APPEND KLEIDIAI_FETCH_ARGS DOWNLOAD_EXTRACT_TIMESTAMP NEW)

ggml/src/ggml-cpu/ggml-cpu.c

@@ -1245,6 +1245,12 @@ void ggml_compute_forward_mul_mat(
     const struct ggml_tensor * src0 = dst->src[0];
     const struct ggml_tensor * src1 = dst->src[1];
 
+    const int32_t hint = ggml_get_op_params_i32(dst, 1);
+    if (hint == GGML_HINT_SRC0_IS_HADAMARD && !params->use_ref) {
+        ggml_compute_forward_fwht(params, dst);
+        return;
+    }
+
     GGML_TENSOR_BINARY_OP_LOCALS
 
     const int ith = params->ith;

ggml/src/ggml-cpu/ops.cpp

@@ -11212,3 +11212,91 @@ void ggml_compute_forward_opt_step_sgd(const ggml_compute_params * params, ggml_
             }
     }
 }
+
+static void ggml_compute_forward_fwht_f32(const ggml_compute_params * params, ggml_tensor * dst) {
+    const ggml_tensor * src0 = dst->src[0];
+    const ggml_tensor * src1 = dst->src[1];
+
+    GGML_ASSERT(src1->type == GGML_TYPE_F32);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32);
+
+    GGML_TENSOR_BINARY_OP_LOCALS
+
+    const int ith = params->ith;
+    const int nth = params->nth;
+
+    const int64_t n = ne10;
+    GGML_ASSERT((n & (n - 1)) == 0); // must be power of 2
+
+    const int64_t nr = ne11 * ne12 * ne13;
+    const int64_t rows_per_thread = (nr + nth - 1) / nth;
+    const int64_t start_row = ith * rows_per_thread;
+    const int64_t end_row = MIN(start_row + rows_per_thread, nr);
+
+    const float scale = 1.0f / sqrtf((float)n);
+
+#if defined(GGML_SIMD)
+    const GGML_F32_VEC v_minus_one = GGML_F32_VEC_SET1(-1.0f);
+#endif
+
+    for (int64_t r = start_row; r < end_row; r++) {
+        const int64_t i13 = r / (ne11 * ne12);
+        const int64_t i12 = (r - i13 * ne11 * ne12) / ne11;
+        const int64_t i11 = r - i13 * ne11 * ne12 - i12 * ne11;
+
+        const float * src_row = (const float *) ((const char *) src1->data + i11 * nb11 + i12 * nb12 + i13 * nb13);
+        float * dst_row = (float *) ((char *) dst->data + i11 * nb1 + i12 * nb2 + i13 * nb3);
+
+        for (int64_t j = 0; j < n; j++) {
+            dst_row[j] = src_row[j] * scale;
+        }
+
+        // Scalar passes
+#if defined(GGML_SIMD)
+        const int step = GGML_F32_EPR;
+#else
+        const int step = n;
+#endif
+        for (int64_t len = 1; len < step && len < n; len <<= 1) {
+            for (int64_t i = 0; i < n; i += 2 * len) {
+                for (int64_t j = 0; j < len; j++) {
+                    float u = dst_row[i + j];
+                    float v = dst_row[i + len + j];
+                    dst_row[i + j] = u + v;
+                    dst_row[i + len + j] = u - v;
+                }
+            }
+        }
+
+        // SIMD passes using GGML_F32_VEC_* macros for multi-architecture support
+#if defined(GGML_SIMD)
+        for (int64_t len = step; len < n; len <<= 1) {
+            for (int64_t i = 0; i < n; i += 2 * len) {
+                for (int64_t j = 0; j < len; j += step) {
+                    GGML_F32_VEC u = GGML_F32_VEC_LOAD(dst_row + i + j);
+                    GGML_F32_VEC v = GGML_F32_VEC_LOAD(dst_row + i + len + j);
+
+                    GGML_F32_VEC_STORE(dst_row + i + j,       GGML_F32_VEC_ADD(u, v));
+                    GGML_F32_VEC_STORE(dst_row + i + len + j, GGML_F32_VEC_FMA(u, v, v_minus_one));
+                }
+            }
+        }
+#endif
+    }
+}
+
+void ggml_compute_forward_fwht(const ggml_compute_params * params, ggml_tensor * dst) {
+    const ggml_tensor * src1 = dst->src[1];
+
+    switch (src1->type) {
+        case GGML_TYPE_F32:
+            {
+                ggml_compute_forward_fwht_f32(params, dst);
+            }
+            break;
+        default:
+            {
+                GGML_ABORT("fatal error - fwht is F32 only");
+            }
+    }
+}

ggml/src/ggml-cpu/ops.h

@@ -111,6 +111,7 @@ void ggml_compute_forward_cross_entropy_loss(const struct ggml_compute_params *
 void ggml_compute_forward_cross_entropy_loss_back(const struct ggml_compute_params * params, struct ggml_tensor * dst);
 void ggml_compute_forward_opt_step_adamw(const struct ggml_compute_params * params, struct ggml_tensor * dst);
 void ggml_compute_forward_mul_mat(const struct ggml_compute_params * params, struct ggml_tensor * dst);
+void ggml_compute_forward_fwht(const struct ggml_compute_params * params, struct ggml_tensor * dst);
 void ggml_compute_forward_opt_step_sgd(const struct ggml_compute_params * params, struct ggml_tensor * dst);
 #ifdef __cplusplus
 }

ggml/src/ggml-cuda/getrows.cu

@@ -6,17 +6,18 @@ template<int qk, int qr, dequantize_kernel_t dequantize_kernel, typename dst_t>
 static __global__ void k_get_rows(
         const void * __restrict__ src0, const int32_t * __restrict__ src1, dst_t * __restrict__ dst,
         const int64_t ne00, /*const int64_t ne01, const int64_t ne02, const int64_t ne03,*/
-        /*const int64_t ne10,*/ const int64_t ne11, const int64_t ne12, /*const int64_t ne13,*/
+        /*const int64_t ne10,*/ const int64_t ne11, const uint3 ne12_fdv, /*const int64_t ne13,*/
         /*const size_t s0,*/ const size_t s1, const size_t s2, const size_t s3,
         /*const size_t nb00,*/ const size_t nb01, const size_t nb02, const size_t nb03,
         const size_t s10, const size_t s11, const size_t s12/*, const size_t s13*/) {
 
-    for (int64_t z = blockIdx.z; z < ne11*ne12; z += gridDim.z) {
+    for (int64_t z = blockIdx.z; z < ne11*(int64_t)ne12_fdv.z; z += gridDim.z) {
         for (int64_t i00 = 2*(blockIdx.y*blockDim.x + threadIdx.x); i00 < ne00; i00 += gridDim.y*blockDim.x) {
             // The x and y dimensions of the grid are swapped because the maximum allowed grid size for x is higher.
             const int i10 =  blockIdx.x;
-            const int i11 =  z / ne12; // TODO fastdiv
-            const int i12 =  z % ne12;
+            const uint2 dm  = fast_div_modulo((uint32_t)z, ne12_fdv);
+            const int i11 =  dm.x;
+            const int i12 =  dm.y;
 
             const int i01 = src1[i10*s10 + i11*s11 + i12*s12];
 
@@ -42,17 +43,18 @@ template<typename src0_t, typename dst_t>
 static __global__ void k_get_rows_float(
         const src0_t * __restrict__ src0, const int32_t * __restrict__ src1, dst_t * __restrict__ dst,
         const int64_t ne00, /*const int64_t ne01, const int64_t ne02, const int64_t ne03,*/
-        /*const int64_t ne10,*/ const int64_t ne11, const int64_t ne12, /*const int64_t ne13,*/
+        /*const int64_t ne10,*/ const int64_t ne11, const uint3 ne12_fdv, /*const int64_t ne13,*/
         /*const size_t s0,*/ const size_t s1, const size_t s2, const size_t s3,
         /*const size_t nb00,*/ const size_t nb01, const size_t nb02, const size_t nb03,
         const size_t s10, const size_t s11, const size_t s12/*, const size_t s13*/) {
 
-    for (int64_t z = blockIdx.z; z < ne11*ne12; z += gridDim.z) {
+    for (int64_t z = blockIdx.z; z < ne11*(int64_t)ne12_fdv.z; z += gridDim.z) {
         for (int64_t i00 = blockIdx.y*blockDim.x + threadIdx.x; i00 < ne00; i00 += gridDim.y*blockDim.x) {
             // The x and y dimensions of the grid are swapped because the maximum allowed grid size for x is higher.
             const int i10 = blockIdx.x;
-            const int i11 = z / ne12; // TODO fastdiv
-            const int i12 = z % ne12;
+            const uint2 dm = fast_div_modulo((uint32_t)z, ne12_fdv);
+            const int i11 = dm.x;
+            const int i12 = dm.y;
 
             if (i00 >= ne00) {
                 return;
@@ -115,10 +117,14 @@ static void get_rows_cuda_q(
 
     GGML_ASSERT(ne00 % 2 == 0);
 
+    GGML_ASSERT(ne12 > 0);
+    GGML_ASSERT(ne11 <= std::numeric_limits<uint32_t>::max() / ne12);
+    const uint3 ne12_fdv = init_fastdiv_values(ne12);
+
     k_get_rows<qk, qr, dq><<<block_nums, block_dims, 0, stream>>>(
         src0_d, src1_d, dst_d,
         ne00, /*ne01, ne02, ne03,*/
-        /*ne10,*/ ne11, ne12, /*ne13,*/
+        /*ne10,*/ ne11, ne12_fdv, /*ne13,*/
         /* s0,*/ s1, s2, s3,
         /* nb00,*/ nb01, nb02, nb03,
         s10, s11, s12/*, s13*/);
@@ -146,10 +152,14 @@ static void get_rows_cuda_float(
     const size_t s12 = nb12 / sizeof(int32_t);
     // const size_t s13 = nb13 / sizeof(int32_t);
 
+    GGML_ASSERT(ne12 > 0);
+    GGML_ASSERT(ne11 <= std::numeric_limits<uint32_t>::max() / ne12);
+    const uint3 ne12_fdv = init_fastdiv_values(ne12);
+
     k_get_rows_float<<<block_nums, block_dims, 0, stream>>>(
         src0_d, src1_d, dst_d,
         ne00, /*ne01, ne02, ne03,*/
-        /*ne10,*/ ne11, ne12, /*ne13,*/
+        /*ne10,*/ ne11, ne12_fdv, /*ne13,*/
         /* s0,*/ s1, s2, s3,
         /* nb00,*/ nb01, nb02, nb03,
         s10, s11, s12/*, s13*/);

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

@@ -5431,8 +5431,8 @@ ggml_backend_reg_t ggml_backend_cuda_reg() {
                 CUDA_CHECK(cudaGetDeviceProperties(&prop, i));
                 dev_ctx->description = prop.name;
 
-                char pci_bus_id[16] = {};
-                snprintf(pci_bus_id, sizeof(pci_bus_id), "%04x:%02x:%02x.0", prop.pciDomainID, prop.pciBusID, prop.pciDeviceID);
+                char pci_bus_id[32] = {};
+                CUDA_CHECK(cudaDeviceGetPCIBusId(pci_bus_id, sizeof(pci_bus_id), i));
                 dev_ctx->pci_bus_id = pci_bus_id;
                 dev_ctx->op_offload_min_batch_size = min_batch_size;
 

ggml/src/ggml-cuda/vendors/hip.h

@@ -55,6 +55,7 @@
 #define cudaDeviceDisablePeerAccess hipDeviceDisablePeerAccess
 #define cudaDeviceEnablePeerAccess hipDeviceEnablePeerAccess
 #define cudaDeviceGetAttribute hipDeviceGetAttribute
+#define cudaDeviceGetPCIBusId hipDeviceGetPCIBusId
 #define cudaDeviceProp hipDeviceProp_t
 #define cudaDeviceSynchronize hipDeviceSynchronize
 #define cudaError_t hipError_t

ggml/src/ggml-cuda/vendors/musa.h

@@ -39,6 +39,7 @@
 #define cudaDeviceCanAccessPeer musaDeviceCanAccessPeer
 #define cudaDeviceDisablePeerAccess musaDeviceDisablePeerAccess
 #define cudaDeviceEnablePeerAccess musaDeviceEnablePeerAccess
+#define cudaDeviceGetPCIBusId musaDeviceGetPCIBusId
 #define cudaDeviceProp musaDeviceProp
 #define cudaDeviceSynchronize musaDeviceSynchronize
 #define cudaError_t musaError_t

ggml/src/ggml-hexagon/CMakeLists.txt

@@ -22,7 +22,8 @@ message(STATUS "hexagon: using ${HEXAGON_SDK_ROOT} and ${HEXAGON_TOOLS_ROOT} for
 include(${HEXAGON_SDK_ROOT}/build/cmake/hexagon_fun.cmake)
 include(ExternalProject)
 
-option(GGML_HEXAGON_HTP_DEBUG "ggml-hexagon: enable HTP debug output" OFF)
+option(GGML_HEXAGON_HTP_DEBUG  "ggml-hexagon: enable HTP debug output" OFF)
+option(GGML_HEXAGON_FA_EXP2_HF "ggml-hexagon: use FP16 exp2 polynomial in FA softmax instead of F32 exp round-trip" OFF)
 set(GGML_HEXAGON_HTP_CERT  "$ENV{HEXAGON_HTP_CERT}" CACHE PATH "ggml-hexagon: enable HTP library signing using certificate")
 set(GGML_HEXAGON_FP32_QUANTIZE_GROUP_SIZE 128 CACHE STRING "ggml-hexagon: quantize group size (32, 64, or 128)")
 

ggml/src/ggml-hexagon/ggml-hexagon.cpp

@@ -2254,8 +2254,7 @@ static bool ggml_hexagon_supported_flash_attn_ext(const struct ggml_hexagon_sess
         return false;
     }
 
-    if (dst->ne[2] != 1 || dst->ne[3] != 1) {
-        // FA during prompt still needs work
+    if (dst->ne[3] != 1) {
         return false;
     }
 
@@ -2421,8 +2420,8 @@ static bool ggml_hexagon_supported_unary(const struct ggml_hexagon_session * ses
         return false;
     }
 
-    // TODO: add support for non-contigiuos tensors
-    if (!ggml_is_contiguous(src0) || !ggml_is_contiguous(dst)) {
+    // TODO: add support for non-contiguous elements within a row
+    if (!ggml_is_contiguous_rows(src0) || !ggml_is_contiguous_rows(dst)) {
         return false;
     }
 

ggml/src/ggml-hexagon/htp/CMakeLists.txt

@@ -44,6 +44,11 @@ target_compile_definitions(${HTP_LIB} PRIVATE
     $<IF:$<BOOL:${HEXAGON_HTP_DEBUG}>,FARF_HIGH=1,>
     FP32_QUANTIZE_GROUP_SIZE=${GGML_HEXAGON_FP32_QUANTIZE_GROUP_SIZE})
 
+if (GGML_HEXAGON_FA_EXP2_HF)
+    message(STATUS "ggml-htp: HMX_FA_USE_EXP2_HF=1 (use FP16 exp2 polynomial in FA softmax)")
+    target_compile_definitions(${HTP_LIB} PRIVATE HMX_FA_USE_EXP2_HF=1)
+endif()
+
 # HMX acceleration: available on v73+ architectures
 set(HTP_HMX_VERSIONS v73 v75 v79 v81)
 list(FIND HTP_HMX_VERSIONS ${DSP_VERSION} _hmx_idx)
@@ -52,11 +57,13 @@ if (_hmx_idx GREATER_EQUAL 0)
     target_sources(${HTP_LIB} PRIVATE
         hmx-queue.c
         hmx-matmul-ops.c
+        hmx-flash-attn-ops.c
     )
 
     # -mhmx enables HMX instruction set (needed by files that include hmx-utils.h)
     set_source_files_properties(
         hmx-matmul-ops.c
+        hmx-flash-attn-ops.c
         PROPERTIES COMPILE_OPTIONS "-mhmx"
     )
 

ggml/src/ggml-hexagon/htp/cmake-toolchain.cmake

@@ -138,15 +138,15 @@ set(CMAKE_SHARED_LIBRARY_SONAME_C_FLAG   "-Wl,-soname,")
 set(CMAKE_SHARED_LIBRARY_SONAME_CXX_FLAG "-Wl,-soname,")
 
 #Compiler Options
-set(COMMON_FLAGS "-mcpu=hexagon${V_ARCH} -m${V_ARCH} -mhvx=${V_ARCH} -fvectorize -Wall -Werror -fno-zero-initialized-in-bss -G0 -fdata-sections -fpic ${XQF_ARGS}")
+set(COMMON_FLAGS "-mcpu=hexagon${V_ARCH} -m${V_ARCH} -mhvx=${V_ARCH} -fvectorize -flto -Wall -Werror -fno-zero-initialized-in-bss -G0 -fdata-sections -fpic ${XQF_ARGS}")
 
 set(CMAKE_CXX_FLAGS_DEBUG          "${COMMON_FLAGS} -O0 -D_DEBUG -g")
-set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "${COMMON_FLAGS} -O3 -g")
-set(CMAKE_CXX_FLAGS_RELEASE        "${COMMON_FLAGS} -O3")
+set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "${COMMON_FLAGS} -O2 -g")
+set(CMAKE_CXX_FLAGS_RELEASE        "${COMMON_FLAGS} -O2")
 
 set(CMAKE_C_FLAGS_DEBUG            "${COMMON_FLAGS} -O0 -D_DEBUG -g")
-set(CMAKE_C_FLAGS_RELWITHDEBINFO   "${COMMON_FLAGS} -O3 -g")
-set(CMAKE_C_FLAGS_RELEASE          "${COMMON_FLAGS} -O3")
+set(CMAKE_C_FLAGS_RELWITHDEBINFO   "${COMMON_FLAGS} -O2 -g")
+set(CMAKE_C_FLAGS_RELEASE          "${COMMON_FLAGS} -O2")
 
 set(CMAKE_ASM_FLAGS_DEBUG          "${COMMON_FLAGS} ${CMAKE_CXX_FLAGS_DEBUG}")
 set(CMAKE_ASM_FLAGS_RELEASE        "${COMMON_FLAGS} ${CMAKE_CXX_FLAGS_RELEASE}")

ggml/src/ggml-hexagon/htp/flash-attn-ops.c

@@ -17,13 +17,14 @@
 #include "htp-ctx.h"
 #include "htp-ops.h"
 #include "htp-ops.h"
+#include "hmx-ops.h"
 
 // Must be multiple of 32
 #define FLASH_ATTN_BLOCK_SIZE (32 * 2)
 
 // This is a bit of a hack because the compiler is strugling to properly inline
 // the default hvx_vec_f32_to_f16 with output into the local array.
-static void __attribute__((noinline)) hvx_vec_f32_to_f16_a(void *ptr, HVX_Vector v0, HVX_Vector v1)
+static __attribute__((noinline)) void hvx_vec_f32_to_f16_a(void *ptr, HVX_Vector v0, HVX_Vector v1)
 {
     *(HVX_Vector *) ptr = hvx_vec_f32_to_f16(v0, v1);
 }
@@ -621,6 +622,17 @@ int op_flash_attn_ext(struct htp_ops_context * octx) {
         return HTP_STATUS_NO_SUPPORT;
     }
 
+#ifdef HTP_HAS_HMX
+    // HMX path: prefill (neq1 >= 32), head_dim multiple of 32, F16 KV
+    if (k->type == HTP_TYPE_F16 && v->type == HTP_TYPE_F16 && k->ne[0] % 32 == 0 && q->ne[1] >= 32) {
+        int ret = hmx_flash_attn_ext(octx);
+        if (ret == HTP_STATUS_OK) {
+            return ret;
+        }
+        // VTCM too small or other failure -> fall through to HVX path
+    }
+#endif
+
     struct htp_fa_context factx;
     factx.octx = octx;
 

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

@@ -74,6 +74,12 @@ static inline size_t hex_smax(size_t a, size_t b) {
     return a > b ? a : b;
 }
 
+static inline void hex_swap_ptr(void ** p1, void ** p2) {
+    void * t = *p1;
+    *p1      = *p2;
+    *p2      = t;
+}
+
 static inline void hex_l2fetch(const void * p, uint32_t width, uint32_t stride, uint32_t height) {
     const uint64_t control = Q6_P_combine_RR(stride, Q6_R_combine_RlRl(width, height));
     Q6_l2fetch_AP((void *) p, control);

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

@@ -61,6 +61,9 @@ int hmx_mat_mul_permuted_qk_0_d16a32(struct htp_context *ctx,
                                       int m, int k, int n,
                                       int weight_type);
 
+// HMX flash attention
+int hmx_flash_attn_ext(struct htp_ops_context * octx);
+
 #ifdef __cplusplus
 }
 #endif

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

@@ -4,6 +4,9 @@
 #ifndef HMX_UTILS_H
 #define HMX_UTILS_H
 
+#include "hvx-base.h"
+
+#include <assert.h>
 #include <hexagon_types.h>
 #include <stddef.h>
 
@@ -12,21 +15,188 @@
 #define HMX_FP16_TILE_N_ELMS 1024
 #define HMX_FP16_TILE_SIZE   2048
 
-#define HMX_INLINE_ALWAYS inline __attribute__((unused, always_inline))
-
 // Initialise aligned 256-byte area with scale vector + zero padding.
-static HMX_INLINE_ALWAYS void hmx_init_column_scales(void *out_scales, HVX_Vector v_scale) {
-    HVX_Vector *pv = (HVX_Vector *)out_scales;
-    *pv++ = v_scale;
-    *pv   = Q6_V_vzero();
+static inline void hmx_init_column_scales(void *out_scales, HVX_Vector v_scale) {
+    volatile HVX_Vector *pv = (HVX_Vector *) out_scales;
+    pv[0] = v_scale;
+    pv[1] = Q6_V_vzero();
 }
 
-// --- VTCM sequential allocator (from htp-ops-lib/include/dsp/vtcm_mgr.h) ---
+// --- Shared scatter offsets and interleave helper ---
 
-static inline uint8_t *vtcm_seq_alloc(uint8_t **vtcm_ptr, size_t size) {
-    uint8_t *p = *vtcm_ptr;
-    *vtcm_ptr += size;
-    return p;
+// vscatter offsets for fused dequant+transpose: write K-values directly to [K][N] tile.
+// word[i] = i*128 maps K-row-pair i to byte offset i*128.
+// Column offset (n*4) is added at runtime.  Entries 0..15 cover one tile (region 2047);
+// entries 16..31 cover the next adjacent tile (region 4095) — pick region size at the
+// call site to scatter into one tile (masked) or two contiguous tiles (unmasked).
+static const int32_t hmx_transpose_scatter_offsets[32] __attribute__((aligned(VLEN))) = {
+    0 * 128,  1 * 128,  2 * 128,  3 * 128,  4 * 128,  5 * 128,  6 * 128,  7 * 128,  8 * 128,  9 * 128,  10 * 128,
+    11 * 128, 12 * 128, 13 * 128, 14 * 128, 15 * 128, 16 * 128, 17 * 128, 18 * 128, 19 * 128, 20 * 128, 21 * 128,
+    22 * 128, 23 * 128, 24 * 128, 25 * 128, 26 * 128, 27 * 128, 28 * 128, 29 * 128, 30 * 128, 31 * 128,
+};
+
+// Scatter row-major FP16 data (in VTCM scratch) into transposed [K][N] tiles.
+// vtcm_src: [n_cols][src_stride] row-major fp16 (only first k elements per row are used)
+// vtcm_dst: [n_col_tiles][n_k_tiles][HMX_FP16_TILE_N_ELMS] tile-major interleaved fp16
+// Processes rows [start_row, end_row) for multi-thread slicing.
+// Full range: start_row=0, end_row=n_cols.
+static inline void hmx_interleave_rows_to_tiles(__fp16 * restrict vtcm_dst,
+                                            const __fp16 * restrict vtcm_src,
+                                            int n_cols,
+                                            int k,
+                                            int src_stride,
+                                            int start_row,
+                                            int end_row) {
+    assert(k % HMX_FP16_TILE_N_COLS == 0);
+
+    const int            n_k_tiles     = k / HMX_FP16_TILE_N_COLS;
+    const HVX_Vector     v_scat_base   = hvx_vmem(hmx_transpose_scatter_offsets);
+    const HVX_Vector     v_scat_step   = Q6_V_vsplat_R(4);
+    const HVX_VectorPred q_mask64      = Q6_Q_vsetq_R(64);
+    // Each hvx_vmemu load brings 64 fp16 = 128 bytes covering 2 adjacent K-tiles.
+    // When n_k_tiles is even, scatter into 2 K-tiles per call (region 4095, no mask)
+    // using the upper half of hmx_transpose_scatter_offsets.  Tail one K-tile (when
+    // n_k_tiles is odd) falls back to single-tile masked scatter.
+    const bool           pair_scatter  = (n_k_tiles & 1) == 0;
+    const size_t         pair_region   = (size_t) (2 * HMX_FP16_TILE_SIZE - 1);
+    const size_t         single_region = (size_t) (HMX_FP16_TILE_SIZE - 1);
+    __builtin_assume(k > 0);
+    __builtin_assume(end_row > start_row);
+
+    if (pair_scatter) {
+        // Step c by 64 fp16 (two K-tiles per scatter), advance dst by 2 tiles per iter.
+        const int    c_step      = 2 * HMX_FP16_TILE_N_COLS;
+        const size_t c_byte_step = (size_t) c_step * sizeof(__fp16);
+        const size_t dst_step    = 2 * (size_t) HMX_FP16_TILE_N_ELMS;
+        const int    n_c_iters   = k / c_step;
+
+        for (int r = start_row; r < end_row; r += 2) {
+            const int        ct             = r / HMX_FP16_TILE_N_ROWS;
+            const int        local_r        = r % HMX_FP16_TILE_N_ROWS;
+            const bool       next_row_valid = (r + 1) < end_row && (r + 1) < n_cols;
+            const HVX_Vector v_off0         = Q6_Vw_vadd_VwVw(v_scat_base, Q6_V_vsplat_R(local_r * 4));
+            const HVX_Vector v_off1         = Q6_Vw_vadd_VwVw(v_off0, v_scat_step);
+
+            __fp16 *        tile_base = vtcm_dst + (size_t) ct * n_k_tiles * HMX_FP16_TILE_N_ELMS;
+            const uint8_t * p0        = (const uint8_t *) (vtcm_src + r * src_stride);
+            const uint8_t * p1        = next_row_valid ? (const uint8_t *) (vtcm_src + (r + 1) * src_stride) : NULL;
+
+            if (p1) {
+                for (int i = 0; i < n_c_iters; ++i) {
+                    HVX_Vector v0 = hvx_vmemu(p0);
+                    p0 += c_byte_step;
+                    HVX_Vector v1 = hvx_vmemu(p1);
+                    p1 += c_byte_step;
+                    Q6_vscatter_RMVwV((size_t) tile_base, pair_region, v_off0, v0);
+                    Q6_vscatter_RMVwV((size_t) tile_base, pair_region, v_off1, v1);
+                    tile_base += dst_step;
+                }
+            } else {
+                const HVX_Vector vzero = Q6_V_vzero();
+                for (int i = 0; i < n_c_iters; ++i) {
+                    HVX_Vector v0 = hvx_vmemu(p0);
+                    p0 += c_byte_step;
+                    Q6_vscatter_RMVwV((size_t) tile_base, pair_region, v_off0, v0);
+                    Q6_vscatter_RMVwV((size_t) tile_base, pair_region, v_off1, vzero);
+                    tile_base += dst_step;
+                }
+            }
+        }
+    } else {
+        // Fallback: scatter one K-tile per call (region 2047, masked).
+        const int    c_step      = HMX_FP16_TILE_N_COLS;
+        const size_t c_byte_step = (size_t) c_step * sizeof(__fp16);
+        const size_t dst_step    = (size_t) HMX_FP16_TILE_N_ELMS;
+        const int    n_c_iters   = k / c_step;
+
+        for (int r = start_row; r < end_row; r += 2) {
+            const int        ct             = r / HMX_FP16_TILE_N_ROWS;
+            const int        local_r        = r % HMX_FP16_TILE_N_ROWS;
+            const bool       next_row_valid = (r + 1) < end_row && (r + 1) < n_cols;
+            const HVX_Vector v_off0         = Q6_Vw_vadd_VwVw(v_scat_base, Q6_V_vsplat_R(local_r * 4));
+            const HVX_Vector v_off1         = Q6_Vw_vadd_VwVw(v_off0, v_scat_step);
+
+            __fp16 *        tile_base = vtcm_dst + (size_t) ct * n_k_tiles * HMX_FP16_TILE_N_ELMS;
+            const uint8_t * p0        = (const uint8_t *) (vtcm_src + r * src_stride);
+            const uint8_t * p1        = next_row_valid ? (const uint8_t *) (vtcm_src + (r + 1) * src_stride) : NULL;
+
+            if (p1) {
+                for (int i = 0; i < n_c_iters; ++i) {
+                    HVX_Vector v0 = hvx_vmemu(p0);
+                    p0 += c_byte_step;
+                    HVX_Vector v1 = hvx_vmemu(p1);
+                    p1 += c_byte_step;
+                    Q6_vscatter_QRMVwV(q_mask64, (size_t) tile_base, single_region, v_off0, v0);
+                    Q6_vscatter_QRMVwV(q_mask64, (size_t) tile_base, single_region, v_off1, v1);
+                    tile_base += dst_step;
+                }
+            } else {
+                const HVX_Vector vzero = Q6_V_vzero();
+                for (int i = 0; i < n_c_iters; ++i) {
+                    HVX_Vector v0 = hvx_vmemu(p0);
+                    p0 += c_byte_step;
+                    Q6_vscatter_QRMVwV(q_mask64, (size_t) tile_base, single_region, v_off0, v0);
+                    Q6_vscatter_QRMVwV(q_mask64, (size_t) tile_base, single_region, v_off1, vzero);
+                    tile_base += dst_step;
+                }
+            }
+        }
+    }
+}
+
+// Interleave row-major FP16 data into column-major tile format.
+// Input: [n_rows, head_dim] row-major.  Output: tile[dim_tile][row_tile].
+// Processes rows [start_row, end_row) for multi-thread slicing.
+// Full range: start_row=0, end_row=n_rows.
+static inline void hmx_interleave_cols_to_tiles(__fp16 * restrict tiles_out,
+                                            const __fp16 * restrict src,
+                                            int n_rows,
+                                            int head_dim,
+                                            int src_stride,
+                                            int n_row_tiles,
+                                            int start_row,
+                                            int end_row) {
+    __builtin_assume(head_dim > 0);
+    const size_t tile_stride_elms = (size_t) n_row_tiles * HMX_FP16_TILE_N_ELMS;
+
+    for (int r = start_row; r < end_row; r += 2) {
+        const bool next_row_valid = (r + 1) < end_row && (r + 1) < n_rows;
+
+        const HVX_Vector * pv_in0 = (const HVX_Vector *) (src + r * src_stride);
+        const HVX_Vector * pv_in1 = next_row_valid ? (const HVX_Vector *) (src + (r + 1) * src_stride) : NULL;
+
+        // Row-pair invariants hoisted out of the c loop.
+        const int r0      = r / HMX_FP16_TILE_N_ROWS;
+        const int r1_half = (r % HMX_FP16_TILE_N_ROWS) / 2;
+
+        // tb0 starts at tile (c0=0, r0); tb1 at the adjacent dim-tile (c0=1, r0).
+        // Each c step (+= 64) advances both by 2 dim-tiles worth of fp16.
+        __fp16 *     tb0     = tiles_out + (size_t) r0 * HMX_FP16_TILE_N_ELMS;
+        __fp16 *     tb1     = tb0 + tile_stride_elms;
+        const size_t tb_step = 2 * tile_stride_elms;
+
+        if (pv_in1) {
+            for (int c = 0; c < head_dim; c += 64) {
+                HVX_Vector     v0             = *pv_in0++;
+                HVX_Vector     v1             = *pv_in1++;
+                HVX_VectorPair vp             = Q6_W_vshuff_VVR(v1, v0, -2);
+                ((HVX_Vector *) tb0)[r1_half] = Q6_V_lo_W(vp);
+                ((HVX_Vector *) tb1)[r1_half] = Q6_V_hi_W(vp);
+                tb0 += tb_step;
+                tb1 += tb_step;
+            }
+        } else {
+            const HVX_Vector vzero = Q6_V_vzero();
+            for (int c = 0; c < head_dim; c += 64) {
+                HVX_Vector     v0             = *pv_in0++;
+                HVX_VectorPair vp             = Q6_W_vshuff_VVR(vzero, v0, -2);
+                ((HVX_Vector *) tb0)[r1_half] = Q6_V_lo_W(vp);
+                ((HVX_Vector *) tb1)[r1_half] = Q6_V_hi_W(vp);
+                tb0 += tb_step;
+                tb1 += tb_step;
+            }
+        }
+    }
 }
 
 #endif // HMX_UTILS_H

ggml/src/ggml-hexagon/htp/hvx-base.h

@@ -77,6 +77,12 @@ static inline int32_t hvx_vec_get_i32(HVX_Vector v) {
     return x;
 }
 
+static inline _Float16 hvx_vec_get_f16(HVX_Vector v) {
+    _Float16 __attribute__((aligned(128))) x;
+    hvx_vec_store_a(&x, 2, v);
+    return x;
+}
+
 static inline HVX_Vector hvx_vec_abs_f16(HVX_Vector v) {
     // abs by clearing the fp16 sign bit
     HVX_Vector mask = Q6_Vh_vsplat_R(0x7fff);

ggml/src/ggml-hexagon/htp/hvx-copy.h

@@ -7,7 +7,8 @@
 
 #include "hvx-base.h"
 
-#define hvx_splat_loop_body(dst_type, vec_store)                 \
+#define hvx_splat_pragma(x) _Pragma(#x)
+#define hvx_splat_loop_body(dst_type, vec_store, unroll_cnt)     \
     do {                                                         \
         dst_type * restrict vdst = (dst_type *) dst;             \
                                                                  \
@@ -16,7 +17,7 @@
                                                                  \
         uint32_t i = 0;                                          \
                                                                  \
-        _Pragma("unroll(4)")                                     \
+        hvx_splat_pragma(unroll(unroll_cnt))                     \
         for (; i < nvec; i++) {                                  \
             vdst[i] = src;                                       \
         }                                                        \
@@ -25,31 +26,47 @@
         }                                                        \
     } while(0)
 
-static inline void hvx_splat_a(uint8_t * restrict dst, HVX_Vector src, uint32_t n, uint32_t elem_size) {
+static inline void hvx_splat_a(void * restrict dst, HVX_Vector src, uint32_t n, uint32_t elem_size) {
     assert((unsigned long) dst % 128 == 0);
-    hvx_splat_loop_body(HVX_Vector, hvx_vec_store_a);
+    hvx_splat_loop_body(HVX_Vector, hvx_vec_store_a, 4);
 }
 
-static inline void hvx_splat_u(uint8_t * restrict dst, HVX_Vector src, uint32_t n, uint32_t elem_size) {
-    hvx_splat_loop_body(HVX_UVector, hvx_vec_store_u);
+static inline void hvx_splat_u(void * restrict dst, HVX_Vector src, uint32_t n, uint32_t elem_size) {
+    hvx_splat_loop_body(HVX_UVector, hvx_vec_store_u, 4);
 }
 
-static inline void hvx_splat_f32_a(uint8_t * restrict dst, float v, uint32_t n) {
+static inline void hvx_splat_f32_a(void * restrict dst, float v, uint32_t n) {
     hvx_splat_a(dst,  hvx_vec_splat_f32(v), n, sizeof(float));
 }
 
-static inline void hvx_splat_f32_u(uint8_t * restrict dst, float v, uint32_t n) {
+static inline void hvx_splat_f32_u(void * restrict dst, float v, uint32_t n) {
     hvx_splat_u(dst,  hvx_vec_splat_f32(v), n, sizeof(float));
 }
 
-static inline void hvx_splat_f16_a(uint8_t * restrict dst, _Float16 v, uint32_t n) {
+static inline void hvx_splat_f16_a(void * restrict dst, _Float16 v, uint32_t n) {
     hvx_splat_u(dst,  hvx_vec_splat_f16(v), n, sizeof(__fp16));
 }
 
-static inline void hvx_splat_f16_u(uint8_t * restrict dst, _Float16 v, uint32_t n) {
+static inline void hvx_splat_f16_u(void * restrict dst, _Float16 v, uint32_t n) {
     hvx_splat_u(dst,  hvx_vec_splat_f16(v), n, sizeof(__fp16));
 }
 
+static inline void hvx_splat_u16_a(void * restrict dst, uint16_t v, uint32_t n) {
+    hvx_splat_a(dst,  Q6_Vh_vsplat_R(v), n, sizeof(uint16_t));
+}
+
+static inline void hvx_splat_u16_u(void * restrict dst, uint16_t v, uint32_t n) {
+    hvx_splat_u(dst,  Q6_Vh_vsplat_R(v), n, sizeof(uint16_t));
+}
+
+static inline void hvx_splat_u8_a(void * restrict dst, uint8_t v, uint32_t n) {
+    hvx_splat_a(dst,  Q6_Vb_vsplat_R(v), n, 1);
+}
+
+static inline void hvx_splat_u8_u(void * restrict dst, uint8_t v, uint32_t n) {
+    hvx_splat_u(dst,  Q6_Vb_vsplat_R(v), n, 1);
+}
+
 #define hvx_copy_loop_body(dst_type, src_type, vec_store)            \
     do {                                                             \
         dst_type * restrict vdst = (dst_type *) dst;                 \

ggml/src/ggml-hexagon/htp/hvx-exp.h

@@ -17,7 +17,7 @@
 #define EXP_LOGN2   (0x3F317218)  // ln(2)   = 0.6931471805
 #define EXP_LOG2E   (0x3FB8AA3B)  // log2(e) = 1/ln(2) = 1.4426950408
 #define EXP_ONE     (0x3f800000)  // 1.0
-#define EXP_RANGE_R (0x42B16666)  // 88.7
+#define EXP_RANGE_R (0x42B17218)  // ln(FLT_MAX) approx = 88.7228
 #define EXP_RANGE_L (0xC2B00000)  // -88.0 (approx log(FLT_MIN))
 
 static inline HVX_Vector hvx_vec_exp_f32(HVX_Vector in_vec) {
@@ -163,7 +163,7 @@ static inline void hvx_exp_f32(uint8_t * restrict dst, const uint8_t * restrict
     HVX_Vector vec_out = Q6_V_vzero();
 
     static const float kInf    = INFINITY;
-    static const float kMaxExp = 88.7f;
+    static const float kMaxExp = 88.7228f;
 
     const HVX_Vector max_exp = hvx_vec_splat_f32(kMaxExp);
     const HVX_Vector inf     = hvx_vec_splat_f32(kInf);

ggml/src/ggml-hexagon/htp/matmul-ops.c

@@ -2991,12 +2991,10 @@ int op_matmul(struct htp_ops_context * octx) {
         return op_matmul_hvx(octx);
     }
 
-    // M alignment: when M > 32 but not 32-aligned, we split into
-    // HMX (first m_hmx = M & ~31 rows) + HVX (remaining m_tail rows).
-    // When M <= 32 and not 32-aligned, fall back entirely to HVX.
+    // M alignment: Use HMX when M >= 32, the last partial tile (m_total % 32 rows)
+    //  is handled by HMX itself; when M < 32  fall back to HVX.
     const int m_total = (int) src1->ne[1];
-    const int m_tail  = m_total % 32;
-    const int m_hmx   = m_total - m_tail;
+    const int m_hmx   = m_total & ~31;   // 0 when M < 32
 
     if (m_hmx == 0) {
         return op_matmul_hvx(octx);
@@ -3009,7 +3007,6 @@ int op_matmul(struct htp_ops_context * octx) {
     int k = (int) src0->ne[0];  // inner dimension
     int n = (int) src0->ne[1];  // weight columns
 
-    // --- Phase 1: HMX on the first m_hmx (32-aligned) rows ---
     int ret = -1;
 
     // Row strides in elements. For compact tensors these equal k; for
@@ -3027,7 +3024,7 @@ int op_matmul(struct htp_ops_context * octx) {
                 .dst             = (float *) dst->data,
                 .activation      = (float *) src1->data,
                 .permuted_weight = (const __fp16 *) src0->data,
-                .m               = m_hmx,
+                .m               = m_total,
                 .k               = k,
                 .n               = n,
                 .act_stride      = act_stride,
@@ -3048,12 +3045,12 @@ int op_matmul(struct htp_ops_context * octx) {
         } else {
             ret = hmx_mat_mul_permuted_w16a32(octx->ctx,
                     (float*) dst->data, (float*) src1->data, (const __fp16 *) src0->data,
-                    m_hmx, k, n, act_stride, wgt_stride);
+                    m_total, k, n, act_stride, wgt_stride);
         }
     } else {
         ret = hmx_mat_mul_permuted_qk_0_d16a32(octx->ctx,
                     (float*) dst->data, (float*) src1->data, (const uint8_t *) src0->data,
-                    m_hmx, k, n, (int) src0->type);
+                    m_total, k, n, (int) src0->type);
     }
 
     if (ret != 0) {
@@ -3061,27 +3058,6 @@ int op_matmul(struct htp_ops_context * octx) {
         return op_matmul(octx);
     }
 
-    // --- Phase 2: HVX on the remaining m_tail rows ---
-    if (m_tail > 0) {
-        // copy of src1 and dst
-        struct htp_tensor src1_tail = *src1;
-        struct htp_tensor dst_tail  = *dst;
-
-        src1_tail.ne[1] = m_tail; // only tail rows
-        dst_tail.ne[1]  = m_tail; // only tail rows
-
-        // Offset activation and dst pointers past the HMX-processed rows.
-        // Use nb[1] (row stride in bytes) to compute the byte offset.
-        src1_tail.data += (uint32_t) m_hmx * src1->nb[1];
-        dst_tail.data  += (uint32_t) m_hmx * dst->nb[1];
-
-        octx->src[1] = &src1_tail;
-        octx->dst    = &dst_tail;
-
-        FARF(HIGH, "hmx-matmul: HVX tail m_tail %d src1 %p dst %p", m_tail, (void *) src1_tail.data, (void *) dst_tail.data);
-        return op_matmul_hvx(octx);
-    }
-
     return 0;
 #endif // HTP_HAS_HMX
 }

ggml/src/ggml-hexagon/htp/unary-ops.c

@@ -26,8 +26,8 @@ struct htp_unary_context {
     const uint8_t *           data_src0;
     uint8_t *                 data_dst;
 
-    size_t                    src0_row_size;
-    size_t                    dst_row_size;
+    size_t                    src0_data_row_size;   // actual data bytes per row
+    size_t                    dst_data_row_size;    // actual data bytes per row
 
     size_t                    src0_row_size_aligned;
     size_t                    dst_row_size_aligned;
@@ -41,6 +41,40 @@ struct htp_unary_context {
     uint32_t                  nc;
 };
 
+// Convert flat row index to DDR byte offset using the tensor's actual strides.
+// ir = i1 + ne1*(i2 + ne2*i3)  =>  offset = i1*nb1 + i2*nb2 + i3*nb3
+static inline size_t unary_row_offset(uint32_t ir,
+                                      uint32_t ne1, uint32_t ne2,
+                                      size_t nb1, size_t nb2, size_t nb3) {
+    const uint32_t i1 = ir % ne1;
+    const uint32_t i2 = (ir / ne1) % ne2;
+    const uint32_t i3 = ir / (ne1 * ne2);
+    return i1 * nb1 + i2 * nb2 + i3 * nb3;
+}
+// Safe DMA block size from row `ir`: clamp to the tighter dim-1 slice
+// boundary of src and dst so the nb1 stride stays valid for all rows.
+static inline uint32_t unary_block_size(uint32_t ir,
+                                        uint32_t end_row,
+                                        uint32_t block,
+                                        bool src_contig,
+                                        bool dst_contig,
+                                        uint32_t src_ne1,
+                                        uint32_t dst_ne1) {
+    uint32_t limit = MIN(block, end_row - ir);
+
+    if (!src_contig) {
+        const uint32_t src_slice_end = (ir / src_ne1 + 1) * src_ne1;
+        limit = MIN(limit, src_slice_end - ir);
+    }
+
+    if (!dst_contig) {
+        const uint32_t dst_slice_end = (ir / dst_ne1 + 1) * dst_ne1;
+        limit = MIN(limit, dst_slice_end - ir);
+    }
+
+    return limit;
+}
+
 #define htp_unary_preamble            \
     const uint32_t ne00 = src->ne[0]; \
     const uint32_t ne01 = src->ne[1]; \
@@ -276,8 +310,8 @@ static void unary_job_f32_per_thread(unsigned int nth, unsigned int ith, void *
     int32_t *                 op_params = octx->op_params;
     uint32_t                  src0_nrows_per_thread = uctx->src0_nrows_per_thread;
 
-    const size_t src0_row_size = uctx->src0_row_size;
-    const size_t dst_row_size  = uctx->dst_row_size;
+    const size_t src0_data_row_size = uctx->src0_data_row_size;
+    const size_t dst_data_row_size  = uctx->dst_data_row_size;
 
     const size_t src0_row_size_aligned = uctx->src0_row_size_aligned;
     const size_t dst_row_size_aligned  = uctx->dst_row_size_aligned;
@@ -303,7 +337,16 @@ static void unary_job_f32_per_thread(unsigned int nth, unsigned int ith, void *
     size_t src0_spad_half_size = uctx->src0_spad_half_size;
     size_t dst_spad_half_size  = uctx->dst_spad_half_size;
 
-    const int BLOCK = uctx->block;
+    // Non-contiguous tensors have gaps at dim-2/3 boundaries that a single-stride
+    // 2D DMA descriptor cannot span. Clamp BLOCK to ne1 (one dim-1 slice) so every
+    // transfer stays within a nb1-uniform region. Skipped for contiguous tensors.
+    const bool src0_contig = (nb02 == (size_t)ne01 * nb01) &&
+                             (nb03 == (size_t)ne02 * nb02);
+    const bool dst_contig  = (nb2  == (size_t)ne1  * nb1)  &&
+                             (nb3  == (size_t)ne2  * nb2);
+    const uint32_t src0_max_block = src0_contig ? uctx->block : MIN((uint32_t)uctx->block, ne01);
+    const uint32_t dst_max_block  = dst_contig  ? uctx->block : MIN((uint32_t)uctx->block, ne1);
+    const uint32_t BLOCK = MIN(src0_max_block, dst_max_block);
     if (BLOCK == 0) {
         FARF(ERROR, "unary-f32 : current VTCM reservation %zu is too small for even 1 row per thread, needed at least %zu\n",
              octx->src0_spad.size_per_thread, src0_row_size_aligned);
@@ -312,21 +355,23 @@ static void unary_job_f32_per_thread(unsigned int nth, unsigned int ith, void *
 
     dma_queue * dma_queue = octx->ctx->dma[ith];
 
-    for (uint32_t ir = src0_start_row, spad_idx = 0; ir < src0_end_row && spad_idx < 2; ir += BLOCK, spad_idx++) {
-        const uint32_t block_size = MIN(BLOCK, src0_end_row - ir);
+    for (uint32_t ir = src0_start_row, spad_idx = 0; ir < src0_end_row && spad_idx < 2; spad_idx++) {
+        const uint32_t block_size = unary_block_size(ir, src0_end_row, BLOCK, src0_contig, dst_contig, ne01, ne1);
 
         // Dummy DMA transation for sequencing (interleaving dst,src,dst,...)
-        dma_queue_push_vtcm_to_ddr(dma_queue,
+        dma_queue_push(dma_queue,
             dma_make_ptr(data_dst, dst_spad_data + (spad_idx * dst_spad_half_size)),
-            dst_row_size, dst_row_size_aligned, 0);
+            nb1, dst_row_size_aligned, dst_data_row_size, 0);
 
-        dma_queue_push_ddr_to_vtcm(dma_queue,
-            dma_make_ptr(src0_spad_data + (spad_idx * src0_spad_half_size), data_src + (ir * src0_row_size)),
-            src0_row_size_aligned, src0_row_size, block_size);
+        const size_t src0_off = unary_row_offset(ir, ne01, ne02, nb01, nb02, nb03);
+        dma_queue_push(dma_queue,
+            dma_make_ptr(src0_spad_data + (spad_idx * src0_spad_half_size), data_src + src0_off),
+            src0_row_size_aligned, nb01, src0_data_row_size, block_size);
+        ir += block_size;
     }
 
-    for (uint32_t ir = src0_start_row; ir < src0_end_row; ir += BLOCK) {
-        const uint32_t block_size = MIN(BLOCK, src0_end_row - ir);
+    for (uint32_t ir = src0_start_row; ir < src0_end_row; ) {
+        const uint32_t block_size = unary_block_size(ir, src0_end_row, BLOCK, src0_contig, dst_contig, ne01, ne1);
 
         float * dst_spad  = (float *) dma_queue_pop(dma_queue).src;
         float * src0_spad = (float *) dma_queue_pop(dma_queue).dst;
@@ -361,18 +406,25 @@ static void unary_job_f32_per_thread(unsigned int nth, unsigned int ith, void *
                 break;
         }
 
-        dma_queue_push_vtcm_to_ddr(dma_queue,
-            dma_make_ptr(data_dst + (ir * dst_row_size), dst_spad),
-            dst_row_size, dst_row_size_aligned, block_size);
+        const size_t dst_off = unary_row_offset(ir, ne1, ne2, nb1, nb2, nb3);
+        dma_queue_push(dma_queue,
+            dma_make_ptr(data_dst + dst_off, dst_spad),
+            nb1, dst_row_size_aligned, dst_data_row_size, block_size);
 
         // prefetch N+2 loop iteration if any
-        const uint32_t pref_block = (ir + BLOCK * 2);
-        if (pref_block < src0_end_row) {
-            const uint32_t pref_block_size = MIN(BLOCK, src0_end_row - pref_block);
-            dma_queue_push_ddr_to_vtcm(dma_queue,
-                dma_make_ptr(src0_spad, data_src + (pref_block * src0_row_size)),
-                src0_row_size_aligned, src0_row_size, pref_block_size);
+        const uint32_t next_ir = ir + block_size;
+        if (next_ir < src0_end_row) {
+            const uint32_t next_block_size = unary_block_size(next_ir, src0_end_row, BLOCK, src0_contig, dst_contig, ne01, ne1);
+            const uint32_t pref_ir = next_ir + next_block_size;
+            if (pref_ir < src0_end_row) {
+                const uint32_t pref_block_size = unary_block_size(pref_ir, src0_end_row, BLOCK, src0_contig, dst_contig, ne01, ne1);
+                const size_t src0_pref_off = unary_row_offset(pref_ir, ne01, ne02, nb01, nb02, nb03);
+            dma_queue_push(dma_queue,
+                dma_make_ptr(src0_spad, data_src + src0_pref_off),
+                src0_row_size_aligned, nb01, src0_data_row_size, pref_block_size);
+            }
         }
+        ir += block_size;
     }
 
     dma_queue_flush(dma_queue);
@@ -426,11 +478,11 @@ static int execute_op_unary_f32(struct htp_ops_context * octx) {
     const uint32_t src0_nrows = src0->ne[1] * src0->ne[2] * src0->ne[3];
     const uint32_t n_threads  = MIN(octx->n_threads, src0_nrows);
 
-    const size_t src0_row_size = src0->nb[1];
-    const size_t dst_row_size  = dst->nb[1];
+    const size_t src0_data_row_size = src0->ne[0] * sizeof(float);
+    const size_t dst_data_row_size  = dst->ne[0]  * sizeof(float);
 
-    const size_t src0_row_size_aligned = hex_round_up(src0_row_size, VLEN);
-    const size_t dst_row_size_aligned  = hex_round_up(dst_row_size, VLEN);
+    const size_t src0_row_size_aligned = hex_round_up(src0_data_row_size, VLEN);
+    const size_t dst_row_size_aligned  = hex_round_up(dst_data_row_size,  VLEN);
 
     // VTCM scratchpads for all tensors
     // N rows per thread, padded to HVX vector size
@@ -468,8 +520,8 @@ static int execute_op_unary_f32(struct htp_ops_context * octx) {
             .data_src0             = (const uint8_t *)src0->data,
             .data_dst              = (uint8_t *)dst->data,
 
-            .src0_row_size         = src0_row_size,
-            .dst_row_size          = dst_row_size,
+            .src0_data_row_size    = src0_data_row_size,
+            .dst_data_row_size     = dst_data_row_size,
 
             .src0_row_size_aligned = src0_row_size_aligned,
             .dst_row_size_aligned  = dst_row_size_aligned,

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

@@ -0,0 +1,16 @@
+#ifndef VTCM_UTILS_H
+#define VTCM_UTILS_H
+
+#include "hex-utils.h"
+
+#include <assert.h>
+#include <stdint.h>
+#include <hexagon_types.h>
+
+static inline uint8_t *vtcm_seq_alloc(uint8_t **vtcm_ptr, size_t size) {
+    uint8_t *p = *vtcm_ptr;
+    *vtcm_ptr += size;
+    return p;
+}
+
+#endif // VTCM_UTILS_H

ggml/src/ggml-metal/ggml-metal-device.h

@@ -282,6 +282,7 @@ bool   ggml_metal_buffer_is_shared(ggml_metal_buffer_t buf);
 void   ggml_metal_buffer_memset_tensor(ggml_metal_buffer_t buf, struct ggml_tensor * tensor, uint8_t value, size_t offset, size_t size);
 void   ggml_metal_buffer_set_tensor   (ggml_metal_buffer_t buf, struct ggml_tensor * tensor, const void * data, size_t offset, size_t size);
 void   ggml_metal_buffer_get_tensor   (ggml_metal_buffer_t buf, const struct ggml_tensor * tensor, void * data, size_t offset, size_t size);
+bool   ggml_metal_buffer_cpy_tensor   (ggml_metal_buffer_t buf, const struct ggml_tensor * src, struct ggml_tensor * dst);
 void   ggml_metal_buffer_clear        (ggml_metal_buffer_t buf, uint8_t value);
 
 // finds the Metal buffer that contains the tensor data on the GPU device

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

@@ -1,6 +1,7 @@
 #import "ggml-metal-device.h"
 
 #import "ggml-impl.h"
+#import "ggml-backend-impl.h"
 
 #include <Foundation/Foundation.h>
 
@@ -1737,6 +1738,47 @@ void ggml_metal_buffer_get_tensor(ggml_metal_buffer_t buf, const struct ggml_ten
     }
 }
 
+bool ggml_metal_buffer_cpy_tensor(ggml_metal_buffer_t buf_dst, const struct ggml_tensor * src, struct ggml_tensor * dst) {
+    ggml_metal_buffer_t buf_src = (ggml_metal_buffer_t)src->buffer->context;
+
+    const size_t size = ggml_nbytes(src);
+
+    // if both buffers are shared, we can use memcpy directly
+    if (buf_dst->is_shared && buf_src->is_shared) {
+        memcpy(dst->data, src->data, size);
+        return true;
+    }
+
+    // for private buffers, we need to use Metal blit commands
+    @autoreleasepool {
+        struct ggml_metal_buffer_id bid_src = ggml_metal_buffer_get_id(buf_src, src);
+        struct ggml_metal_buffer_id bid_dst = ggml_metal_buffer_get_id(buf_dst, dst);
+
+        if (bid_src.metal == nil || bid_dst.metal == nil) {
+            return false;
+        }
+
+        id<MTLCommandBuffer> cmd_buf = [buf_dst->dev->mtl_queue commandBufferWithUnretainedReferences];
+
+        {
+            id<MTLBlitCommandEncoder> encoder = [cmd_buf blitCommandEncoder];
+
+            [encoder copyFromBuffer:bid_src.metal
+                       sourceOffset:bid_src.offs
+                           toBuffer:bid_dst.metal
+                  destinationOffset:bid_dst.offs
+                               size:size];
+
+            [encoder endEncoding];
+        }
+
+        [cmd_buf commit];
+        [cmd_buf waitUntilCompleted];
+    }
+
+    return true;
+}
+
 void ggml_metal_buffer_clear(ggml_metal_buffer_t buf, uint8_t value) {
     if (buf->is_shared) {
         memset(buf->all_data, value, buf->all_size);

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

@@ -17,6 +17,9 @@
 // note: can be overridden with GGML_METAL_DEVICES env to simulate virtual devices
 static int g_devices = 1;
 
+// forward declaration
+static bool ggml_backend_buffer_is_metal(ggml_backend_buffer_t buffer);
+
 ////////////////////////////////////////////////////////////////////////////////
 // backend interface
 ////////////////////////////////////////////////////////////////////////////////
@@ -68,11 +71,11 @@ static bool ggml_backend_metal_buffer_shared_cpy_tensor(ggml_backend_buffer_t bu
 
     GGML_ASSERT(ggml_metal_buffer_is_shared(ctx));
 
-    GGML_UNUSED(buffer);
-    GGML_UNUSED(src);
-    GGML_UNUSED(dst);
+    if (!ggml_backend_buffer_is_metal(src->buffer)) {
+        return false;
+    }
 
-    return false;
+    return ggml_metal_buffer_cpy_tensor(ctx, src, dst);
 }
 
 static void ggml_backend_metal_buffer_shared_clear(ggml_backend_buffer_t buffer, uint8_t value) {
@@ -144,11 +147,11 @@ static bool ggml_backend_metal_buffer_private_cpy_tensor(ggml_backend_buffer_t b
 
     GGML_ASSERT(!ggml_metal_buffer_is_shared(ctx));
 
-    GGML_UNUSED(buffer);
-    GGML_UNUSED(src);
-    GGML_UNUSED(dst);
+    if (!ggml_backend_buffer_is_metal(src->buffer)) {
+        return false;
+    }
 
-    return false;
+    return ggml_metal_buffer_cpy_tensor(ctx, src, dst);
 }
 
 static void ggml_backend_metal_buffer_private_clear(ggml_backend_buffer_t buffer, uint8_t value) {

ggml/src/ggml-opencl/CMakeLists.txt

@@ -66,8 +66,6 @@ set(GGML_OPENCL_KERNELS
     diag
     div
     gelu
-    gemv_noshuffle_general
-    gemv_noshuffle
     get_rows
     glu
     group_norm
@@ -75,7 +73,6 @@ set(GGML_OPENCL_KERNELS
     im2col_f32
     im2col_f16
     mean
-    mul_mat_Ab_Bi_8x4
     mul_mv_f16_f16
     mul_mv_f16_f32_1row
     mul_mv_f16_f32_l4
@@ -107,6 +104,10 @@ set(GGML_OPENCL_KERNELS
     mul_mv_id_mxfp4_f32_flat
     gemm_moe_mxfp4_f32
     gemv_moe_mxfp4_f32
+    gemm_moe_mxfp4_f32_ns
+    gemv_moe_mxfp4_f32_ns
+    moe_reorder_b
+    moe_sort_by_expert
     mul_mm_f32_f32_l4_lm
     mul_mm_f16_f32_l4_lm
     mul_mm_q4_0_f32_l4_lm
@@ -116,12 +117,15 @@ set(GGML_OPENCL_KERNELS
     mul_mm_q4_k_f32_l4_lm
     mul_mm_q5_k_f32_l4_lm
     mul_mm_q6_k_f32_l4_lm
-    mul_mm_q8_0_f32_8x4
+    gemv_noshuffle_q4_0_f32
+    gemv_noshuffle_q4_0_f32_spec
+    gemm_noshuffle_q4_0_f32
     gemv_noshuffle_q4_1_f32
     gemm_noshuffle_q4_1_f32
     gemv_noshuffle_iq4_nl_f32
     gemm_noshuffle_iq4_nl_f32
-    gemv_noshuffle_general_q8_0_f32
+    gemv_noshuffle_q8_0_f32
+    gemm_noshuffle_q8_0_f32
     gemv_noshuffle_q4_k_f32
     gemm_noshuffle_q4_k_f32
     gemv_noshuffle_q6_k_f32

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

@@ -371,6 +371,93 @@ kernel void kernel_restore_block_mxfp4_trans(
     b->e = src_e[src_blk_offset];
 }
 
+kernel void kernel_convert_block_mxfp4_trans4_ns(
+    global struct block_mxfp4 * src0,
+    __global uint * dst_q,
+    __global uchar * dst_e,
+    uint ne00,
+    uint ne01
+) {
+    uint i00 = get_global_id(1);
+    uint i01 = get_global_id(0);
+    uint i02 = get_global_id(2);
+
+    uint ne00_blk = ne00 / QK_MXFP4;
+    uint src_blk_offset = i00 + i01 * ne00_blk + i02 * ne00_blk * ne01;
+    uint dst_blk_offset = i01 + i00 * ne01 + i02 * ne00_blk * ne01;
+
+    global struct block_mxfp4 * b = src0 + src_blk_offset;
+    dst_e[dst_blk_offset] = b->e;
+
+    // extract quantization and unshuffle
+    ushort8 pre_block = ((global ushort8 *)(&(b->qs[0])))[0];
+
+    ushort8 post_block = (ushort8)(0);
+
+    uchar * pre_block_ptr = (uchar *)(&pre_block);
+    uchar * post_block_ptr = (uchar *)(&post_block);
+
+    for (int i = 0; i < QK_MXFP4 / 4; ++i) {
+        uchar x0 = pre_block_ptr[2*i + 0];
+        uchar x1 = pre_block_ptr[2*i + 1];
+
+        post_block_ptr[i + 0        ] = convert_uchar(x0 & 0x0F) | convert_uchar((x1 & 0x0F) << 4);
+        post_block_ptr[i + QK_MXFP4 / 4] = convert_uchar((x0 & 0xF0) >> 4) | convert_uchar(x1 & 0xF0);
+    }
+
+    uint4 q_block = as_uint4(post_block);
+
+    uint offset = i02 * ne00_blk * ne01 * 4 + i00 * ne01 * 4 + i01;
+    dst_q[offset] = q_block.x;
+    dst_q[offset + ne01] = q_block.y;
+    dst_q[offset + ne01 * 2] = q_block.z;
+    dst_q[offset + ne01 * 3] = q_block.w;
+}
+
+kernel void kernel_restore_block_mxfp4_trans4_ns(
+    __global uint * src_q,
+    __global uchar * src_e,
+    __global struct block_mxfp4 * dst0,
+    uint ne00,
+    uint ne01
+) {
+    uint i00 = get_global_id(1);
+    uint i01 = get_global_id(0);
+    uint i02 = get_global_id(2);
+
+    uint ne00_blk = ne00 / QK_MXFP4;
+    uint dst_blk_offset = i00 + i01 * ne00_blk + i02 * ne00_blk * ne01;
+    uint src_d_offset = i01 + i00 * ne01 + i02 * ne00_blk * ne01;
+
+    __global struct block_mxfp4 * b = dst0 + dst_blk_offset;
+    b->e = src_e[src_d_offset];
+
+    // collect transposed quantization parts for a block
+    uint src_q_offset = i02 * ne00_blk * ne01 * 4 + i00 * ne01 * 4 + i01;
+    uint4 q_block;
+    q_block.x = src_q[src_q_offset];
+    q_block.y = src_q[src_q_offset + ne01];
+    q_block.z = src_q[src_q_offset + ne01 * 2];
+    q_block.w = src_q[src_q_offset + ne01 * 3];
+
+    ushort8 post_block = as_ushort8(q_block);
+    ushort8 pre_block = (ushort8)(0);
+
+    uchar * pre_block_ptr = (uchar *)(&pre_block);
+    uchar * post_block_ptr = (uchar *)(&post_block);
+
+    for (int i = 0; i < QK_MXFP4 / 4; ++i) {
+        uchar x0 = post_block_ptr[i + 0];
+        uchar x1 = post_block_ptr[i + QK_MXFP4 / 4];
+
+        pre_block_ptr[2 * i + 0] = convert_uchar(x0 & 0x0F) | convert_uchar((x1 & 0x0F) << 4);
+        pre_block_ptr[2 * i + 1] = convert_uchar((x0 & 0xF0) >> 4) | convert_uchar(x1 & 0xF0);
+    }
+
+    ((__global ushort8 *)(&(b->qs[0])))[0] = pre_block;
+}
+
+
 //------------------------------------------------------------------------------
 // block_q8_0
 //------------------------------------------------------------------------------

ggml/src/ggml-opencl/kernels/gemm_moe_mxfp4_f32_ns.cl

@@ -0,0 +1,302 @@
+#pragma OPENCL EXTENSION cl_khr_fp16 : enable
+#pragma OPENCL EXTENSION cl_khr_subgroups : enable
+#pragma OPENCL EXTENSION cl_qcom_subgroup_uniform_load: enable
+#pragma OPENCL EXTENSION cl_qcom_subgroup_constant_load: enable
+#pragma OPENCL EXTENSION cl_qcom_extra_vector_types : enable
+
+#define TILESIZE_K 16
+#define TILESIZE_M 64
+#define TILESIZE_N 32
+
+
+static inline half8 mxfp4_to_fp16_packed8(ushort2 fp4x8) {
+    ushort2 fp16_packed_a_0, fp16_packed_b_0, bias_a, bias_b, sign_a, sign_b;
+    fp16_packed_a_0.lo = (fp4x8.s0 << 9) & 0x0E00;
+    fp16_packed_a_0.hi = (fp4x8.s0 << 5) & 0x0E00;
+    fp16_packed_b_0.lo = (fp4x8.s0 << 1) & 0x0E00;
+    fp16_packed_b_0.hi = (fp4x8.s0 >> 3) & 0x0E00;
+
+    bias_a.lo = (fp16_packed_a_0.lo != 0) ? 0x3800 : 0x0;
+    bias_a.hi = (fp16_packed_a_0.hi != 0) ? 0x3800 : 0x0;
+    bias_b.lo = (fp16_packed_b_0.lo != 0) ? 0x3800 : 0x0;
+    bias_b.hi = (fp16_packed_b_0.hi != 0) ? 0x3800 : 0x0;
+
+    fp16_packed_a_0.lo = (fp16_packed_a_0.lo != 0x0200) ? fp16_packed_a_0.lo : 0x0;
+    fp16_packed_a_0.hi = (fp16_packed_a_0.hi != 0x0200) ? fp16_packed_a_0.hi : 0x0;
+    fp16_packed_b_0.lo = (fp16_packed_b_0.lo != 0x0200) ? fp16_packed_b_0.lo : 0x0;
+    fp16_packed_b_0.hi = (fp16_packed_b_0.hi != 0x0200) ? fp16_packed_b_0.hi : 0x0;
+
+    sign_a.lo = (fp4x8.s0 << 12) & 0x8000;
+    sign_a.hi = (fp4x8.s0 << 8) & 0x8000;
+    sign_b.lo = (fp4x8.s0 << 4) & 0x8000;
+    sign_b.hi = fp4x8.s0 & 0x8000;
+
+    fp16_packed_a_0 = sign_a + bias_a + fp16_packed_a_0;
+    fp16_packed_b_0 = sign_b + bias_b + fp16_packed_b_0;
+
+    ushort2 fp16_packed_a_1, fp16_packed_b_1;
+    fp16_packed_a_1.lo = (fp4x8.s1 << 9) & 0x0E00;
+    fp16_packed_a_1.hi = (fp4x8.s1 << 5) & 0x0E00;
+    fp16_packed_b_1.lo = (fp4x8.s1 << 1) & 0x0E00;
+    fp16_packed_b_1.hi = (fp4x8.s1 >> 3) & 0x0E00;
+
+    bias_a.lo = (fp16_packed_a_1.lo != 0) ? 0x3800 : 0x0;
+    bias_a.hi = (fp16_packed_a_1.hi != 0) ? 0x3800 : 0x0;
+    bias_b.lo = (fp16_packed_b_1.lo != 0) ? 0x3800 : 0x0;
+    bias_b.hi = (fp16_packed_b_1.hi != 0) ? 0x3800 : 0x0;
+
+    fp16_packed_a_1.lo = (fp16_packed_a_1.lo != 0x0200) ? fp16_packed_a_1.lo : 0x0;
+    fp16_packed_a_1.hi = (fp16_packed_a_1.hi != 0x0200) ? fp16_packed_a_1.hi : 0x0;
+    fp16_packed_b_1.lo = (fp16_packed_b_1.lo != 0x0200) ? fp16_packed_b_1.lo : 0x0;
+    fp16_packed_b_1.hi = (fp16_packed_b_1.hi != 0x0200) ? fp16_packed_b_1.hi : 0x0;
+
+    sign_a.lo = (fp4x8.s1 << 12) & 0x8000;
+    sign_a.hi = (fp4x8.s1 << 8) & 0x8000;
+    sign_b.lo = (fp4x8.s1 << 4) & 0x8000;
+    sign_b.hi = fp4x8.s1 & 0x8000;
+
+    fp16_packed_a_1 = sign_a + bias_a + fp16_packed_a_1;
+    fp16_packed_b_1 = sign_b + bias_b + fp16_packed_b_1;
+
+    return as_half8((ushort8)(fp16_packed_a_0, fp16_packed_b_0, fp16_packed_a_1, fp16_packed_b_1));
+}
+
+
+#define dotx16_reduce8(a_reg, b_lm, c_reg, lm_offset) \
+    acc.s0 = dot(a_reg.s0123, b_lm[lm_offset + 0]); \
+    acc.s1 = dot(a_reg.s0123, b_lm[lm_offset + 1]); \
+    acc.s2 = dot(a_reg.s0123, b_lm[lm_offset + 2]); \
+    acc.s3 = dot(a_reg.s0123, b_lm[lm_offset + 3]); \
+    acc.s4 = dot(a_reg.s0123, b_lm[lm_offset + 4]); \
+    acc.s5 = dot(a_reg.s0123, b_lm[lm_offset + 5]); \
+    acc.s6 = dot(a_reg.s0123, b_lm[lm_offset + 6]); \
+    acc.s7 = dot(a_reg.s0123, b_lm[lm_offset + 7]); \
+    acc.s8 = dot(a_reg.s0123, b_lm[lm_offset + 8]); \
+    acc.s9 = dot(a_reg.s0123, b_lm[lm_offset + 9]); \
+    acc.sa = dot(a_reg.s0123, b_lm[lm_offset + 10]); \
+    acc.sb = dot(a_reg.s0123, b_lm[lm_offset + 11]); \
+    acc.sc = dot(a_reg.s0123, b_lm[lm_offset + 12]); \
+    acc.sd = dot(a_reg.s0123, b_lm[lm_offset + 13]); \
+    acc.se = dot(a_reg.s0123, b_lm[lm_offset + 14]); \
+    acc.sf = dot(a_reg.s0123, b_lm[lm_offset + 15]); \
+    acc.s0 += dot(a_reg.s4567, b_lm[lm_offset + 32]); \
+    acc.s1 += dot(a_reg.s4567, b_lm[lm_offset + 33]); \
+    acc.s2 += dot(a_reg.s4567, b_lm[lm_offset + 34]); \
+    acc.s3 += dot(a_reg.s4567, b_lm[lm_offset + 35]); \
+    acc.s4 += dot(a_reg.s4567, b_lm[lm_offset + 36]); \
+    acc.s5 += dot(a_reg.s4567, b_lm[lm_offset + 37]); \
+    acc.s6 += dot(a_reg.s4567, b_lm[lm_offset + 38]); \
+    acc.s7 += dot(a_reg.s4567, b_lm[lm_offset + 39]); \
+    acc.s8 += dot(a_reg.s4567, b_lm[lm_offset + 40]); \
+    acc.s9 += dot(a_reg.s4567, b_lm[lm_offset + 41]); \
+    acc.sa += dot(a_reg.s4567, b_lm[lm_offset + 42]); \
+    acc.sb += dot(a_reg.s4567, b_lm[lm_offset + 43]); \
+    acc.sc += dot(a_reg.s4567, b_lm[lm_offset + 44]); \
+    acc.sd += dot(a_reg.s4567, b_lm[lm_offset + 45]); \
+    acc.se += dot(a_reg.s4567, b_lm[lm_offset + 46]); \
+    acc.sf += dot(a_reg.s4567, b_lm[lm_offset + 47]); \
+    c_reg.lo += convert_float8(acc.lo); \
+    c_reg.hi += convert_float8(acc.hi); \
+    acc.s0 = dot(a_reg.s89ab, b_lm[lm_offset + 64]); \
+    acc.s1 = dot(a_reg.s89ab, b_lm[lm_offset + 65]); \
+    acc.s2 = dot(a_reg.s89ab, b_lm[lm_offset + 66]); \
+    acc.s3 = dot(a_reg.s89ab, b_lm[lm_offset + 67]); \
+    acc.s4 = dot(a_reg.s89ab, b_lm[lm_offset + 68]); \
+    acc.s5 = dot(a_reg.s89ab, b_lm[lm_offset + 69]); \
+    acc.s6 = dot(a_reg.s89ab, b_lm[lm_offset + 70]); \
+    acc.s7 = dot(a_reg.s89ab, b_lm[lm_offset + 71]); \
+    acc.s8 = dot(a_reg.s89ab, b_lm[lm_offset + 72]); \
+    acc.s9 = dot(a_reg.s89ab, b_lm[lm_offset + 73]); \
+    acc.sa = dot(a_reg.s89ab, b_lm[lm_offset + 74]); \
+    acc.sb = dot(a_reg.s89ab, b_lm[lm_offset + 75]); \
+    acc.sc = dot(a_reg.s89ab, b_lm[lm_offset + 76]); \
+    acc.sd = dot(a_reg.s89ab, b_lm[lm_offset + 77]); \
+    acc.se = dot(a_reg.s89ab, b_lm[lm_offset + 78]); \
+    acc.sf = dot(a_reg.s89ab, b_lm[lm_offset + 79]); \
+    acc.s0 += dot(a_reg.scdef, b_lm[lm_offset + 96]); \
+    acc.s1 += dot(a_reg.scdef, b_lm[lm_offset + 97]); \
+    acc.s2 += dot(a_reg.scdef, b_lm[lm_offset + 98]); \
+    acc.s3 += dot(a_reg.scdef, b_lm[lm_offset + 99]); \
+    acc.s4 += dot(a_reg.scdef, b_lm[lm_offset + 100]); \
+    acc.s5 += dot(a_reg.scdef, b_lm[lm_offset + 101]); \
+    acc.s6 += dot(a_reg.scdef, b_lm[lm_offset + 102]); \
+    acc.s7 += dot(a_reg.scdef, b_lm[lm_offset + 103]); \
+    acc.s8 += dot(a_reg.scdef, b_lm[lm_offset + 104]); \
+    acc.s9 += dot(a_reg.scdef, b_lm[lm_offset + 105]); \
+    acc.sa += dot(a_reg.scdef, b_lm[lm_offset + 106]); \
+    acc.sb += dot(a_reg.scdef, b_lm[lm_offset + 107]); \
+    acc.sc += dot(a_reg.scdef, b_lm[lm_offset + 108]); \
+    acc.sd += dot(a_reg.scdef, b_lm[lm_offset + 109]); \
+    acc.se += dot(a_reg.scdef, b_lm[lm_offset + 110]); \
+    acc.sf += dot(a_reg.scdef, b_lm[lm_offset + 111]); \
+    c_reg.lo += convert_float8(acc.lo); \
+    c_reg.hi += convert_float8(acc.hi); \
+
+
+static inline half e8m0_to_fp16(uchar x) {
+    ushort bits;
+    bits = (ushort)(x) - (ushort)(112);
+    bits = ((bits & 0x00E0) != 0) ? 0x7C00 : (bits << 10);
+    return as_half(bits);
+}
+
+static inline float e8m0_to_fp32(uchar x) {
+    int bits;
+    bits = (x == 0) ? 0x00400000 : ((uint) x << 23);
+    return as_float(bits);
+}
+
+
+__attribute__((qcom_wave_pair_mode(1))) // 1=force single 2=force pair
+kernel void kernel_gemm_moe_mxfp4_f32_ns(
+        __read_only  image1d_buffer_t src0_q,
+        __global     uchar *          src0_d,
+        __read_only  image1d_buffer_t src1,
+        __global     uint *           src2,
+        __global     ushort *         src2_emap,
+        __write_only image1d_buffer_t dst,
+        __global     int *            total_tiles,
+        uint ne00,
+        uint ne01
+) {
+    uint block_id_m = get_global_id(1); // m_tile
+    uint block_id_n = get_global_id(2); // n_tile
+
+    // Boundary check
+    if (((get_global_id(0) + block_id_m * TILESIZE_M) >= ne01) || (block_id_n >= total_tiles[0])) {
+        return;
+    }
+
+    __private half16 reg_a;
+    __private float32 reg_c = (float32)(0);
+    __local half4 shared_b[128];
+
+    const ushort expert_id = src2_emap[block_id_n];
+
+    const uint row = block_id_m * TILESIZE_M;
+    const uint col = block_id_n * TILESIZE_N;
+
+    uint sub_block_id_m = get_local_id(0);
+    uint2 b_global_offset;
+    b_global_offset.x = ((sub_block_id_m & 3) << 2) + (sub_block_id_m >> 2) * ne00;
+    b_global_offset.y = b_global_offset.x + (16 * ne00);
+    uint2 b_local_offset;
+    b_local_offset.x = (sub_block_id_m & 3) * 32 + (sub_block_id_m >> 2);
+    b_local_offset.y = b_local_offset.x + 16;
+
+    // Loop along K axis, 32 elements (one block) for each iteration, divided into 2 sub-blocks
+    for (uint step = 0; step < ne00; step += TILESIZE_K * 2) {
+        // First sub-block
+        uint q_sub_offset = row + ((ne01 * step) >> 3) + ((expert_id * ne00 * ne01) >> 3);
+        uint s_sub_offset = row + ((ne01 * step) >> 5) + ((expert_id * ne00 * ne01) >> 5);
+        uint b_sub_offset = col * ne00 + step;
+
+        // Load scale for current mxfp4 block
+        uint s_offset = s_sub_offset + get_global_id(0);
+        float s = e8m0_to_fp32(src0_d[s_offset]);
+
+        // Load 16 fp4 (64-bits) in transposed layout
+        uint2 mxfp4x16;
+        mxfp4x16.x = read_imageui(src0_q, q_sub_offset + sub_block_id_m).x;
+        mxfp4x16.y = read_imageui(src0_q, q_sub_offset + sub_block_id_m + ne01).x;
+
+        // Load 16x32 floats from matrix B, each fiber out of 64 in a sub-group loads 8 elements
+        float8 bx8_f32;
+        bx8_f32.lo = read_imagef(src1, (b_sub_offset + b_global_offset.x) / 4);
+        bx8_f32.hi = read_imagef(src1, (b_sub_offset + b_global_offset.y) / 4);
+        // Convert to half and store to LM to share within the subgroup
+        half8 bx8_f16 = convert_half8(bx8_f32);
+        shared_b[b_local_offset.x] = bx8_f16.lo;
+        shared_b[b_local_offset.y] = bx8_f16.hi;
+
+        // Dequantization
+        reg_a.lo = mxfp4_to_fp16_packed8(as_ushort2(mxfp4x16.lo)) * s;
+        reg_a.hi = mxfp4_to_fp16_packed8(as_ushort2(mxfp4x16.hi)) * s;
+
+        sub_group_barrier(CLK_LOCAL_MEM_FENCE);
+
+        // 32 16x16 fp16 dot product with 8 elements reduction for better precision
+        half16 acc;
+        dotx16_reduce8(reg_a, shared_b, reg_c.lo, 0);
+        dotx16_reduce8(reg_a, shared_b, reg_c.hi, 16);
+
+        // Repeat for second sub-block
+        uint half_step = step + TILESIZE_K;
+        q_sub_offset = row + ((ne01 * half_step) >> 3) + ((expert_id * ne00 * ne01) >> 3);
+        b_sub_offset = col * ne00 + half_step;
+
+        // Load next 16 fp4 (64-bits) in transposed layout
+        mxfp4x16.x = read_imageui(src0_q, q_sub_offset + sub_block_id_m).x;
+        mxfp4x16.y = read_imageui(src0_q, q_sub_offset + sub_block_id_m + ne01).x;
+
+        // Load 16x32 floats from matrix B, each fiber out of 64 in a sub-group loads 8 elements
+        bx8_f32.lo = read_imagef(src1, (b_sub_offset + b_global_offset.x) / 4);
+        bx8_f32.hi = read_imagef(src1, (b_sub_offset + b_global_offset.y) / 4);
+        // Convert to half and store to LM to share within the subgroup
+        bx8_f16 = convert_half8(bx8_f32);
+        shared_b[b_local_offset.x] = bx8_f16.lo;
+        shared_b[b_local_offset.y] = bx8_f16.hi;
+
+        // Dequantization
+        reg_a.lo = mxfp4_to_fp16_packed8(as_ushort2(mxfp4x16.lo)) * s;
+        reg_a.hi = mxfp4_to_fp16_packed8(as_ushort2(mxfp4x16.hi)) * s;
+
+        sub_group_barrier(CLK_LOCAL_MEM_FENCE);
+
+        // 32 16x16 fp16 dot product with 3-levels reduction for better precision
+        dotx16_reduce8(reg_a, shared_b, reg_c.lo, 0);
+        dotx16_reduce8(reg_a, shared_b, reg_c.hi, 16);
+    }
+
+    // Load poster router and share in LM
+    __local uint out_idx[TILESIZE_N];
+
+    if (get_local_id(0) < TILESIZE_N) {
+        uint idx = src2[block_id_n * TILESIZE_N + get_local_id(0)];
+        if (idx == 0xFFFFFFFF) {
+            idx = src2[block_id_n * TILESIZE_N + 0];
+        }
+        out_idx[get_local_id(0)] = idx * ne01;
+    }
+
+    barrier(CLK_LOCAL_MEM_FENCE);
+
+    // Scatter results back to original position in output grid
+    uint m_offset = row + get_local_id(0);
+
+    write_imagef(dst, out_idx[1] + m_offset, (reg_c.s1));
+    write_imagef(dst, out_idx[2] + m_offset, (reg_c.s2));
+    write_imagef(dst, out_idx[3] + m_offset, (reg_c.s3));
+    write_imagef(dst, out_idx[4] + m_offset, (reg_c.s4));
+    write_imagef(dst, out_idx[5] + m_offset, (reg_c.s5));
+    write_imagef(dst, out_idx[6] + m_offset, (reg_c.s6));
+    write_imagef(dst, out_idx[7] + m_offset, (reg_c.s7));
+    write_imagef(dst, out_idx[8] + m_offset, (reg_c.s8));
+    write_imagef(dst, out_idx[9] + m_offset, (reg_c.s9));
+    write_imagef(dst, out_idx[10] + m_offset, (reg_c.sa));
+    write_imagef(dst, out_idx[11] + m_offset, (reg_c.sb));
+    write_imagef(dst, out_idx[12] + m_offset, (reg_c.sc));
+    write_imagef(dst, out_idx[13] + m_offset, (reg_c.sd));
+    write_imagef(dst, out_idx[14] + m_offset, (reg_c.se));
+    write_imagef(dst, out_idx[15] + m_offset, (reg_c.sf));
+    write_imagef(dst, out_idx[16] + m_offset, (reg_c.sg));
+    write_imagef(dst, out_idx[17] + m_offset, (reg_c.sh));
+    write_imagef(dst, out_idx[18] + m_offset, (reg_c.si));
+    write_imagef(dst, out_idx[19] + m_offset, (reg_c.sj));
+    write_imagef(dst, out_idx[20] + m_offset, (reg_c.sk));
+    write_imagef(dst, out_idx[21] + m_offset, (reg_c.sl));
+    write_imagef(dst, out_idx[22] + m_offset, (reg_c.sm));
+    write_imagef(dst, out_idx[23] + m_offset, (reg_c.sn));
+    write_imagef(dst, out_idx[24] + m_offset, (reg_c.so));
+    write_imagef(dst, out_idx[25] + m_offset, (reg_c.sp));
+    write_imagef(dst, out_idx[26] + m_offset, (reg_c.sq));
+    write_imagef(dst, out_idx[27] + m_offset, (reg_c.sr));
+    write_imagef(dst, out_idx[28] + m_offset, (reg_c.ss));
+    write_imagef(dst, out_idx[29] + m_offset, (reg_c.st));
+    write_imagef(dst, out_idx[30] + m_offset, (reg_c.su));
+    write_imagef(dst, out_idx[31] + m_offset, (reg_c.sv));
+
+    // Store zero padding parts to the index of first output in tile, override correct result in the end
+    barrier(CLK_GLOBAL_MEM_FENCE);
+    write_imagef(dst, out_idx[0] + m_offset, (reg_c.s0));
+}

ggml/src/ggml-opencl/kernels/gemm_noshuffle_q4_0_f32.cl

@@ -17,7 +17,7 @@
 REQD_SUBGROUP_SIZE_128
 #endif
 
-kernel void kernel_mul_mat_Ab_Bi_8x4(
+kernel void kernel_gemm_noshuffle_q4_0_f32(
         global const ushort * src0_q,       // quantized A
         global const half  * src0_d,        // A scales
         __read_only image1d_buffer_t src1,  // B (1d image)

ggml/src/ggml-opencl/kernels/gemm_noshuffle_q8_0_f32.cl

@@ -11,7 +11,7 @@
 REQD_SUBGROUP_SIZE_128
 #endif
 
-kernel void kernel_mul_mm_q8_0_f32_8x4(
+kernel void kernel_gemm_noshuffle_q8_0_f32(
         global const uint * src0_q,
         global const half  * src0_d,
         __read_only image1d_buffer_t src1,

ggml/src/ggml-opencl/kernels/gemv_moe_mxfp4_f32_ns.cl

@@ -0,0 +1,161 @@
+#pragma OPENCL EXTENSION cl_khr_fp16 : enable
+#pragma OPENCL EXTENSION cl_khr_subgroups : enable
+#pragma OPENCL EXTENSION cl_qcom_reqd_sub_group_size : enable
+
+#define QK_MXFP4 32
+#define N_SIMDGROUP 4
+#define SIMDGROUP_WIDTH 64
+
+static inline half8 mxfp4_to_fp16_packed8(ushort2 fp4x8) {
+    ushort2 fp16_packed_a_0, fp16_packed_b_0, bias_a, bias_b, sign_a, sign_b;
+    fp16_packed_a_0.lo = (fp4x8.s0 << 9) & 0x0E00;
+    fp16_packed_a_0.hi = (fp4x8.s0 << 5) & 0x0E00;
+    fp16_packed_b_0.lo = (fp4x8.s0 << 1) & 0x0E00;
+    fp16_packed_b_0.hi = (fp4x8.s0 >> 3) & 0x0E00;
+
+    bias_a.lo = (fp16_packed_a_0.lo != 0) ? 0x3800 : 0x0;
+    bias_a.hi = (fp16_packed_a_0.hi != 0) ? 0x3800 : 0x0;
+    bias_b.lo = (fp16_packed_b_0.lo != 0) ? 0x3800 : 0x0;
+    bias_b.hi = (fp16_packed_b_0.hi != 0) ? 0x3800 : 0x0;
+
+    fp16_packed_a_0.lo = (fp16_packed_a_0.lo != 0x0200) ? fp16_packed_a_0.lo : 0x0;
+    fp16_packed_a_0.hi = (fp16_packed_a_0.hi != 0x0200) ? fp16_packed_a_0.hi : 0x0;
+    fp16_packed_b_0.lo = (fp16_packed_b_0.lo != 0x0200) ? fp16_packed_b_0.lo : 0x0;
+    fp16_packed_b_0.hi = (fp16_packed_b_0.hi != 0x0200) ? fp16_packed_b_0.hi : 0x0;
+
+    sign_a.lo = (fp4x8.s0 << 12) & 0x8000;
+    sign_a.hi = (fp4x8.s0 << 8) & 0x8000;
+    sign_b.lo = (fp4x8.s0 << 4) & 0x8000;
+    sign_b.hi = fp4x8.s0 & 0x8000;
+
+    fp16_packed_a_0 = sign_a + bias_a + fp16_packed_a_0;
+    fp16_packed_b_0 = sign_b + bias_b + fp16_packed_b_0;
+
+    ushort2 fp16_packed_a_1, fp16_packed_b_1;
+    fp16_packed_a_1.lo = (fp4x8.s1 << 9) & 0x0E00;
+    fp16_packed_a_1.hi = (fp4x8.s1 << 5) & 0x0E00;
+    fp16_packed_b_1.lo = (fp4x8.s1 << 1) & 0x0E00;
+    fp16_packed_b_1.hi = (fp4x8.s1 >> 3) & 0x0E00;
+
+    bias_a.lo = (fp16_packed_a_1.lo != 0) ? 0x3800 : 0x0;
+    bias_a.hi = (fp16_packed_a_1.hi != 0) ? 0x3800 : 0x0;
+    bias_b.lo = (fp16_packed_b_1.lo != 0) ? 0x3800 : 0x0;
+    bias_b.hi = (fp16_packed_b_1.hi != 0) ? 0x3800 : 0x0;
+
+    fp16_packed_a_1.lo = (fp16_packed_a_1.lo != 0x0200) ? fp16_packed_a_1.lo : 0x0;
+    fp16_packed_a_1.hi = (fp16_packed_a_1.hi != 0x0200) ? fp16_packed_a_1.hi : 0x0;
+    fp16_packed_b_1.lo = (fp16_packed_b_1.lo != 0x0200) ? fp16_packed_b_1.lo : 0x0;
+    fp16_packed_b_1.hi = (fp16_packed_b_1.hi != 0x0200) ? fp16_packed_b_1.hi : 0x0;
+
+    sign_a.lo = (fp4x8.s1 << 12) & 0x8000;
+    sign_a.hi = (fp4x8.s1 << 8) & 0x8000;
+    sign_b.lo = (fp4x8.s1 << 4) & 0x8000;
+    sign_b.hi = fp4x8.s1 & 0x8000;
+
+    fp16_packed_a_1 = sign_a + bias_a + fp16_packed_a_1;
+    fp16_packed_b_1 = sign_b + bias_b + fp16_packed_b_1;
+
+    return as_half8((ushort8)(fp16_packed_a_0, fp16_packed_b_0, fp16_packed_a_1, fp16_packed_b_1));
+}
+
+static inline float e8m0_to_fp32(uchar x) {
+    int bits;
+    bits = (x == 0) ? 0x00400000 : ((uint) x << 23);
+    return as_float(bits);
+}
+
+
+__attribute__((qcom_reqd_sub_group_size("half")))
+__kernel void kernel_gemv_moe_mxfp4_f32_ns(
+    __global uint * src0_q,
+    __global uchar * src0_e,
+    __read_only image1d_buffer_t src1,
+    __global uint * src2,
+    __global float * dst,
+    ulong         offsetd,
+    int           ne00,
+    int           ne01,
+    int           ne11
+) {
+    uint i01  = get_global_id(0);
+    uint i20  = get_global_id(2);
+    uint sgid = get_local_id(1);
+    uint slid = get_sub_group_local_id();
+
+    uint i11 = i20 % ne11;
+
+    uint expert_id = src2[i20];
+    uint expert_offset = expert_id * ne00 * ne01 / 32;
+
+    __private float sum = 0.0f; // each thread calculate partial sum of one output
+
+    // loop along ne00 in block granularity, skip 4 blocks every iter
+    for (uint ib00 = sgid; ib00 < (ne00 / QK_MXFP4); ib00 += N_SIMDGROUP) {
+
+        // load one block of q
+        uint4 regQ;
+        uint block_offset = expert_offset * 4 + ib00 * ne01 * 4 + i01;
+
+        regQ.s0 = src0_q[block_offset];
+        regQ.s1 = src0_q[block_offset + ne01];
+        regQ.s2 = src0_q[block_offset + ne01 * 2];
+        regQ.s3 = src0_q[block_offset + ne01 * 3];
+
+        uint offset = i11 * ne00 / 4 + ib00 * 8;
+
+        half8 fp16x8 = mxfp4_to_fp16_packed8(as_ushort2(regQ.s0));
+
+        float4 shared_y4;
+        shared_y4 = read_imagef(src1, (offset + 0));
+        float4 acc = shared_y4 * convert_float4(fp16x8.lo);
+
+        shared_y4 = read_imagef(src1, (offset + 1));
+        acc += shared_y4 * convert_float4(fp16x8.hi);
+
+        fp16x8 = mxfp4_to_fp16_packed8(as_ushort2(regQ.s1));
+
+        shared_y4 = read_imagef(src1, (offset + 2));
+        acc += shared_y4 * convert_float4(fp16x8.lo);
+
+        shared_y4 = read_imagef(src1, (offset + 3));
+        acc += shared_y4 * convert_float4(fp16x8.hi);
+
+
+        fp16x8 = mxfp4_to_fp16_packed8(as_ushort2(regQ.s2));
+
+        shared_y4 = read_imagef(src1, (offset + 4));
+        acc += shared_y4 * convert_float4(fp16x8.lo);
+
+        shared_y4 = read_imagef(src1, (offset + 5));
+        acc += shared_y4 * convert_float4(fp16x8.hi);
+
+
+        fp16x8 = mxfp4_to_fp16_packed8(as_ushort2(regQ.s3));
+
+        shared_y4 = read_imagef(src1, (offset + 6));
+        acc += shared_y4 * convert_float4(fp16x8.lo);
+
+        shared_y4 = read_imagef(src1, (offset + 7));
+        acc += shared_y4 * convert_float4(fp16x8.hi);
+
+        uchar regE = src0_e[ib00 * ne01 + i01 + expert_offset];
+        sum += e8m0_to_fp32(regE) * ((acc.s0 + acc.s1) + (acc.s2 + acc.s3));
+    }
+
+    // reduction in local memory, assumes #subgroups=4
+    __local float reduceLM[SIMDGROUP_WIDTH * (N_SIMDGROUP - 1)];
+    if (sgid == 1) reduceLM[SIMDGROUP_WIDTH * 0 + slid] = sum;
+    if (sgid == 2) reduceLM[SIMDGROUP_WIDTH * 1 + slid] = sum;
+    if (sgid == 3) reduceLM[SIMDGROUP_WIDTH * 2 + slid] = sum;
+    barrier(CLK_LOCAL_MEM_FENCE);
+    if (sgid == 0) sum += reduceLM[SIMDGROUP_WIDTH * 0 + slid];
+    if (sgid == 0) sum += reduceLM[SIMDGROUP_WIDTH * 1 + slid];
+    if (sgid == 0) sum += reduceLM[SIMDGROUP_WIDTH * 2 + slid];
+
+    // 1 outputs per thread in subgroup 0
+    if (sgid == 0) {
+        dst = dst + (offsetd >> 2);
+        dst[i01 + i20 * ne01] = sum;
+    }
+
+}

ggml/src/ggml-opencl/kernels/gemv_noshuffle_q4_0_f32.cl

@@ -191,7 +191,7 @@
 #ifdef ADRENO_GPU
 REQD_SUBGROUP_SIZE_64
 #endif
-__kernel void kernel_gemv_noshuffle(
+__kernel void kernel_gemv_noshuffle_q4_0_f32(
         __read_only  image1d_buffer_t src0_q,  // quantized A
         global half2  * src0_d,  // A scales
         __read_only  image1d_buffer_t src1,    // B
@@ -238,21 +238,21 @@ __kernel void kernel_gemv_noshuffle(
         regA.s1 = read_imageui(src0_q, (gid + k * BLOCK_STRIDE_A + LINE_STRIDE_A * 1)).x;
         regA.s2 = read_imageui(src0_q, (gid + k * BLOCK_STRIDE_A + LINE_STRIDE_A * 2)).x;
         regA.s3 = read_imageui(src0_q, (gid + k * BLOCK_STRIDE_A + LINE_STRIDE_A * 3)).x;
-#ifdef VECTOR_SUB_GROUP_BROADCAT
+#ifdef VECTOR_SUB_GROUP_BROADCAST
         dequantizeBlockAccum_ns_sgbroadcast_8_hi(totalSum, as_ushort8(regA), regS, regB);
 #else
         dequantizeBlockAccum_ns_sgbroadcast_1_hi(totalSum, as_ushort8(regA), regS, regB);
-#endif // VECTOR_SUB_GROUP_BROADCAT
+#endif // VECTOR_SUB_GROUP_BROADCAST
 
         regA.s0 = read_imageui(src0_q, (gid + k * BLOCK_STRIDE_A + LINE_STRIDE_A * 4)).x;
         regA.s1 = read_imageui(src0_q, (gid + k * BLOCK_STRIDE_A + LINE_STRIDE_A * 5)).x;
         regA.s2 = read_imageui(src0_q, (gid + k * BLOCK_STRIDE_A + LINE_STRIDE_A * 6)).x;
         regA.s3 = read_imageui(src0_q, (gid + k * BLOCK_STRIDE_A + LINE_STRIDE_A * 7)).x;
-#ifdef VECTOR_SUB_GROUP_BROADCAT
+#ifdef VECTOR_SUB_GROUP_BROADCAST
         dequantizeBlockAccum_ns_sgbroadcast_8_lo(totalSum, as_ushort8(regA), regS, regB);
 #else
         dequantizeBlockAccum_ns_sgbroadcast_1_lo(totalSum, as_ushort8(regA), regS, regB);
-#endif // VECTOR_SUB_GROUP_BROADCAT
+#endif // VECTOR_SUB_GROUP_BROADCAST
     }
 
     // reduction in local memory, assumes #wave=4

ggml/src/ggml-opencl/kernels/gemv_noshuffle_q4_0_f32_spec.cl

@@ -191,7 +191,7 @@
 #ifdef ADRENO_GPU
 REQD_SUBGROUP_SIZE_64
 #endif
-__kernel void kernel_gemv_noshuffle(
+__kernel void kernel_gemv_noshuffle_q4_0_f32(
         __read_only  image1d_buffer_t src0_q,  // quantized A
         global half2  * src0_d,  // A scales
         __read_only  image1d_buffer_t src1,    // B
@@ -232,21 +232,21 @@ __kernel void kernel_gemv_noshuffle(
         regA.s1 = read_imageui(src0_q, (gid + k * BLOCK_STRIDE_A + LINE_STRIDE_A * 1)).x;
         regA.s2 = read_imageui(src0_q, (gid + k * BLOCK_STRIDE_A + LINE_STRIDE_A * 2)).x;
         regA.s3 = read_imageui(src0_q, (gid + k * BLOCK_STRIDE_A + LINE_STRIDE_A * 3)).x;
-#ifdef VECTOR_SUB_GROUP_BROADCAT
+#ifdef VECTOR_SUB_GROUP_BROADCAST
         dequantizeBlockAccum_ns_sgbroadcast_8_hi(totalSum, as_ushort8(regA), regS, regB);
 #else
         dequantizeBlockAccum_ns_sgbroadcast_1_hi(totalSum, as_ushort8(regA), regS, regB);
-#endif // VECTOR_SUB_GROUP_BROADCAT
+#endif // VECTOR_SUB_GROUP_BROADCAST
 
         regA.s0 = read_imageui(src0_q, (gid + k * BLOCK_STRIDE_A + LINE_STRIDE_A * 4)).x;
         regA.s1 = read_imageui(src0_q, (gid + k * BLOCK_STRIDE_A + LINE_STRIDE_A * 5)).x;
         regA.s2 = read_imageui(src0_q, (gid + k * BLOCK_STRIDE_A + LINE_STRIDE_A * 6)).x;
         regA.s3 = read_imageui(src0_q, (gid + k * BLOCK_STRIDE_A + LINE_STRIDE_A * 7)).x;
-#ifdef VECTOR_SUB_GROUP_BROADCAT
+#ifdef VECTOR_SUB_GROUP_BROADCAST
         dequantizeBlockAccum_ns_sgbroadcast_8_lo(totalSum, as_ushort8(regA), regS, regB);
 #else
         dequantizeBlockAccum_ns_sgbroadcast_1_lo(totalSum, as_ushort8(regA), regS, regB);
-#endif // VECTOR_SUB_GROUP_BROADCAT
+#endif // VECTOR_SUB_GROUP_BROADCAST
     }
 
     // reduction in local memory, assumes #wave=4

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

@@ -0,0 +1,30 @@
+#pragma OPENCL EXTENSION cl_khr_fp16 : enable
+
+#define QK4_0 32
+
+kernel void kernel_moe_reorder_b(
+    global float4 * src,
+    global uint * router,
+    global float4 * dst,
+    global int * total_tiles,
+    uint K,
+    ushort map_ratio,
+    uint tile_size
+) {
+    uint k_4 = get_global_id(0);
+    uint post_router_idx = get_global_id(1);
+
+    if ((k_4 >= (K / 4)) || (post_router_idx >= total_tiles[0] * tile_size)) {
+        return;
+    }
+
+    uint router_idx = router[post_router_idx];
+
+    float4 out = (float4)(0);
+    if (router_idx != 0xFFFFFFFF) {
+        ushort activation_idx = router_idx / map_ratio;
+        out = src[activation_idx * K / 4 + k_4];
+    }
+
+    dst[post_router_idx * K / 4 + k_4] = out;
+}

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

@@ -0,0 +1,82 @@
+#pragma OPENCL EXTENSION cl_khr_fp16 : enable
+
+__kernel void kernel_moe_histogram(
+    __global const int * input,
+    __global int * hist,
+    uint N,
+    uint topK,
+    uint n_experts
+) {
+    uint n = get_global_id(0);
+    uint k = get_global_id(1);
+
+    if (n >= N || k >= topK) {
+        return;
+    }
+
+    int expert_id = input[n * n_experts + k];
+    atomic_inc(&hist[expert_id]);
+}
+
+__kernel void kernel_moe_scan(
+    __global int * hist,
+    __global int * tile_offset,
+    __global int * total_tiles,
+    __global int * slot_counter,
+    int tile_size,
+    uint n_experts
+) {
+    int offset = 0;
+    for (int v = 0; v < n_experts; v++) {
+        int count = hist[v];
+        int tiles = (count + tile_size - 1) / tile_size;
+        tile_offset[v] = offset;
+        offset += tiles;
+        hist[v] = 0;
+        slot_counter[v] = 0;
+    }
+
+    *total_tiles = offset;
+}
+
+__kernel void kernel_moe_scatter(
+    __global const int * input,
+    __global int * post_router,
+    __global ushort * emap,
+    __global const int * tile_offset,
+    __global int * slot_counter,
+    int N,
+    int topK,
+    uint n_experts
+) {
+    uint n = get_global_id(0);
+    uint k = get_global_id(1);
+
+    if (n >= N || k >= topK) {
+        return;
+    }
+
+    int val = input[n * n_experts + k];
+
+    int local_slot = atomic_inc(&slot_counter[val]);
+
+    int tile_idx  = tile_offset[val] + (local_slot / 32);
+    int lane      = local_slot % 32;
+    int out_pos   = tile_idx * 32 + lane;
+
+    post_router[out_pos] = n * topK + k;
+    emap[tile_idx] = val;
+}
+
+__kernel void kernel_moe_fill(
+    __global int * post_router,
+    __global int * total_tiles,
+    int tile_size
+) {
+    int tile_id = get_global_id(0);
+    int vec_id_in_tile = get_global_id(1);
+
+    if (tile_id < total_tiles[0]) {
+        post_router[tile_id * tile_size + vec_id_in_tile] = 0xFFFFFFFF;
+    }
+}

ggml/src/ggml-rpc/ggml-rpc.cpp

@@ -207,35 +207,11 @@ struct ggml_backend_rpc_buffer_type_context {
     size_t      max_size;
 };
 
-struct graph_cache {
-
-    bool is_cached(const ggml_cgraph * cgraph) {
-        if ((int)last_graph.size() != cgraph->n_nodes) {
-            return false;
-        }
-        for (int i = 0; i < cgraph->n_nodes; i++) {
-            if (memcmp(&last_graph[i], cgraph->nodes[i], sizeof(ggml_tensor)) != 0) {
-                return false;
-            }
-        }
-        return true;
-    }
-
-    void add(const ggml_cgraph * cgraph) {
-        last_graph.resize(cgraph->n_nodes);
-        for (int i = 0; i < cgraph->n_nodes; i++) {
-            memcpy(&last_graph[i], cgraph->nodes[i], sizeof(ggml_tensor));
-        }
-    }
-
-    std::vector<ggml_tensor> last_graph;
-};
-
 struct ggml_backend_rpc_context {
     std::string endpoint;
     uint32_t    device;
     std::string name;
-    graph_cache gc;
+    uint64_t    last_graph_uid;
 };
 
 struct ggml_backend_rpc_buffer_context {
@@ -717,7 +693,7 @@ static enum ggml_status ggml_backend_rpc_graph_compute(ggml_backend_t backend, g
     ggml_backend_rpc_context * rpc_ctx = (ggml_backend_rpc_context *)backend->context;
 
     GGML_ASSERT(cgraph->n_nodes > 0);
-    bool reuse = rpc_ctx->gc.is_cached(cgraph);
+    bool reuse = cgraph->uid != 0 && rpc_ctx->last_graph_uid == cgraph->uid;
     if (reuse) {
         rpc_msg_graph_recompute_req request;
         request.device = rpc_ctx->device;
@@ -725,7 +701,7 @@ static enum ggml_status ggml_backend_rpc_graph_compute(ggml_backend_t backend, g
         bool status = send_rpc_cmd(sock, RPC_CMD_GRAPH_RECOMPUTE, &request, sizeof(request));
         RPC_STATUS_ASSERT(status);
     } else {
-        rpc_ctx->gc.add(cgraph);
+        rpc_ctx->last_graph_uid = cgraph->uid;
         std::vector<uint8_t> input;
         serialize_graph(rpc_ctx->device, cgraph, input);
         auto sock = get_socket(rpc_ctx->endpoint);
@@ -791,10 +767,10 @@ ggml_backend_buffer_type_t ggml_backend_rpc_buffer_type(const char * endpoint, u
 ggml_backend_t ggml_backend_rpc_init(const char * endpoint, uint32_t device) {
     std::string dev_name = "RPC" + std::to_string(device) + "[" + std::string(endpoint) + "]";
     ggml_backend_rpc_context * ctx = new ggml_backend_rpc_context {
-        /* .endpoint = */ endpoint,
-        /* .device   = */ device,
-        /* .name     = */ dev_name,
-        /* .gc       = */ {},
+        /* .endpoint       = */ endpoint,
+        /* .device         = */ device,
+        /* .name           = */ dev_name,
+        /* .last_graph_uid = */ 0,
     };
     auto reg = ggml_backend_rpc_add_server(endpoint);
     ggml_backend_t backend = new ggml_backend {

ggml/src/ggml-virtgpu/virtgpu-shm.cpp

@@ -1,6 +1,7 @@
 #include "virtgpu-shm.h"
 
 #include "virtgpu.h"
+#include "ggml-remoting.h"
 
 #include <assert.h>
 

ggml/src/ggml-virtgpu/virtgpu.cpp

@@ -1,4 +1,5 @@
 #include "virtgpu.h"
+#include "ggml-remoting.h"
 
 #include <stdio.h>
 #include <unistd.h>

ggml/src/ggml-virtgpu/virtgpu.h

@@ -18,8 +18,6 @@
 
 #include <cstring>
 
-#include "ggml-remoting.h"
-
 #define VIRGL_RENDERER_UNSTABLE_APIS 1
 #include "apir_hw.h"
 #include <drm/virtgpu_drm.h>

ggml/src/ggml-vulkan/ggml-vulkan.cpp

@@ -111,8 +111,6 @@ static bool is_pow2(uint32_t x) { return x > 1 && (x & (x-1)) == 0; }
 
 #define VK_DEVICE_DESCRIPTOR_POOL_SIZE 256
 
-#define GGML_VK_MAX_NODES 8192
-
 #define VK_CHECK(err, msg)                                          \
     do {                                                            \
         vk::Result err_ = (err);                                    \
@@ -440,10 +438,12 @@ struct vk_fa_pipeline_state {
     bool f32acc;
     uint32_t flags;
     uint32_t limit_occupancy_shmem;
+    ggml_type k_type;
+    ggml_type v_type;
 
     bool operator<(const vk_fa_pipeline_state &b) const {
-        return std::tie(HSK, HSV, Br, Bc, D_split, row_split, shmem_staging, path, workgroup_size, subgroup_size, aligned, f32acc, flags, limit_occupancy_shmem) <
-               std::tie(b.HSK, b.HSV, b.Br, b.Bc, b.D_split, b.row_split, b.shmem_staging, b.path, b.workgroup_size, b.subgroup_size, b.aligned, b.f32acc, b.flags, b.limit_occupancy_shmem);
+        return std::tie(HSK, HSV, Br, Bc, D_split, row_split, shmem_staging, path, workgroup_size, subgroup_size, aligned, f32acc, flags, limit_occupancy_shmem, k_type, v_type) <
+               std::tie(b.HSK, b.HSV, b.Br, b.Bc, b.D_split, b.row_split, b.shmem_staging, b.path, b.workgroup_size, b.subgroup_size, b.aligned, b.f32acc, b.flags, b.limit_occupancy_shmem, b.k_type, b.v_type);
     }
 };
 
@@ -3041,7 +3041,7 @@ static vk_fa_tuning_params get_fa_tuning_params_coopmat1(const vk_device& device
     return result;
 }
 
-static vk_fa_tuning_params get_fa_tuning_params_coopmat2(const vk_device& device, uint32_t hsk, uint32_t hsv, uint32_t n_rows, uint32_t n_kv, ggml_type kv_type, bool f32acc) {
+static vk_fa_tuning_params get_fa_tuning_params_coopmat2(const vk_device& device, uint32_t hsk, uint32_t hsv, uint32_t n_rows, uint32_t n_kv, ggml_type k_type, ggml_type v_type, bool f32acc) {
     GGML_UNUSED(n_kv);
     GGML_UNUSED(f32acc);
 
@@ -3055,7 +3055,7 @@ static vk_fa_tuning_params get_fa_tuning_params_coopmat2(const vk_device& device
     if (small_rows) {
         result.block_rows = 32;
         result.block_cols = 32;
-    } else if (ggml_is_quantized(kv_type) || hsk >= 256 || hsv >= 256) {
+    } else if (ggml_is_quantized(k_type) || ggml_is_quantized(v_type) || hsk >= 256 || hsv >= 256) {
         result.block_rows = (hsk >= 512 || hsv >= 512) ? 32 : 64;
         result.block_cols = 32;
     } else {
@@ -3069,7 +3069,13 @@ static vk_fa_tuning_params get_fa_tuning_params_coopmat2(const vk_device& device
     return result;
 }
 
-static vk_fa_tuning_params get_fa_tuning_params(const vk_device& device, uint32_t hsk, uint32_t hsv, uint32_t n_rows, uint32_t n_kv, ggml_type kv_type, bool f32acc) {
+static vk_fa_tuning_params get_fa_tuning_params(const vk_device& device, uint32_t hsk, uint32_t hsv, uint32_t n_rows, uint32_t n_kv, ggml_type k_type, ggml_type v_type, bool f32acc) {
+    // Mixed K/V is only implemented on the coopmat2 (flash_attn_cm2) path; never use scalar/cm1.
+    if (k_type != v_type) {
+        GGML_ASSERT(device->coopmat2);
+        return get_fa_tuning_params_coopmat2(device, hsk, hsv, n_rows, n_kv, k_type, v_type, f32acc);
+    }
+
     FaCodePath path = device->coopmat2 ? FA_COOPMAT2 :
                       device->coopmat1_fa_support ? FA_COOPMAT1 : FA_SCALAR;
 
@@ -3081,7 +3087,7 @@ static vk_fa_tuning_params get_fa_tuning_params(const vk_device& device, uint32_
     if (path == FA_COOPMAT1) {
         bool shape_ok = (f32acc && device->coopmat_support_16x16x16_f32acc) ||
                         (!f32acc && device->coopmat_support_16x16x16_f16acc);
-        const vk_fa_tuning_params params = get_fa_tuning_params_coopmat1(device, hsk, hsv, n_rows, n_kv, kv_type, f32acc);
+        const vk_fa_tuning_params params = get_fa_tuning_params_coopmat1(device, hsk, hsv, n_rows, n_kv, k_type, f32acc);
         bool shmem_ok = ggml_vk_flash_attn_coopmat_shmem_support(device, params, hsk, hsv, f32acc);
 
         if (!shape_ok || !shmem_ok) {
@@ -3094,20 +3100,25 @@ static vk_fa_tuning_params get_fa_tuning_params(const vk_device& device, uint32_
         path = FA_SCALAR;
     }
 
+    // Q1_0 K/V is only implemented on coopmat2 (flash_attn_cm2); there is no scalar FA shader for it.
+    if ((k_type == GGML_TYPE_Q1_0 || v_type == GGML_TYPE_Q1_0) && device->coopmat2) {
+        path = FA_COOPMAT2;
+    }
+
     switch (path) {
     case FA_SCALAR:
-        return get_fa_tuning_params_scalar(device, hsk, hsv, n_rows, n_kv, kv_type, f32acc);
+        return get_fa_tuning_params_scalar(device, hsk, hsv, n_rows, n_kv, k_type, f32acc);
     case FA_COOPMAT1:
-        return get_fa_tuning_params_coopmat1(device, hsk, hsv, n_rows, n_kv, kv_type, f32acc);
+        return get_fa_tuning_params_coopmat1(device, hsk, hsv, n_rows, n_kv, k_type, f32acc);
     case FA_COOPMAT2:
-        return get_fa_tuning_params_coopmat2(device, hsk, hsv, n_rows, n_kv, kv_type, f32acc);
+        return get_fa_tuning_params_coopmat2(device, hsk, hsv, n_rows, n_kv, k_type, v_type, f32acc);
     default:
         throw std::runtime_error("unsupported FaCodePath");
     }
 }
 
 static vk_fa_pipeline_state get_fa_pipeline_state(const vk_device& device, const vk_fa_tuning_params& params, uint32_t hsk, uint32_t hsv, bool aligned, bool f32acc,
-                                                  bool use_mask, bool use_mask_opt, bool use_logit_softcap) {
+                                                  bool use_mask, bool use_mask_opt, bool use_logit_softcap, ggml_type k_type, ggml_type v_type) {
     const bool old_amd_windows = device->vendor_id == VK_VENDOR_ID_AMD && device->driver_id == vk::DriverId::eAmdProprietary &&
                                  (device->architecture == AMD_GCN || device->architecture == AMD_RDNA1 || device->architecture == AMD_RDNA2);
 
@@ -3118,12 +3129,32 @@ static vk_fa_pipeline_state get_fa_pipeline_state(const vk_device& device, const
 
     const uint32_t subgroup_size = params.disable_subgroups ? 0 : params.subgroup_size;
 
-    return vk_fa_pipeline_state{hsk, hsv, params.block_rows, params.block_cols, params.d_split, params.row_split, params.shmem_staging, params.path, params.workgroup_size, subgroup_size, aligned, f32acc, flags, params.limit_occupancy_shmem};
+    return vk_fa_pipeline_state{hsk, hsv, params.block_rows, params.block_cols, params.d_split, params.row_split, params.shmem_staging, params.path, params.workgroup_size, subgroup_size, aligned, f32acc, flags, params.limit_occupancy_shmem, k_type, v_type};
 }
 
 static std::vector<uint32_t> get_fa_spec_constants(const vk_fa_pipeline_state& state) {
-    return {state.workgroup_size, state.Br, state.Bc, state.HSK, state.HSV, !state.aligned, state.D_split,
-            state.row_split, state.subgroup_size, state.shmem_staging ? 1u : 0u, state.flags, state.limit_occupancy_shmem};
+    const auto fa_block_bytes = [](ggml_type t) -> uint32_t {
+        // decodeBufF32 uses a block of vec4s for a better memory access pattern.
+        return t == GGML_TYPE_F32 ? 16u : (uint32_t) ggml_type_size(t);
+    };
+    return {
+        /* 0 WorkGroupSize   */ state.workgroup_size,
+        /* 1 Br              */ state.Br,
+        /* 2 Bc              */ state.Bc,
+        /* 3 HSK             */ state.HSK,
+        /* 4 HSV             */ state.HSV,
+        /* 5 Clamp           */ static_cast<uint32_t>(!state.aligned),
+        /* 6 D_split         */ state.D_split,
+        /* 7 row_split       */ state.row_split,
+        /* 8 SubGroupSize    */ state.subgroup_size,
+        /* 9 SHMEM_STAGING   */ state.shmem_staging ? 1u : 0u,
+        /*10 Flags           */ state.flags,
+        /*11 LIMIT_OCCUPANCY_SHMEM */ state.limit_occupancy_shmem,
+        /*12 FaTypeK         */ static_cast<uint32_t>(state.k_type),
+        /*13 FaTypeV         */ static_cast<uint32_t>(state.v_type),
+        /*14 FaBlockBytesK   */ fa_block_bytes(state.k_type),
+        /*15 FaBlockBytesV   */ fa_block_bytes(state.v_type),
+    };
 }
 
 static bool ggml_vk_matmul_shmem_support(const vk_device& device, const std::vector<uint32_t>& warptile, bool mul_mat_id, ggml_type src0_type) {
@@ -3578,16 +3609,35 @@ static void ggml_vk_load_shaders(vk_device& device) {
     }
 #endif
 #if defined(VK_NV_cooperative_matrix2) && defined(GGML_VULKAN_COOPMAT2_GLSLC_SUPPORT)
+#define CREATE_FA_CM2_MIXED() \
+        for (int fa_k_ty = 0; fa_k_ty < (int)GGML_TYPE_COUNT; ++fa_k_ty) { \
+        for (auto &fa : device->pipeline_flash_attn_f32_f16[fa_k_ty]) { \
+            FaCodePath path = fa.first.path; \
+            uint32_t Br = fa.first.Br; \
+            uint32_t Bc = fa.first.Bc; \
+            bool aligned = fa.first.aligned; \
+            bool f32acc = fa.first.f32acc; \
+            if (path == FA_COOPMAT2) { \
+                if (aligned) { \
+                    if (f32acc) { \
+                        ggml_vk_create_pipeline(device, fa.second, "flash_attn_f32_f16_mixed_aligned_f32acc_cm2", flash_attn_f32_f16_mixed_cm2_len, flash_attn_f32_f16_mixed_cm2_data, "main", 7, sizeof(vk_flash_attn_push_constants), {Br, 1, 1}, get_fa_spec_constants(fa.first), Bc, true, false, 0); \
+                    } else { \
+                        ggml_vk_create_pipeline(device, fa.second, "flash_attn_f32_f16_mixed_aligned_f16acc_cm2", flash_attn_f32_f16_mixed_f16acc_cm2_len, flash_attn_f32_f16_mixed_f16acc_cm2_data, "main", 7, sizeof(vk_flash_attn_push_constants), {Br, 1, 1}, get_fa_spec_constants(fa.first), Bc, true, false, 0); \
+                    } \
+                } else { \
+                    if (f32acc) { \
+                        ggml_vk_create_pipeline(device, fa.second, "flash_attn_f32_f16_mixed_f32acc_cm2", flash_attn_f32_f16_mixed_cm2_len, flash_attn_f32_f16_mixed_cm2_data, "main", 7, sizeof(vk_flash_attn_push_constants), {Br, 1, 1}, get_fa_spec_constants(fa.first), 1, true, false, 0); \
+                    } else { \
+                        ggml_vk_create_pipeline(device, fa.second, "flash_attn_f32_f16_mixed_f16acc_cm2", flash_attn_f32_f16_mixed_f16acc_cm2_len, flash_attn_f32_f16_mixed_f16acc_cm2_data, "main", 7, sizeof(vk_flash_attn_push_constants), {Br, 1, 1}, get_fa_spec_constants(fa.first), 1, true, false, 0); \
+                    } \
+                } \
+            } \
+        } \
+        }
     if (device->coopmat2) {
-        CREATE_FA(GGML_TYPE_F32, f32, FA_COOPMAT2, _cm2)
-        CREATE_FA(GGML_TYPE_F16, f16, FA_COOPMAT2, _cm2)
-        CREATE_FA(GGML_TYPE_Q4_0, q4_0, FA_COOPMAT2, _cm2)
-        CREATE_FA(GGML_TYPE_Q4_1, q4_1, FA_COOPMAT2, _cm2)
-        CREATE_FA(GGML_TYPE_Q5_0, q5_0, FA_COOPMAT2, _cm2)
-        CREATE_FA(GGML_TYPE_Q5_1, q5_1, FA_COOPMAT2, _cm2)
-        CREATE_FA(GGML_TYPE_Q8_0, q8_0, FA_COOPMAT2, _cm2)
-        CREATE_FA(GGML_TYPE_IQ4_NL, iq4_nl, FA_COOPMAT2, _cm2)
+        CREATE_FA_CM2_MIXED();
     }
+#undef CREATE_FA_CM2_MIXED
 #endif
 #undef CREATE_FA
 
@@ -9042,8 +9092,6 @@ static void ggml_vk_flash_attn(ggml_backend_vk_context * ctx, vk_context& subctx
 
     assert(dst->type == GGML_TYPE_F32);
     assert(q->type == GGML_TYPE_F32);
-    assert(k->type == v->type);
-
     uint32_t gqa_ratio = 1;
     uint32_t qk_ratio = neq2 / nek2;
     uint32_t workgroups_x = (uint32_t)neq1;
@@ -9054,7 +9102,7 @@ static void ggml_vk_flash_attn(ggml_backend_vk_context * ctx, vk_context& subctx
 
     // For scalar/coopmat1 FA, we can use the "large" size to accommodate qga.
     // For coopmat2 FA, we always use the small size (which is still pretty large for gqa).
-    vk_fa_tuning_params tuning_params = get_fa_tuning_params(ctx->device, HSK, HSV, 512, KV, k->type, f32acc);
+    vk_fa_tuning_params tuning_params = get_fa_tuning_params(ctx->device, HSK, HSV, 512, KV, k->type, v->type, f32acc);
     const uint32_t max_gqa = std::min(tuning_params.block_rows, 32u);
 
     if (N <= 8 && qk_ratio > 1 && qk_ratio <= max_gqa &&
@@ -9067,7 +9115,11 @@ static void ggml_vk_flash_attn(ggml_backend_vk_context * ctx, vk_context& subctx
         workgroups_y /= gqa_ratio;
     }
 
-    tuning_params = get_fa_tuning_params(ctx->device, HSK, HSV, N, KV, k->type, f32acc);
+    tuning_params = get_fa_tuning_params(ctx->device, HSK, HSV, N, KV, k->type, v->type, f32acc);
+
+    if (tuning_params.path != FA_COOPMAT2) {
+        GGML_ASSERT(k->type == v->type);
+    }
 
     const uint32_t q_stride = (uint32_t)(nbq1 / ggml_type_size(q->type));
     uint32_t k_stride = (uint32_t)(nbk1 / ggml_type_size(k->type));
@@ -9106,7 +9158,7 @@ static void ggml_vk_flash_attn(ggml_backend_vk_context * ctx, vk_context& subctx
     // Only use mask opt when the mask is fairly large. This hasn't been tuned extensively.
     bool use_mask_opt = mask && nem1 >= 32 && nem0 * nem1 > 32768 && nem0 >= tuning_params.block_cols * 16;
     vk_fa_pipeline_state fa_pipeline_state = get_fa_pipeline_state(ctx->device, tuning_params, HSK, HSV, aligned, f32acc,
-                                                                   mask != nullptr, use_mask_opt, logit_softcap != 0);
+                                                                   mask != nullptr, use_mask_opt, logit_softcap != 0, k->type, v->type);
 
     vk_pipeline pipeline = nullptr;
 
@@ -15590,38 +15642,27 @@ static bool ggml_backend_vk_device_supports_op(ggml_backend_dev_t dev, const ggm
                 if (op->src[3] && op->src[3]->type != GGML_TYPE_F16) {
                     return false;
                 }
-                // It's straightforward to support different K/V dequant, but would
-                // significantly increase the number of pipelines
-                if (op->src[1]->type != op->src[2]->type) {
+                // mismatching K/V type is currently supported for coopmat2 only.
+                if (op->src[1]->type != op->src[2]->type && !coopmat2) {
                     return false;
                 }
-                switch (op->src[1]->type) {
-                case GGML_TYPE_F16:
-                case GGML_TYPE_F32:
-                case GGML_TYPE_Q4_0:
-                case GGML_TYPE_Q8_0:
-                case GGML_TYPE_Q4_1:
-                case GGML_TYPE_Q5_0:
-                case GGML_TYPE_Q5_1:
-                case GGML_TYPE_IQ4_NL:
-                    // supported in scalar and coopmat2 paths
-                    break;
-                // K dequants currently disabled because D dimension is rounded up to 256 and runs inefficiently
-                //case GGML_TYPE_Q2_K:
-                //case GGML_TYPE_Q3_K:
-                //case GGML_TYPE_Q4_K:
-                //case GGML_TYPE_Q5_K:
-                //case GGML_TYPE_Q6_K:
-                //case GGML_TYPE_IQ1_S:
-                //case GGML_TYPE_IQ1_M:
-                //case GGML_TYPE_IQ2_XXS:
-                //case GGML_TYPE_IQ2_XS:
-                //case GGML_TYPE_IQ2_S:
-                //case GGML_TYPE_IQ3_XXS:
-                //case GGML_TYPE_IQ3_S:
-                //case GGML_TYPE_IQ4_XS:
-
-                default:
+                auto fa_kv_ok = [coopmat2](ggml_type t) {
+                    switch (t) {
+                    case GGML_TYPE_F32:
+                    case GGML_TYPE_F16:
+                    case GGML_TYPE_Q8_0:
+                    case GGML_TYPE_Q5_1:
+                    case GGML_TYPE_Q5_0:
+                    case GGML_TYPE_Q4_1:
+                    case GGML_TYPE_Q4_0:
+                        return true;
+                    case GGML_TYPE_Q1_0:
+                        return coopmat2;
+                    default:
+                        return false;
+                    }
+                };
+                if (!fa_kv_ok(op->src[1]->type) || !fa_kv_ok(op->src[2]->type)) {
                     return false;
                 }
                 if (!coopmat2 && !(device->subgroup_shuffle && device->subgroup_vote)) {

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

@@ -13,6 +13,12 @@ layout (constant_id =  8) const uint32_t SubGroupSize = 32;
 layout (constant_id =  9) const uint32_t SHMEM_STAGING = 0;
 layout (constant_id = 10) const uint32_t Flags = 0;
 layout (constant_id = 11) const uint32_t LIMIT_OCCUPANCY_SHMEM = 0;
+// ggml_type enumerant for K/V
+layout (constant_id = 12) const uint32_t FaTypeK = 0;
+layout (constant_id = 13) const uint32_t FaTypeV = 0;
+// sizeof(decode buffer): quants -> ggml block size; F32 -> 16 (decodeBufF32 vec4).
+layout (constant_id = 14) const uint32_t FaBlockBytesK = 2;
+layout (constant_id = 15) const uint32_t FaBlockBytesV = 2;
 
 const bool USE_MASK_OPT    = (Flags & 1) != 0;
 const bool MASK_ENABLE     = (Flags & 2) != 0;

ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp

@@ -17,8 +17,57 @@
 #extension GL_EXT_null_initializer : enable
 
 #include "types.glsl"
-#include "dequant_funcs_cm2.glsl"
 #include "flash_attn_base.glsl"
+#include "dequant_funcs_cm2.glsl"
+
+// buffer_reference stride = sizeof(struct) = FaBlockBytesK/V.
+layout(buffer_reference, std430, buffer_reference_align = 1) buffer decodeBufFA_K {
+    uint8_t raw[FaBlockBytesK];
+};
+layout(buffer_reference, std430, buffer_reference_align = 1) buffer decodeBufFA_V {
+    uint8_t raw[FaBlockBytesV];
+};
+
+uint fa_block_elems(uint ty) {
+    switch (ty) {
+        case 0u:  return 4u;  // GGML_TYPE_F32: vec4 block (matches decodeBufF32 / dequantFuncF32)
+        case 1u:  return 1u;  // GGML_TYPE_F16
+        case 2u:  return uint(QUANT_K_Q4_0);
+        case 3u:  return uint(QUANT_K_Q4_1);
+        case 6u:  return uint(QUANT_K_Q5_0);
+        case 7u:  return uint(QUANT_K_Q5_1);
+        case 8u:  return uint(QUANT_K_Q8_0);
+        case 41u: return uint(QUANT_K_Q1_0);
+        default:
+            return 1u;
+    }
+}
+
+float16_t faDecodeK(const decodeBufFA_K bl_in, const uint blockCoords[2], const uint coordInBlock[2]) {
+    switch (FaTypeK) {
+        case 0u: return dequantFuncF32(decodeBufF32(bl_in), blockCoords, coordInBlock);
+        case 2u: return dequantFuncQ4_0(decodeBufQ4_0(bl_in), blockCoords, coordInBlock);
+        case 3u: return dequantFuncQ4_1(decodeBufQ4_1(bl_in), blockCoords, coordInBlock);
+        case 6u: return dequantFuncQ5_0(decodeBufQ5_0(bl_in), blockCoords, coordInBlock);
+        case 7u: return dequantFuncQ5_1(decodeBufQ5_1(bl_in), blockCoords, coordInBlock);
+        case 8u: return dequantFuncQ8_0(decodeBufQ8_0(bl_in), blockCoords, coordInBlock);
+        case 41u: return dequantFuncQ1_0(decodeBufQ1_0(bl_in), blockCoords, coordInBlock);
+        default: return float16_t(0);
+    }
+}
+
+float16_t faDecodeV(const decodeBufFA_V bl_in, const uint blockCoords[2], const uint coordInBlock[2]) {
+    switch (FaTypeV) {
+        case 0u: return dequantFuncF32(decodeBufF32(bl_in), blockCoords, coordInBlock);
+        case 2u: return dequantFuncQ4_0(decodeBufQ4_0(bl_in), blockCoords, coordInBlock);
+        case 3u: return dequantFuncQ4_1(decodeBufQ4_1(bl_in), blockCoords, coordInBlock);
+        case 6u: return dequantFuncQ5_0(decodeBufQ5_0(bl_in), blockCoords, coordInBlock);
+        case 7u: return dequantFuncQ5_1(decodeBufQ5_1(bl_in), blockCoords, coordInBlock);
+        case 8u: return dequantFuncQ8_0(decodeBufQ8_0(bl_in), blockCoords, coordInBlock);
+        case 41u: return dequantFuncQ1_0(decodeBufQ1_0(bl_in), blockCoords, coordInBlock);
+        default: return float16_t(0);
+    }
+}
 
 layout (binding = 0) readonly buffer Q {uint8_t data_q[];};
 layout (binding = 1) readonly buffer K {uint8_t data_k[];};
@@ -55,12 +104,6 @@ ACC_TYPE Max(const in uint32_t row, const in uint32_t col, const in ACC_TYPE ele
     return max(elem0, elem1);
 }
 
-#if BLOCK_SIZE > 1
-#define DECODEFUNC , DEQUANTFUNC
-#else
-#define DECODEFUNC
-#endif
-
 // Store the output when doing grouped query attention.
 // Rows index by Q's dimension 2, and the first N rows are valid.
 D_TYPE perElemOpGqaStore(const in uint32_t r, const in uint32_t c, const in D_TYPE elem, const in uint32_t o_offset, const in uint32_t iq2, const in uint32_t N)
@@ -95,10 +138,6 @@ ACC_TYPE perElemOpNonGqaSplitKStoreCol0(const in uint32_t r, const in uint32_t c
 }
 
 void main() {
-#ifdef NEEDS_INIT_IQ_SHMEM
-    init_iq_shmem(gl_WorkGroupSize);
-#endif
-
     init_indices();
 
     tensorLayoutNV<2, gl_CooperativeMatrixClampModeConstantNV> tensorLayoutQ = createTensorLayoutNV(2, gl_CooperativeMatrixClampModeConstantNV);
@@ -107,10 +146,10 @@ void main() {
 
     tensorViewNV<2, false, 1, 0> tensorViewTranspose = createTensorViewNV(2, false, 1, 0);
 
-#if BLOCK_SIZE > 1
-    tensorLayoutK = setTensorLayoutBlockSizeNV(tensorLayoutK, 1, BLOCK_SIZE);
-    tensorLayoutV = setTensorLayoutBlockSizeNV(tensorLayoutV, 1, BLOCK_SIZE);
-#endif
+    const uint bs_k = fa_block_elems(FaTypeK);
+    const uint bs_v = fa_block_elems(FaTypeV);
+    tensorLayoutK = setTensorLayoutBlockSizeNV(tensorLayoutK, 1, bs_k);
+    tensorLayoutV = setTensorLayoutBlockSizeNV(tensorLayoutV, 1, bs_v);
 
     tensorLayoutQ = setTensorLayoutDimensionNV(tensorLayoutQ, N, HSK);
     tensorLayoutK = setTensorLayoutDimensionNV(tensorLayoutK, KV, HSK);
@@ -120,10 +159,12 @@ void main() {
     if (Clamp != gl_CooperativeMatrixClampModeConstantNV)
     {
         q_stride &= ~7;
-#if BLOCK_SIZE == 1
-        k_stride &= ~7;
-        v_stride &= ~7;
-#endif
+        if (bs_k == 1u) {
+            k_stride &= ~7;
+        }
+        if (bs_v == 1u) {
+            v_stride &= ~7;
+        }
         m_stride &= ~7;
     }
     tensorLayoutQ = setTensorLayoutStrideNV(tensorLayoutQ, q_stride, 1);
@@ -230,7 +271,13 @@ void main() {
         coopmat<float16_t, gl_ScopeWorkgroup, HSK_pad, Bc, gl_MatrixUseB> K_T;
 
         uint32_t k_offset = ik2*p.nb12 + ik3*p.nb13;
-        coopMatLoadTensorNV(K_T, data_k, k_offset, sliceTensorLayoutNV(tensorLayoutK, j * Bc, Bc, 0, HSK_pad), tensorViewTranspose DECODEFUNC);
+        // F16: bs_k==1 (direct load). F32: bs_k==4 (vec4 / dequantFuncF32). Q4/Q8 family: bs_k==32. Q1_0: bs_k==128.
+        const bool k_use_decode = (bs_k > 1u);
+        if (k_use_decode) {
+            coopMatLoadTensorNV(K_T, data_k, k_offset, sliceTensorLayoutNV(tensorLayoutK, j * Bc, Bc, 0, HSK_pad), tensorViewTranspose, faDecodeK);
+        } else {
+            coopMatLoadTensorNV(K_T, data_k, k_offset, sliceTensorLayoutNV(tensorLayoutK, j * Bc, Bc, 0, HSK_pad), tensorViewTranspose);
+        }
         S = coopMatMulAdd(Qf16, K_T, S);
 
         if (LOGIT_SOFTCAP) {
@@ -291,7 +338,12 @@ void main() {
 
         coopmat<float16_t, gl_ScopeWorkgroup, Bc, HSV_pad, gl_MatrixUseB> V;
         uint32_t v_offset = iv2*p.nb22 + iv3*p.nb23;
-        coopMatLoadTensorNV(V,  data_v, v_offset, sliceTensorLayoutNV(tensorLayoutV, j * Bc, Bc, 0, HSV_pad) DECODEFUNC);
+        const bool v_use_decode = (bs_v > 1u);
+        if (v_use_decode) {
+            coopMatLoadTensorNV(V, data_v, v_offset, sliceTensorLayoutNV(tensorLayoutV, j * Bc, Bc, 0, HSV_pad), faDecodeV);
+        } else {
+            coopMatLoadTensorNV(V, data_v, v_offset, sliceTensorLayoutNV(tensorLayoutV, j * Bc, Bc, 0, HSV_pad));
+        }
 
         L = eM*L + rowsum;
 

ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp

@@ -641,20 +641,17 @@ void process_shaders() {
                 fa_base_dict["ACC_TYPE_MAX"] = "float16_t(65504.0)";
             }
 
+            if (fp16) {
+#if defined(GGML_VULKAN_COOPMAT2_GLSLC_SUPPORT)
+                string_to_spv("flash_attn_f32_f16_mixed", "flash_attn_cm2.comp",
+                    merge_maps(fa_base_dict, {{"Q_TYPE", "float"}, {"D_TYPE", "float"}, {"D_TYPEV4", "vec4"}}), fp16, false, true, f16acc);
+#endif
+            }
+
             for (const auto& tname : type_names) {
                 if (tname == "bf16") continue;
 
                 if (fp16) {
-#if defined(GGML_VULKAN_COOPMAT2_GLSLC_SUPPORT)
-                if (tname == "f16") {
-                    string_to_spv("flash_attn_f32_f16_" + tname, "flash_attn_cm2.comp",
-                        merge_maps(fa_base_dict, {{"Q_TYPE", "float"}, {"D_TYPE", "float"}, {"D_TYPEV4", "vec4"}}), fp16, false, true, f16acc);
-                } else {
-                    std::string data_a_key = "DATA_A_" + to_uppercase(tname);
-                    string_to_spv("flash_attn_f32_f16_" + tname, "flash_attn_cm2.comp",
-                        merge_maps(fa_base_dict, {{data_a_key, "1"}, {"Q_TYPE", "float"}, {"D_TYPE", "float"}, {"D_TYPEV4", "vec4"}, {"DEQUANTFUNC", "dequantFunc"+to_uppercase(tname) }, {"BLOCK_SIZE", "QUANT_K_"+to_uppercase(tname) }}), fp16, false, true, f16acc);
-                }
-#endif
 #if defined(GGML_VULKAN_COOPMAT_GLSLC_SUPPORT)
                 if (tname == "f16") {
                     string_to_spv("flash_attn_f32_f16_" + tname, "flash_attn_cm1.comp",

ggml/src/ggml-webgpu/ggml-webgpu-shader-lib.hpp

@@ -228,11 +228,13 @@ struct ggml_webgpu_get_rows_pipeline_key_hash {
 /** Row Norm **/
 
 struct ggml_webgpu_row_norm_pipeline_key {
-    ggml_op op;
-    bool    inplace;
+    ggml_op   op;
+    ggml_type src_type;
+    ggml_type dst_type;
+    bool      inplace;
 
     bool operator==(const ggml_webgpu_row_norm_pipeline_key & other) const {
-        return op == other.op && inplace == other.inplace;
+        return op == other.op && src_type == other.src_type && dst_type == other.dst_type && inplace == other.inplace;
     }
 };
 
@@ -240,6 +242,8 @@ struct ggml_webgpu_row_norm_pipeline_key_hash {
     size_t operator()(const ggml_webgpu_row_norm_pipeline_key & key) const {
         size_t seed = 0;
         ggml_webgpu_hash_combine(seed, key.op);
+        ggml_webgpu_hash_combine(seed, key.src_type);
+        ggml_webgpu_hash_combine(seed, key.dst_type);
         ggml_webgpu_hash_combine(seed, key.inplace);
         return seed;
     }
@@ -1097,6 +1101,8 @@ class ggml_webgpu_shader_lib {
     webgpu_pipeline get_row_norm_pipeline(const ggml_webgpu_shader_lib_context & context) {
         ggml_webgpu_row_norm_pipeline_key key = {};
         key.op                                = context.dst->op;
+        key.src_type                          = context.src0->type;
+        key.dst_type                          = context.dst->type;
         key.inplace                           = ggml_webgpu_tensor_equal(context.src0, context.dst);
 
         auto it = row_norm_pipelines.find(key);
@@ -1111,6 +1117,10 @@ class ggml_webgpu_shader_lib {
                 defines.push_back("RMS_NORM");
                 variant = "rms_norm";
                 break;
+            case GGML_OP_NORM:
+                defines.push_back("NORM");
+                variant = "norm";
+                break;
             case GGML_OP_L2_NORM:
                 defines.push_back("L2_NORM");
                 variant = "l2_norm";
@@ -1124,6 +1134,22 @@ class ggml_webgpu_shader_lib {
             variant += "_inplace";
         }
 
+        if (key.src_type == GGML_TYPE_F32) {
+            defines.push_back("SRC_F32");
+            variant += "_src_f32";
+        } else if (key.src_type == GGML_TYPE_F16) {
+            defines.push_back("SRC_F16");
+            variant += "_src_f16";
+        }
+
+        if (key.dst_type == GGML_TYPE_F32) {
+            defines.push_back("DST_F32");
+            variant += "_dst_f32";
+        } else if (key.dst_type == GGML_TYPE_F16) {
+            defines.push_back("DST_F16");
+            variant += "_dst_f16";
+        }
+
         const uint32_t row_norm_wg_size = 128u;
         uint32_t       wg_size          = std::min(context.max_wg_size, row_norm_wg_size);
         defines.push_back(std::string("WG_SIZE=") + std::to_string(wg_size));
@@ -1779,12 +1805,12 @@ class ggml_webgpu_shader_lib {
 
     webgpu_pipeline get_mul_mat_fast_pipeline(const ggml_webgpu_shader_lib_context & context) {
         ggml_webgpu_mul_mat_pipeline_key key = {};
-        key.src0_type                        = context.src0->type;
-        key.src1_type                        = context.src1->type;
-        key.vectorized = (context.src0->ne[0] % 4 == 0 && context.dst->ne[0] % 4 == 0 && context.dst->ne[1] % 4 == 0 &&
-                          (context.src0->type == GGML_TYPE_F32 || context.src0->type == GGML_TYPE_F16)) ?
-                             1 :
-                             0;
+        key.src0_type           = context.src0->type;
+        key.src1_type           = context.src1->type;
+        key.vectorized          = (context.src0->ne[0] % 4 == 0 && context.dst->ne[0] % 4 == 0 &&
+                                   (context.src0->type == GGML_TYPE_F32 || context.src0->type == GGML_TYPE_F16)) ?
+                                      1 :
+                                      0;
         key.use_subgroup_matrix = context.supports_subgroup_matrix;
 
         auto it = mul_mat_fast_pipelines.find(key);
@@ -2143,6 +2169,9 @@ class ggml_webgpu_shader_lib {
 
         // variant suffix for src1 type
         variant += std::string("_") + (context.src1->type == GGML_TYPE_F32 ? "f32" : "f16");
+        if (key.vectorized) {
+            variant += "_vectorized";
+        }
 
         auto processed = preprocessor.preprocess(wgsl_mul_mat_id, defines);
 

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

@@ -2927,6 +2927,7 @@ static std::optional<webgpu_encoded_op> ggml_webgpu_encode(webgpu_context ctx,
             } else {
                 return ggml_webgpu_row_norm(ctx, src0, node);
             }
+        case GGML_OP_NORM:
         case GGML_OP_L2_NORM:
             return ggml_webgpu_row_norm(ctx, src0, node);
         case GGML_OP_ROPE:
@@ -4071,6 +4072,7 @@ static bool ggml_backend_webgpu_device_supports_op(ggml_backend_dev_t dev, const
                 break;
             }
         case GGML_OP_RMS_NORM:
+        case GGML_OP_NORM:
         case GGML_OP_L2_NORM:
             supports_op = op->type == GGML_TYPE_F32 && src0->type == GGML_TYPE_F32;
             break;

ggml/src/ggml-webgpu/wgsl-shaders/row_norm.wgsl

@@ -1,20 +1,17 @@
-#ifdef INPLACE
-fn update(src_offset: u32, dst_offset: u32, scale: f32) {
-    src[dst_offset] = scale * src[src_offset];
-}
+#if defined(SRC_F16) || defined(DST_F16)
+enable f16;
+#endif
 
-@group(0) @binding(1)
-var<uniform> params: Params;
+#ifdef SRC_F16
+#define SRC_TYPE f16
 #else
-fn update(src_offset: u32, dst_offset: u32, scale: f32) {
-    dst[dst_offset] = scale * src[src_offset];
-}
+#define SRC_TYPE f32
+#endif
 
-@group(0) @binding(1)
-var<storage, read_write> dst: array<f32>;
-
-@group(0) @binding(2)
-var<uniform> params: Params;
+#ifdef DST_F16
+#define DST_TYPE f16
+#else
+#define DST_TYPE f32
 #endif
 
 struct Params {
@@ -40,9 +37,20 @@ struct Params {
 };
 
 @group(0) @binding(0)
-var<storage, read_write> src: array<f32>;
+var<storage, read_write> src: array<SRC_TYPE>;
 
-var<workgroup> scratch: array<f32, WG_SIZE>;
+#ifdef INPLACE
+@group(0) @binding(1)
+var<uniform> params: Params;
+#else
+@group(0) @binding(1)
+var<storage, read_write> dst: array<DST_TYPE>;
+
+@group(0) @binding(2)
+var<uniform> params: Params;
+#endif
+
+var<workgroup> scratch: array<f32, WG_SIZE * 2u>;
 
 @compute @workgroup_size(WG_SIZE)
 fn main(@builtin(workgroup_id) wid: vec3<u32>,
@@ -65,34 +73,81 @@ fn main(@builtin(workgroup_id) wid: vec3<u32>,
         if (col >= params.ne0) {
             break;
         }
-        sum += pow(src[i_src_row + col], 2.0);
+        let v = f32(src[i_src_row + col]);
+#ifdef NORM
+        sum += v;
+#else
+        sum += v * v;
+#endif
         col += WG_SIZE;
     }
 
     scratch[lid.x] = sum;
     workgroupBarrier();
-    var offset: u32 = WG_SIZE / 2;
+
+    var offset: u32 = WG_SIZE / 2u;
     while (offset > 0) {
         if (lid.x < offset) {
             scratch[lid.x] += scratch[lid.x + offset];
         }
-        offset = offset / 2;
+        offset /= 2u;
         workgroupBarrier();
     }
     sum = scratch[0];
 
-#ifdef RMS_NORM
+#ifdef NORM
+    let mean = sum / f32(params.ne0);
+    var sq_sum = 0.0f;
+    col = lid.x;
+    for (var j: u32 = 0; j < elems; j++) {
+        if (col >= params.ne0) {
+            break;
+        }
+        let v = f32(src[i_src_row + col]);
+        let d = v - mean;
+        sq_sum += d * d;
+        col += WG_SIZE;
+    }
+
+    workgroupBarrier();
+    scratch[lid.x] = sq_sum;
+    workgroupBarrier();
+    offset = WG_SIZE / 2u;
+    while (offset > 0) {
+        if (lid.x < offset) {
+            scratch[lid.x] += scratch[lid.x + offset];
+        }
+        offset /= 2u;
+        workgroupBarrier();
+    }
+
+    let variance = scratch[0] / f32(params.ne0);
+    let scale = 1.0 / sqrt(variance + params.eps);
+#elif defined(RMS_NORM)
     let scale = 1.0/sqrt(sum/f32(params.ne0) + params.eps);
 #elif defined(L2_NORM)
     let scale = 1.0/max(sqrt(sum), params.eps);
 #endif
 
+#ifdef NORM
+    let mean_val = mean;
+#else
+    let mean_val = 0.0f;
+#endif
+
     col = lid.x;
     for (var j: u32 = 0; j < elems; j++) {
         if (col >= params.ne0) {
             break;
         }
-        update(i_src_row + col, i_dst_row + col, scale);
+        let i_src = i_src_row + col;
+        let i_dst = i_dst_row + col;
+        let v = src[i_src];
+#ifdef INPLACE
+        src[i_dst] = scale * (v - mean_val);
+#else
+        dst[i_dst] = scale * (v - mean_val);
+#endif
         col += WG_SIZE;
     }
 }

ggml/src/ggml.c

@@ -55,8 +55,13 @@
 
 uint64_t ggml_graph_next_uid(void) {
 #ifdef _MSC_VER
+#if defined(_WIN32)
+    static volatile LONG counter = 1;
+    return (uint64_t) InterlockedIncrement(&counter) - 1;
+#else
     static volatile long long counter = 1;
     return (uint64_t) _InterlockedIncrement64(&counter) - 1;
+#endif
 #else
     static uint64_t counter = 1;
     return __atomic_fetch_add(&counter, 1, __ATOMIC_RELAXED);
@@ -3259,6 +3264,16 @@ void ggml_mul_mat_set_prec(
     ggml_set_op_params_i32(a, 0, prec_i32);
 }
 
+void ggml_mul_mat_set_hint(
+        struct ggml_tensor * a,
+        enum ggml_op_hint    hint) {
+    GGML_ASSERT(a->op == GGML_OP_MUL_MAT);
+
+    const int32_t hint_i32 = (int32_t) hint;
+
+    ggml_set_op_params_i32(a, 1, hint_i32);
+}
+
 // ggml_mul_mat_id
 
 /*

include/llama.h

@@ -864,6 +864,9 @@ extern "C" {
 // work only with partial states, such as SWA KV cache or recurrent cache (e.g. Mamba)
 #define LLAMA_STATE_SEQ_FLAGS_PARTIAL_ONLY 1
 
+// keeps the tensor data on device buffers (i.e. not accessible in host memory, but faster save/load)
+#define LLAMA_STATE_SEQ_FLAGS_ON_DEVICE 2
+
     typedef uint32_t llama_state_seq_flags;
 
     LLAMA_API size_t llama_state_seq_get_size_ext(

scripts/server-test-function-call.py

@@ -79,7 +79,7 @@ def print_info(msg):
 # ---------------------------------------------------------------------------
 
 
-def chat_completion(url, messages, tools=None, stream=False):
+def chat_completion(url, messages, tools=None, stream=False, force_tools=False):
     payload = {
         "messages": messages,
         "stream": stream,
@@ -87,7 +87,10 @@ def chat_completion(url, messages, tools=None, stream=False):
     }
     if tools:
         payload["tools"] = tools
-        payload["tool_choice"] = "auto"
+        if force_tools:
+            payload["tool_choice"] = "required"
+        else:
+            payload["tool_choice"] = "auto"
 
     try:
         response = requests.post(url, json=payload, stream=stream)
@@ -160,7 +163,13 @@ def chat_completion(url, messages, tools=None, stream=False):
     return result
 
 
-def run_agentic_loop(url, messages, tools, mock_tool_responses, stream, max_turns=6):
+def all_tools_called(tools, all_tool_calls):
+    all_tool_names = set([tc["function"]["name"] for tc in tools])
+    all_called_tool_names = set([tc["function"]["name"] for tc in all_tool_calls])
+    return all_tool_names == all_called_tool_names
+
+
+def run_agentic_loop(url, messages, tools, mock_tool_responses, stream, max_turns=6, force_tools=False):
     """
     Drive the multi-turn tool-call loop:
       1. Send messages to model.
@@ -172,8 +181,8 @@ def run_agentic_loop(url, messages, tools, mock_tool_responses, stream, max_turn
     msgs = list(messages)
     all_tool_calls: list[dict] = []
 
-    for _ in range(max_turns):
-        result = chat_completion(url, msgs, tools=tools, stream=stream)
+    for t in range(max_turns):
+        result = chat_completion(url, msgs, tools=tools, stream=stream, force_tools=(force_tools and not all_tools_called(tools, all_tool_calls)))
         if result is None:
             return all_tool_calls, None
 
@@ -235,10 +244,10 @@ def run_agentic_loop(url, messages, tools, mock_tool_responses, stream, max_turn
 # ---------------------------------------------------------------------------
 
 
-def run_test(url, test_case, stream):
+def run_test(url, test_case, stream, force_tools):
     name = test_case["name"]
     mode = f"{'stream' if stream else 'non-stream'}"
-    print_header(f"{name}  [{mode}]")
+    print_header(f"{name} [{mode}, force_tools={force_tools}] ")
 
     all_tool_calls, final_content = run_agentic_loop(
         url,
@@ -246,6 +255,7 @@ def run_test(url, test_case, stream):
         tools=test_case["tools"],
         mock_tool_responses=test_case["mock_tool_responses"],
         stream=stream,
+        force_tools=force_tools
     )
 
     if final_content is None and not all_tool_calls:
@@ -1093,6 +1103,9 @@ def main():
     parser.add_argument(
         "--stream-only", action="store_true", help="Only run streaming mode tests"
     )
+    parser.add_argument(
+        "--force-tools", action="store_true", help="Change tool mode to forced instead of auto"
+    )
     parser.add_argument(
         "--test",
         help="Run only the test whose name contains this substring (case-insensitive)",
@@ -1103,10 +1116,13 @@ def main():
     print_info(f"Testing server at {url}")
 
     modes = []
+    force_tools = False
     if not args.stream_only:
         modes.append(False)
     if not args.no_stream:
         modes.append(True)
+    if args.force_tools:
+        force_tools = True
 
     cases: list[dict] = ALL_TEST_CASES
     if args.test:
@@ -1121,7 +1137,7 @@ def main():
     for stream in modes:
         for case in cases:
             total += 1
-            if run_test(url, case, stream=stream):
+            if run_test(url, case, stream=stream, force_tools=force_tools):
                 passed += 1
 
     color = GREEN if passed == total else RED

scripts/sync-ggml.last

@@ -1 +1 @@
-387fa29fbbf3149f06a631c7850b6c35c24b0232
+ac6f7b44f60fde0091f0b3d99afde48f8c99b13a

scripts/sync_vendor.py

@@ -5,7 +5,7 @@ import os
 import sys
 import subprocess
 
-HTTPLIB_VERSION = "refs/tags/v0.43.2"
+HTTPLIB_VERSION = "refs/tags/v0.43.3"
 
 vendor = {
     "https://github.com/nlohmann/json/releases/latest/download/json.hpp":     "vendor/nlohmann/json.hpp",

src/llama-context.cpp

@@ -2230,13 +2230,17 @@ llm_graph_cb llama_context::graph_get_cb() const {
 
 class llama_io_write_dummy : public llama_io_write_i {
 public:
-    llama_io_write_dummy() = default;
+    llama_io_write_dummy(bool skip_tensors) : skip_tensors(skip_tensors) {}
 
     void write(const void * /* src */, size_t size) override {
         size_written += size;
     }
 
-    void write_tensor(const ggml_tensor * /* tensor */, size_t /* offset */, size_t size) override {
+    void write_tensor(ggml_tensor * /* tensor */, size_t /* offset */, size_t size) override {
+        if (skip_tensors) {
+            return;
+        }
+
         size_written += size;
     }
 
@@ -2245,14 +2249,23 @@ public:
     }
 
 private:
+    const bool skip_tensors;
+
     size_t size_written = 0;
 };
 
-class llama_io_write_buffer : public llama_io_write_i {
+class llama_io_write_host : public llama_io_write_i {
 public:
-    llama_io_write_buffer(
+    llama_io_write_host(
             uint8_t * p, size_t len) : ptr(p), buf_size(len) {}
 
+    ~llama_io_write_host() {
+        // TODO: add backend support to batch tensor_get? or some other way to speed this up
+        for (const auto & winfo : winfos) {
+            ggml_backend_tensor_get(winfo.tensor, winfo.ptr, winfo.offset, winfo.size);
+        }
+    }
+
     void write(const void * src, size_t size) override {
         if (size > buf_size) {
             throw std::runtime_error("unexpectedly reached end of buffer");
@@ -2263,11 +2276,14 @@ public:
         buf_size -= size;
     }
 
-    void write_tensor(const ggml_tensor * tensor, size_t offset, size_t size) override {
+    void write_tensor(ggml_tensor * tensor, size_t offset, size_t size) override {
         if (size > buf_size) {
             throw std::runtime_error("unexpectedly reached end of buffer");
         }
-        ggml_backend_tensor_get(tensor, ptr, offset, size);
+
+        // save the write for later during destruction
+        winfos.push_back({tensor, ptr, size, offset});
+
         ptr += size;
         size_written += size;
         buf_size -= size;
@@ -2281,25 +2297,48 @@ private:
     uint8_t * ptr;
     size_t buf_size = 0;
     size_t size_written = 0;
+
+    struct write_info {
+        ggml_tensor * tensor;
+        uint8_t * ptr;
+        size_t size;
+        size_t offset;
+    };
+    std::vector<write_info> winfos;
 };
 
-class llama_io_read_buffer : public llama_io_read_i {
+class llama_io_read_host : public llama_io_read_i {
 public:
-    llama_io_read_buffer(const uint8_t * p, size_t len) : ptr(p), buf_size(len) {}
+    llama_io_read_host(const uint8_t * p, size_t len) : ptr(p), buf_size(len) {}
 
-    const uint8_t * read(size_t size) override {
-        const uint8_t * base_ptr = ptr;
+    ~llama_io_read_host() {
+        // flush the reads
+        for (const auto & rinfo : rinfos) {
+            ggml_backend_tensor_set(rinfo.tensor, rinfo.ptr, rinfo.offset, rinfo.size);
+        }
+    }
+
+    void read(void * dst, size_t size) override {
         if (size > buf_size) {
             throw std::runtime_error("unexpectedly reached end of buffer");
         }
+        memcpy(dst, ptr, size);
         ptr += size;
         size_read += size;
         buf_size -= size;
-        return base_ptr;
     }
 
-    void read_to(void * dst, size_t size) override {
-        memcpy(dst, read(size), size);
+    void read_tensor(ggml_tensor * tensor, size_t offset, size_t size) override {
+        if (size > buf_size) {
+            throw std::runtime_error("unexpectedly reached end of buffer");
+        }
+
+        // save for later during destruction
+        rinfos.push_back({tensor, ptr, size, offset});
+
+        ptr += size;
+        size_read += size;
+        buf_size -= size;
     }
 
     size_t n_bytes() override {
@@ -2310,6 +2349,14 @@ private:
     const uint8_t * ptr;
     size_t buf_size = 0;
     size_t size_read = 0;
+
+    struct read_info {
+        ggml_tensor * tensor;
+        const uint8_t * ptr;
+        size_t size;
+        size_t offset;
+    };
+    std::vector<read_info> rinfos;
 };
 
 class llama_io_write_file : public llama_io_write_i {
@@ -2321,7 +2368,7 @@ public:
         size_written += size;
     }
 
-    void write_tensor(const ggml_tensor * tensor, size_t offset, size_t size) override {
+    void write_tensor(ggml_tensor * tensor, size_t offset, size_t size) override {
         temp_buffer.resize(size);
         ggml_backend_tensor_get(tensor, temp_buffer.data(), offset, size);
         write(temp_buffer.data(), temp_buffer.size());
@@ -2341,15 +2388,15 @@ class llama_io_read_file : public llama_io_read_i {
 public:
     llama_io_read_file(llama_file * f) : file(f) {}
 
-    void read_to(void * dst, size_t size) override {
+    void read(void * dst, size_t size) override {
         file->read_raw(dst, size);
         size_read += size;
     }
 
-    const uint8_t * read(size_t size) override {
+    void read_tensor(ggml_tensor * tensor, size_t offset, size_t size) override {
         temp_buffer.resize(size);
-        read_to(temp_buffer.data(), size);
-        return temp_buffer.data();
+        read(temp_buffer.data(), size);
+        ggml_backend_tensor_set(tensor, temp_buffer.data(), offset, size);
     }
 
     size_t n_bytes() override {
@@ -2362,8 +2409,162 @@ private:
     std::vector<uint8_t> temp_buffer;
 };
 
+class llama_io_write_device : public llama_io_write_i {
+public:
+    llama_io_write_device(uint8_t * p, size_t len, llama_memory_buffers & mbufs) : ptr(p), buf_size(len), mbufs(mbufs)  {
+    }
+
+    ~llama_io_write_device() {
+        llama_memory_buffers mbufs_new;
+
+        for (const auto & winfo : winfos) {
+            auto * buft = ggml_backend_buffer_get_type(winfo.tensor->buffer);
+
+            mbufs_new[buft].n_tensors++;
+            mbufs_new[buft].total_size += winfo.size;
+        }
+
+        for (auto & [buft, mbuf] : mbufs_new) {
+            ggml_init_params params = {
+                /*.mem_size   =*/ 2*mbuf.n_tensors*ggml_tensor_overhead(),
+                /*.mem_buffer =*/ NULL,
+                /*.no_alloc   =*/ true,
+            };
+
+            mbuf.ctx.reset(ggml_init(params));
+
+            mbuf.org.reserve(mbuf.n_tensors);
+            mbuf.cpy.reserve(mbuf.n_tensors);
+        }
+
+        for (const auto & winfo : winfos) {
+            auto * buft = ggml_backend_buffer_get_type(winfo.tensor->buffer);
+
+            const int64_t n = winfo.size/ggml_element_size(winfo.tensor);
+
+            auto & mbuf = mbufs_new[buft];
+
+            mbuf.org.push_back(ggml_view_1d      (mbuf.ctx.get(), winfo.tensor, n, winfo.offset));
+            mbuf.cpy.push_back(ggml_new_tensor_1d(mbuf.ctx.get(), winfo.tensor->type, n));
+        }
+
+        for (auto & [buft, mbuf] : mbufs_new) {
+            auto & mbuf_cur = mbufs[buft];
+
+            if (!mbuf_cur.buf || mbuf_cur.org.size() != mbuf.org.size() || mbuf_cur.total_size != mbuf.total_size) {
+                mbuf_cur = std::move(mbuf);
+
+                mbuf_cur.buf.reset(ggml_backend_alloc_ctx_tensors_from_buft(mbuf_cur.ctx.get(), buft));
+
+                LLAMA_LOG_INFO("%s: allocated '%s' buffer %.3f MiB\n", __func__, ggml_backend_buft_name(buft), mbuf.total_size/1024.0/1024.0);
+            }
+
+            for (size_t i = 0; i < mbuf_cur.org.size(); ++i) {
+                ggml_backend_tensor_copy(mbuf_cur.org[i], mbuf_cur.cpy[i]);
+            }
+        }
+    }
+
+    void write(const void * src, size_t size) override {
+        if (size > buf_size) {
+            throw std::runtime_error("unexpectedly reached end of buffer");
+        }
+        memcpy(ptr, src, size);
+        ptr += size;
+        size_written += size;
+        buf_size -= size;
+    }
+
+    void write_tensor(ggml_tensor * tensor, size_t offset, size_t size) override {
+        // save the write for later during destruction
+        winfos.push_back({tensor, ptr, size, offset});
+    }
+
+    size_t n_bytes() override {
+        return size_written;
+    }
+
+private:
+    uint8_t * ptr;
+    size_t buf_size = 0;
+    size_t size_written = 0;
+
+    struct write_info {
+        ggml_tensor * tensor;
+        uint8_t * ptr;
+        size_t size;
+        size_t offset;
+    };
+    std::vector<write_info> winfos;
+
+    llama_memory_buffers & mbufs;
+};
+
+class llama_io_read_device : public llama_io_read_i {
+public:
+    llama_io_read_device(const uint8_t * p, size_t len, const llama_memory_buffers & mbufs) : ptr(p), buf_size(len), mbufs(mbufs) {
+    }
+
+    ~llama_io_read_device() {
+        llama_memory_buffers mbufs_new;
+
+        for (const auto & rinfo : rinfos) {
+            auto * buft = ggml_backend_buffer_get_type(rinfo.tensor->buffer);
+
+            mbufs_new[buft].n_tensors++;
+            mbufs_new[buft].total_size += rinfo.size;
+        }
+
+        for (auto & [buft, mbuf] : mbufs_new) {
+            const auto & mbuf_cur = mbufs.at(buft);
+
+            if (!mbuf_cur.buf || mbuf_cur.n_tensors != mbuf.n_tensors || mbuf_cur.total_size != mbuf.total_size) {
+                GGML_ABORT("%s: memory buffer mismatch\n", __func__);
+            }
+
+            for (size_t i = 0; i < mbuf_cur.org.size(); ++i) {
+                ggml_backend_tensor_copy(mbuf_cur.cpy[i], mbuf_cur.org[i]);
+            }
+        }
+    }
+
+    void read(void * dst, size_t size) override {
+        if (size > buf_size) {
+            throw std::runtime_error("unexpectedly reached end of buffer");
+        }
+        memcpy(dst, ptr, size);
+        ptr += size;
+        size_read += size;
+        buf_size -= size;
+    }
+
+    void read_tensor(ggml_tensor * tensor, size_t offset, size_t size) override {
+        // save for later during destruction
+        rinfos.push_back({tensor, ptr, size, offset});
+    }
+
+    size_t n_bytes() override {
+        return size_read;
+    }
+
+private:
+    const uint8_t * ptr;
+    size_t buf_size = 0;
+    size_t size_read = 0;
+
+    struct read_info {
+        ggml_tensor * tensor;
+        const uint8_t * ptr;
+        size_t size;
+        size_t offset;
+    };
+    std::vector<read_info> rinfos;
+
+    const llama_memory_buffers & mbufs;
+};
+
 size_t llama_context::state_get_size() {
-    llama_io_write_dummy io;
+    llama_io_write_dummy io(false);
     try {
         return state_write_data(io);
     } catch (const std::exception & err) {
@@ -2373,7 +2574,7 @@ size_t llama_context::state_get_size() {
 }
 
 size_t llama_context::state_get_data(uint8_t * dst, size_t size) {
-    llama_io_write_buffer io(dst, size);
+    llama_io_write_host io(dst, size);
     try {
         return state_write_data(io);
     } catch (const std::exception & err) {
@@ -2383,7 +2584,7 @@ size_t llama_context::state_get_data(uint8_t * dst, size_t size) {
 }
 
 size_t llama_context::state_set_data(const uint8_t * src, size_t size) {
-    llama_io_read_buffer io(src, size);
+    llama_io_read_host io(src, size);
     try {
         return state_read_data(io);
     } catch (const std::exception & err) {
@@ -2392,9 +2593,14 @@ size_t llama_context::state_set_data(const uint8_t * src, size_t size) {
     }
 }
 
+static constexpr uint32_t io_magic = 0xaf143cd8;
+
 size_t llama_context::state_seq_get_size(llama_seq_id seq_id, llama_state_seq_flags flags) {
-    llama_io_write_dummy io;
+    llama_io_write_dummy io(flags & LLAMA_STATE_SEQ_FLAGS_ON_DEVICE);
     try {
+        io.write(&io_magic, sizeof(io_magic));
+        io.write(&seq_id, sizeof(seq_id));
+
         return state_seq_write_data(io, seq_id, flags);
     } catch (const std::exception & err) {
         LLAMA_LOG_ERROR("%s: error getting state size: %s\n", __func__, err.what());
@@ -2403,9 +2609,18 @@ size_t llama_context::state_seq_get_size(llama_seq_id seq_id, llama_state_seq_fl
 }
 
 size_t llama_context::state_seq_get_data(llama_seq_id seq_id, uint8_t * dst, size_t size, llama_state_seq_flags flags) {
-    llama_io_write_buffer io(dst, size);
+    std::unique_ptr<llama_io_write_i> io;
+    if (flags & LLAMA_STATE_SEQ_FLAGS_ON_DEVICE) {
+        io = std::make_unique<llama_io_write_device>(dst, size, mem_storage[seq_id]);
+    } else {
+        io = std::make_unique<llama_io_write_host>(dst, size);
+    }
+
     try {
-        return state_seq_write_data(io, seq_id, flags);
+        io->write(&io_magic, sizeof(io_magic));
+        io->write(&seq_id, sizeof(seq_id));
+
+        return state_seq_write_data(*io, seq_id, flags);
     } catch (const std::exception & err) {
         LLAMA_LOG_ERROR("%s: error saving state: %s\n", __func__, err.what());
         return 0;
@@ -2413,9 +2628,43 @@ size_t llama_context::state_seq_get_data(llama_seq_id seq_id, uint8_t * dst, siz
 }
 
 size_t llama_context::state_seq_set_data(llama_seq_id seq_id, const uint8_t * src, size_t size, llama_state_seq_flags flags) {
-    llama_io_read_buffer io(src, size);
+    std::unique_ptr<llama_io_read_i> io;
+    if (flags & LLAMA_STATE_SEQ_FLAGS_ON_DEVICE) {
+        // create a temporary io to read the magic and the src seq_id
+        io = std::make_unique<llama_io_read_host>(src, size);
+
+        uint32_t magic_read;
+        io->read(&magic_read, sizeof(magic_read));
+        if (io_magic != magic_read) {
+            throw std::runtime_error("wrong sequence state magic");
+        }
+
+        llama_seq_id seq_id_read;
+        io->read(&seq_id_read, sizeof(seq_id_read));
+
+        GGML_ASSERT(mem_storage.find(seq_id_read) != mem_storage.end());
+
+        io = std::make_unique<llama_io_read_device>(src, size, mem_storage[seq_id_read]);
+    } else {
+        io = std::make_unique<llama_io_read_host>(src, size);
+    }
+
     try {
-        return state_seq_read_data(io, seq_id, flags);
+        uint32_t magic_read;
+        io->read(&magic_read, sizeof(magic_read));
+        if (io_magic != magic_read) {
+            throw std::runtime_error("wrong sequence state magic");
+        }
+
+        const bool need_seq_match = (flags & LLAMA_STATE_SEQ_FLAGS_PARTIAL_ONLY);
+
+        llama_seq_id seq_id_read;
+        io->read(&seq_id_read, sizeof(seq_id_read));
+        if (need_seq_match && seq_id != seq_id_read) {
+            throw std::runtime_error("wrong sequence id");
+        }
+
+        return state_seq_read_data(*io, seq_id, flags);
     } catch (const std::exception & err) {
         LLAMA_LOG_ERROR("%s: error loading state: %s\n", __func__, err.what());
         return 0;
@@ -3406,7 +3655,6 @@ size_t llama_state_seq_get_data_ext(llama_context * ctx, uint8_t * dst, size_t s
 
     return ctx->state_seq_get_data(seq_id, dst, size, flags);
 }
-
 size_t llama_state_seq_set_data_ext(llama_context * ctx, const uint8_t * src, size_t size, llama_seq_id seq_id, llama_state_seq_flags flags) {
     ctx->synchronize();
 

src/llama-context.h

@@ -23,6 +23,21 @@ class llama_io_write_i;
 struct llama_memory_i;
 struct llama_memory_context_i;
 
+// stores copy of the memory in device buffer. used for fast state save/load
+struct llama_memory_buffer {
+    int n_tensors = 0;
+    size_t total_size = 0;
+
+    ggml_backend_buffer_ptr buf;
+
+    ggml_context_ptr ctx;
+
+    std::vector<ggml_tensor *> org;
+    std::vector<ggml_tensor *> cpy;
+};
+
+using llama_memory_buffers = std::map<ggml_backend_buffer_type_t, llama_memory_buffer>;
+
 struct llama_context {
     // init scheduler and compute buffers, reserve worst-case graphs
     llama_context(
@@ -128,6 +143,7 @@ struct llama_context {
     size_t state_set_data(const uint8_t * src, size_t size);
 
     size_t state_seq_get_size(llama_seq_id seq_id, llama_state_seq_flags flags);
+
     size_t state_seq_get_data(llama_seq_id seq_id,       uint8_t * dst, size_t size, llama_state_seq_flags flags);
     size_t state_seq_set_data(llama_seq_id seq_id, const uint8_t * src, size_t size, llama_state_seq_flags flags);
 
@@ -328,6 +344,9 @@ private:
     // host buffer for the model output (logits and embeddings)
     ggml_backend_buffer_ptr buf_output;
 
+    // keep copies of the per-sequence memory on the device
+    std::map<llama_seq_id, llama_memory_buffers> mem_storage;
+
     bool has_evaluated_once = false;
 
     // env: LLAMA_GRAPH_REUSE_DISABLE

src/llama-graph.cpp

@@ -65,8 +65,13 @@ static ggml_tensor * ggml_mul_mat_aux(
 
     ggml_tensor * res;
 
-    res = ggml_reshape_2d(ctx, cur, n, ggml_nelements(cur)/n);
+    if (!ggml_is_contiguous(cur)) {
+        res = ggml_cont_2d   (ctx, cur, n, ggml_nelements(cur)/n);
+    } else {
+        res = ggml_reshape_2d(ctx, cur, n, ggml_nelements(cur)/n);
+    }
     res = ggml_mul_mat   (ctx, rot, res);
+    ggml_mul_mat_set_hint(res, GGML_HINT_SRC0_IS_HADAMARD);
     res = ggml_reshape_4d(ctx, res, cur->ne[0], cur->ne[1], cur->ne[2], cur->ne[3]);
 
     return res;

src/llama-io.cpp

@@ -1,5 +1,7 @@
 #include "llama-io.h"
 
+#include <vector>
+
 void llama_io_write_i::write_string(const std::string & str) {
     uint32_t str_size = str.size();
 
@@ -9,7 +11,10 @@ void llama_io_write_i::write_string(const std::string & str) {
 
 void llama_io_read_i::read_string(std::string & str) {
     uint32_t str_size;
-    read_to(&str_size, sizeof(str_size));
+    read(&str_size, sizeof(str_size));
 
-    str.assign((const char *) read(str_size), str_size);
+    std::vector<char> buf(str_size);
+    read(buf.data(), str_size);
+
+    str.assign(buf.data(), str_size);
 }

src/llama-io.h

@@ -12,7 +12,7 @@ public:
     virtual ~llama_io_write_i() = default;
 
     virtual void write(const void * src, size_t size) = 0;
-    virtual void write_tensor(const ggml_tensor * tensor, size_t offset, size_t size) = 0;
+    virtual void write_tensor(ggml_tensor * tensor, size_t offset, size_t size) = 0;
 
     // bytes written so far
     virtual size_t n_bytes() = 0;
@@ -25,8 +25,8 @@ public:
     llama_io_read_i() = default;
     virtual ~llama_io_read_i() = default;
 
-    virtual const uint8_t * read(size_t size) = 0;
-    virtual void read_to(void * dst, size_t size) = 0;
+    virtual void read(void * dst, size_t size) = 0;
+    virtual void read_tensor(ggml_tensor * tensor, size_t offset, size_t size) = 0;
 
     // bytes read so far
     virtual size_t n_bytes() = 0;

src/llama-kv-cache.cpp

@@ -67,6 +67,7 @@ static ggml_tensor * ggml_mul_mat_aux(
 
     res = ggml_reshape_2d(ctx, cur, n, ggml_nelements(cur)/n);
     res = ggml_mul_mat   (ctx, rot, res);
+    ggml_mul_mat_set_hint(res, GGML_HINT_SRC0_IS_HADAMARD);
     res = ggml_reshape_4d(ctx, res, cur->ne[0], cur->ne[1], cur->ne[2], cur->ne[3]);
 
     return res;
@@ -1900,14 +1901,14 @@ void llama_kv_cache::state_read(llama_io_read_i & io, llama_seq_id seq_id, llama
     GGML_ASSERT(seq_id == -1 || (seq_id >= 0 && (size_t) seq_id < seq_to_stream.size()));
 
     uint32_t n_stream_cur;
-    io.read_to(&n_stream_cur, sizeof(n_stream_cur));
+    io.read(&n_stream_cur, sizeof(n_stream_cur));
     if (n_stream_cur != n_stream) {
         throw std::runtime_error("n_stream mismatch");
     }
 
     for (uint32_t s = 0; s < n_stream; ++s) {
         uint32_t cell_count;
-        io.read_to(&cell_count, sizeof(cell_count));
+        io.read(&cell_count, sizeof(cell_count));
 
         if (cell_count == 0) {
             continue;
@@ -2082,8 +2083,8 @@ bool llama_kv_cache::state_read_meta(llama_io_read_i & io, uint32_t strm, uint32
             llama_pos pos;
             uint32_t n_seq_id;
 
-            io.read_to(&pos,      sizeof(pos));
-            io.read_to(&n_seq_id, sizeof(n_seq_id));
+            io.read(&pos,      sizeof(pos));
+            io.read(&n_seq_id, sizeof(n_seq_id));
 
             if (n_seq_id != 1) {
                 LLAMA_LOG_ERROR("%s: invalid seq_id-agnostic kv cell\n", __func__);
@@ -2092,7 +2093,7 @@ bool llama_kv_cache::state_read_meta(llama_io_read_i & io, uint32_t strm, uint32
 
             if (hparams.n_pos_per_embd() > 1) {
                 llama_kv_cell_ext ext;
-                io.read_to(&ext, sizeof(ext));
+                io.read(&ext, sizeof(ext));
 
                 ubatch.pos[i + ubatch.n_tokens]   = ext.y;
                 ubatch.pos[i + ubatch.n_tokens*2] = ext.x;
@@ -2101,7 +2102,7 @@ bool llama_kv_cache::state_read_meta(llama_io_read_i & io, uint32_t strm, uint32
             // read the sequence id, but directly discard it - we will use dest_seq_id instead
             {
                 llama_seq_id seq_id;
-                io.read_to(&seq_id, sizeof(seq_id));
+                io.read(&seq_id, sizeof(seq_id));
             }
 
             ubatch.pos[i]      = pos;
@@ -2143,20 +2144,20 @@ bool llama_kv_cache::state_read_meta(llama_io_read_i & io, uint32_t strm, uint32
             llama_pos pos;
             uint32_t  n_seq_id;
 
-            io.read_to(&pos,      sizeof(pos));
-            io.read_to(&n_seq_id, sizeof(n_seq_id));
+            io.read(&pos,      sizeof(pos));
+            io.read(&n_seq_id, sizeof(n_seq_id));
 
             cells.pos_set(i, pos);
 
             if (hparams.n_pos_per_embd() > 1) {
                 llama_kv_cell_ext ext;
-                io.read_to(&ext, sizeof(ext));
+                io.read(&ext, sizeof(ext));
                 cells.ext_set(i, ext);
             }
 
             for (uint32_t j = 0; j < n_seq_id; ++j) {
                 llama_seq_id seq_id;
-                io.read_to(&seq_id, sizeof(seq_id));
+                io.read(&seq_id, sizeof(seq_id));
 
                 if (seq_id < 0 || (uint32_t) seq_id >= n_seq_max) {
                     LLAMA_LOG_ERROR("%s: invalid seq_id, %d is out of range [0, %u)\n", __func__, seq_id, n_seq_max);
@@ -2189,8 +2190,8 @@ bool llama_kv_cache::state_read_data(llama_io_read_i & io, uint32_t strm, uint32
     uint32_t v_trans;
     uint32_t n_layer;
 
-    io.read_to(&v_trans, sizeof(v_trans));
-    io.read_to(&n_layer, sizeof(n_layer));
+    io.read(&v_trans, sizeof(v_trans));
+    io.read(&n_layer, sizeof(n_layer));
 
     if (n_layer != layers.size()) {
         LLAMA_LOG_ERROR("%s: mismatched layer count (%u instead of %u)\n", __func__, n_layer, (uint32_t) layers.size());
@@ -2217,7 +2218,7 @@ bool llama_kv_cache::state_read_data(llama_io_read_i & io, uint32_t strm, uint32
 
         // Read type of key
         int32_t k_type_i_ref;
-        io.read_to(&k_type_i_ref, sizeof(k_type_i_ref));
+        io.read(&k_type_i_ref, sizeof(k_type_i_ref));
         const int32_t k_type_i = (int32_t) k->type;
         if (k_type_i != k_type_i_ref) {
             LLAMA_LOG_ERROR("%s: mismatched key type (%d != %d, layer %d)\n", __func__, k_type_i, k_type_i_ref, il);
@@ -2226,7 +2227,7 @@ bool llama_kv_cache::state_read_data(llama_io_read_i & io, uint32_t strm, uint32
 
         // Read row size of key
         uint64_t k_size_row_ref;
-        io.read_to(&k_size_row_ref, sizeof(k_size_row_ref));
+        io.read(&k_size_row_ref, sizeof(k_size_row_ref));
         const size_t k_size_row = ggml_row_size(k->type, n_embd_k_gqa);
         if (k_size_row != k_size_row_ref) {
             LLAMA_LOG_ERROR("%s: mismatched key row size (%zu != %zu, layer %d)\n", __func__, k_size_row, (size_t) k_size_row_ref, il);
@@ -2236,13 +2237,12 @@ bool llama_kv_cache::state_read_data(llama_io_read_i & io, uint32_t strm, uint32
         if (cell_count) {
             if (sinfo.is_contiguous()) {
                 // Fast path: contiguous cells, single memcpy
-                ggml_backend_tensor_set(k, io.read(cell_count * k_size_row), sinfo.head() * k_size_row, cell_count * k_size_row);
+                io.read_tensor(k, sinfo.head() * k_size_row, cell_count * k_size_row);
             } else {
                 // Slow path: scatter to non-contiguous positions
-                const void * src = io.read(cell_count * k_size_row);
                 for (uint32_t i = 0; i < cell_count; ++i) {
                     const size_t dst_offset = sinfo.idxs[0][i] * k_size_row;
-                    ggml_backend_tensor_set(k, (const char*)src + i * k_size_row, dst_offset, k_size_row);
+                    io.read_tensor(k, dst_offset, k_size_row);
                 }
             }
         }
@@ -2261,7 +2261,7 @@ bool llama_kv_cache::state_read_data(llama_io_read_i & io, uint32_t strm, uint32
 
             // Read type of value
             int32_t v_type_i_ref;
-            io.read_to(&v_type_i_ref, sizeof(v_type_i_ref));
+            io.read(&v_type_i_ref, sizeof(v_type_i_ref));
             const int32_t v_type_i = (int32_t) v->type;
             if (v_type_i != v_type_i_ref) {
                 LLAMA_LOG_ERROR("%s: mismatched value type (%d != %d, layer %d)\n", __func__, v_type_i, v_type_i_ref, il);
@@ -2270,7 +2270,7 @@ bool llama_kv_cache::state_read_data(llama_io_read_i & io, uint32_t strm, uint32
 
             // Read row size of value
             uint64_t v_size_row_ref;
-            io.read_to(&v_size_row_ref, sizeof(v_size_row_ref));
+            io.read(&v_size_row_ref, sizeof(v_size_row_ref));
             const size_t v_size_row = ggml_row_size(v->type, n_embd_v_gqa);
             if (v_size_row != v_size_row_ref) {
                 LLAMA_LOG_ERROR("%s: mismatched value row size (%zu != %zu, layer %d)\n", __func__, v_size_row, (size_t) v_size_row_ref, il);
@@ -2280,13 +2280,12 @@ bool llama_kv_cache::state_read_data(llama_io_read_i & io, uint32_t strm, uint32
             if (cell_count) {
                 if (sinfo.is_contiguous()) {
                     // Fast path: contiguous cells, single memcpy
-                    ggml_backend_tensor_set(v, io.read(cell_count * v_size_row), sinfo.head() * v_size_row, cell_count * v_size_row);
+                    io.read_tensor(v, sinfo.head() * v_size_row, cell_count * v_size_row);
                 } else {
                     // Slow path: scatter to non-contiguous positions
-                    const void * src = io.read(cell_count * v_size_row);
                     for (uint32_t i = 0; i < cell_count; ++i) {
                         const size_t dst_offset = sinfo.idxs[0][i] * v_size_row;
-                        ggml_backend_tensor_set(v, (const char*)src + i * v_size_row, dst_offset, v_size_row);
+                        io.read_tensor(v, dst_offset, v_size_row);
                     }
                 }
             }
@@ -2305,7 +2304,7 @@ bool llama_kv_cache::state_read_data(llama_io_read_i & io, uint32_t strm, uint32
 
             // Read type of value
             int32_t v_type_i_ref;
-            io.read_to(&v_type_i_ref, sizeof(v_type_i_ref));
+            io.read(&v_type_i_ref, sizeof(v_type_i_ref));
             const int32_t v_type_i = (int32_t) v->type;
             if (v_type_i != v_type_i_ref) {
                 LLAMA_LOG_ERROR("%s: mismatched value type (%d != %d, layer %d)\n", __func__, v_type_i, v_type_i_ref, il);
@@ -2314,7 +2313,7 @@ bool llama_kv_cache::state_read_data(llama_io_read_i & io, uint32_t strm, uint32
 
             // Read element size of value
             uint32_t v_size_el_ref;
-            io.read_to(&v_size_el_ref, sizeof(v_size_el_ref));
+            io.read(&v_size_el_ref, sizeof(v_size_el_ref));
             const size_t v_size_el = ggml_type_size(v->type);
             if (v_size_el != v_size_el_ref) {
                 LLAMA_LOG_ERROR("%s: mismatched value element size (%zu != %zu, layer %d)\n", __func__, v_size_el, (size_t) v_size_el_ref, il);
@@ -2323,7 +2322,7 @@ bool llama_kv_cache::state_read_data(llama_io_read_i & io, uint32_t strm, uint32
 
             // Read GQA embedding size
             uint32_t n_embd_v_gqa_ref;
-            io.read_to(&n_embd_v_gqa_ref, sizeof(n_embd_v_gqa_ref));
+            io.read(&n_embd_v_gqa_ref, sizeof(n_embd_v_gqa_ref));
             if (n_embd_v_gqa != n_embd_v_gqa_ref) {
                 LLAMA_LOG_ERROR("%s: mismatched GQA embedding size (%u != %u, layer %d)\n", __func__, n_embd_v_gqa, n_embd_v_gqa_ref, il);
                 return false;
@@ -2335,15 +2334,14 @@ bool llama_kv_cache::state_read_data(llama_io_read_i & io, uint32_t strm, uint32
                     const uint32_t h = sinfo.head();
                     for (uint32_t j = 0; j < n_embd_v_gqa; ++j) {
                         const size_t dst_offset = (h + j * cells.size()) * v_size_el;
-                        ggml_backend_tensor_set(v, io.read(cell_count * v_size_el), dst_offset, cell_count * v_size_el);
+                        io.read_tensor(v, dst_offset, cell_count * v_size_el);
                     }
                 } else {
                     // Slow path: scatter to non-contiguous positions
                     for (uint32_t j = 0; j < n_embd_v_gqa; ++j) {
-                        const void * src = io.read(cell_count * v_size_el);
                         for (uint32_t i = 0; i < cell_count; ++i) {
                             const size_t dst_offset = (sinfo.idxs[0][i] + j * cells.size()) * v_size_el;
-                            ggml_backend_tensor_set(v, (const char*)src + i * v_size_el, dst_offset, v_size_el);
+                            io.read_tensor(v, dst_offset, v_size_el);
                         }
                     }
                 }

src/llama-memory-recurrent.cpp

@@ -743,7 +743,7 @@ void llama_memory_recurrent::state_read(llama_io_read_i & io, llama_seq_id seq_i
     GGML_UNUSED(flags);
 
     uint32_t cell_count;
-    io.read_to(&cell_count, sizeof(cell_count));
+    io.read(&cell_count, sizeof(cell_count));
 
     bool res = true;
 
@@ -784,7 +784,7 @@ void llama_memory_recurrent::state_write_data(llama_io_write_i & io, const std::
     const uint32_t n_layer = hparams.n_layer;
 
     io.write(&s_trans, sizeof(s_trans));
-    io.write(&n_layer,   sizeof(n_layer));
+    io.write(&n_layer, sizeof(n_layer));
 
     // Iterate and write all the R tensors first, each row is a cell
     // Get whole range at a time
@@ -879,8 +879,8 @@ bool llama_memory_recurrent::state_read_meta(llama_io_read_i & io, uint32_t cell
             llama_pos pos;
             uint32_t n_seq_id;
 
-            io.read_to(&pos,      sizeof(pos));
-            io.read_to(&n_seq_id, sizeof(n_seq_id));
+            io.read(&pos,      sizeof(pos));
+            io.read(&n_seq_id, sizeof(n_seq_id));
 
             if (n_seq_id != 0) {
                 LLAMA_LOG_ERROR("%s: invalid seq_id-agnostic kv cell\n", __func__);
@@ -920,14 +920,14 @@ bool llama_memory_recurrent::state_read_meta(llama_io_read_i & io, uint32_t cell
             llama_pos pos;
             uint32_t  n_seq_id;
 
-            io.read_to(&pos,      sizeof(pos));
-            io.read_to(&n_seq_id, sizeof(n_seq_id));
+            io.read(&pos,      sizeof(pos));
+            io.read(&n_seq_id, sizeof(n_seq_id));
 
             cell.pos = pos;
 
             for (uint32_t j = 0; j < n_seq_id; ++j) {
                 llama_seq_id seq_id;
-                io.read_to(&seq_id, sizeof(seq_id));
+                io.read(&seq_id, sizeof(seq_id));
 
                 if (seq_id < 0 || (uint32_t) seq_id >= this->n_seq_max) {
                     LLAMA_LOG_ERROR("%s: invalid seq_id, %d is out of range [0, %u)\n", __func__, seq_id, this->n_seq_max);
@@ -961,8 +961,8 @@ bool llama_memory_recurrent::state_read_meta(llama_io_read_i & io, uint32_t cell
 bool llama_memory_recurrent::state_read_data(llama_io_read_i & io, uint32_t cell_count) {
     uint32_t s_trans;
     uint32_t n_layer;
-    io.read_to(&s_trans, sizeof(s_trans));
-    io.read_to(&n_layer, sizeof(n_layer));
+    io.read(&s_trans, sizeof(s_trans));
+    io.read(&n_layer, sizeof(n_layer));
 
     if (n_layer != hparams.n_layer) {
         LLAMA_LOG_ERROR("%s: mismatched layer count (%u instead of %u)\n", __func__, n_layer, hparams.n_layer);
@@ -984,7 +984,7 @@ bool llama_memory_recurrent::state_read_data(llama_io_read_i & io, uint32_t cell
 
         // Read type of key
         int32_t r_type_i_ref;
-        io.read_to(&r_type_i_ref, sizeof(r_type_i_ref));
+        io.read(&r_type_i_ref, sizeof(r_type_i_ref));
         const int32_t r_type_i = (int32_t) r_l[il]->type;
         if (r_type_i != r_type_i_ref) {
             LLAMA_LOG_ERROR("%s: mismatched r type (%d != %d, layer %d)\n", __func__, r_type_i, r_type_i_ref, il);
@@ -993,7 +993,7 @@ bool llama_memory_recurrent::state_read_data(llama_io_read_i & io, uint32_t cell
 
         // Read row size of key
         uint64_t r_size_row_ref;
-        io.read_to(&r_size_row_ref, sizeof(r_size_row_ref));
+        io.read(&r_size_row_ref, sizeof(r_size_row_ref));
         const size_t r_size_row = ggml_row_size(r_l[il]->type, hparams.n_embd_r());
         if (r_size_row != r_size_row_ref) {
             LLAMA_LOG_ERROR("%s: mismatched r row size (%zu != %zu, layer %d)\n", __func__, r_size_row, (size_t) r_size_row_ref, il);
@@ -1002,7 +1002,7 @@ bool llama_memory_recurrent::state_read_data(llama_io_read_i & io, uint32_t cell
 
         if (cell_count) {
             // Read and set the keys for the whole cell range
-            ggml_backend_tensor_set(r_l[il], io.read(cell_count * r_size_row), head * r_size_row, cell_count * r_size_row);
+            io.read_tensor(r_l[il], head * r_size_row, cell_count * r_size_row);
         }
     }
 
@@ -1013,7 +1013,7 @@ bool llama_memory_recurrent::state_read_data(llama_io_read_i & io, uint32_t cell
 
             // Read type of value
             int32_t s_type_i_ref;
-            io.read_to(&s_type_i_ref, sizeof(s_type_i_ref));
+            io.read(&s_type_i_ref, sizeof(s_type_i_ref));
             const int32_t s_type_i = (int32_t)s_l[il]->type;
 
             if (s_type_i != s_type_i_ref) {
@@ -1023,7 +1023,7 @@ bool llama_memory_recurrent::state_read_data(llama_io_read_i & io, uint32_t cell
 
             // Read row size of value
             uint64_t s_size_row_ref;
-            io.read_to(&s_size_row_ref, sizeof(s_size_row_ref));
+            io.read(&s_size_row_ref, sizeof(s_size_row_ref));
             const size_t s_size_row = ggml_row_size(s_l[il]->type, hparams.n_embd_s());
             if (s_size_row != s_size_row_ref) {
                 LLAMA_LOG_ERROR("%s: mismatched s row size (%zu != %zu, layer %d)\n", __func__, s_size_row, (size_t) s_size_row_ref, il);
@@ -1032,7 +1032,7 @@ bool llama_memory_recurrent::state_read_data(llama_io_read_i & io, uint32_t cell
 
             if (cell_count) {
                 // Read and set the values for the whole cell range
-                ggml_backend_tensor_set(s_l[il], io.read(cell_count * s_size_row), head * s_size_row, cell_count * s_size_row);
+                io.read_tensor(s_l[il], head * s_size_row, cell_count * s_size_row);
             }
         }
     } else {
@@ -1045,7 +1045,7 @@ bool llama_memory_recurrent::state_read_data(llama_io_read_i & io, uint32_t cell
 
             // Read type of value
             int32_t s_type_i_ref;
-            io.read_to(&s_type_i_ref, sizeof(s_type_i_ref));
+            io.read(&s_type_i_ref, sizeof(s_type_i_ref));
             const int32_t s_type_i = (int32_t)s_l[il]->type;
             if (s_type_i != s_type_i_ref) {
                 LLAMA_LOG_ERROR("%s: mismatched s type (%d != %d, layer %d)\n", __func__, s_type_i, s_type_i_ref, il);
@@ -1054,7 +1054,7 @@ bool llama_memory_recurrent::state_read_data(llama_io_read_i & io, uint32_t cell
 
             // Read element size of value
             uint32_t s_size_el_ref;
-            io.read_to(&s_size_el_ref, sizeof(s_size_el_ref));
+            io.read(&s_size_el_ref, sizeof(s_size_el_ref));
             const size_t s_size_el = ggml_type_size(s_l[il]->type);
             if (s_size_el != s_size_el_ref) {
                 LLAMA_LOG_ERROR("%s: mismatched s element size (%zu != %zu, layer %d)\n", __func__, s_size_el, (size_t) s_size_el_ref, il);
@@ -1063,7 +1063,7 @@ bool llama_memory_recurrent::state_read_data(llama_io_read_i & io, uint32_t cell
 
             // Read state embedding size
             uint32_t n_embd_s_ref;
-            io.read_to(&n_embd_s_ref, sizeof(n_embd_s_ref));
+            io.read(&n_embd_s_ref, sizeof(n_embd_s_ref));
             if (n_embd_s != n_embd_s_ref) {
                 LLAMA_LOG_ERROR("%s: mismatched s embedding size (%u != %u, layer %d)\n", __func__, n_embd_s, n_embd_s_ref, il);
                 return false;
@@ -1073,7 +1073,7 @@ bool llama_memory_recurrent::state_read_data(llama_io_read_i & io, uint32_t cell
                 // For each row in the transposed matrix, read the values for the whole cell range
                 for (uint32_t j = 0; j < n_embd_s; ++j) {
                     const size_t dst_offset = (head + j * size) * s_size_el;
-                    ggml_backend_tensor_set(s_l[il], io.read(cell_count * s_size_el), dst_offset, cell_count * s_size_el);
+                    io.read_tensor(s_l[il], dst_offset, cell_count * s_size_el);
                 }
             }
         }

src/llama-model.h

@@ -577,14 +577,8 @@ struct llama_model {
     int64_t t_load_us  = 0;
     int64_t t_start_us = 0;
 
-    explicit llama_model(const struct llama_model_params & params);
-    ~llama_model();
-
-    void load_stats  (llama_model_loader & ml);
-    void load_arch   (llama_model_loader & ml);
-    void load_hparams(llama_model_loader & ml);
-    void load_vocab  (llama_model_loader & ml);
-    bool load_tensors(llama_model_loader & ml); // returns false if cancelled by progress_callback
+    explicit llama_model(const llama_model_params & params);
+    virtual ~llama_model();
 
     std::string arch_name() const;
     std::string type_name() const;
@@ -620,21 +614,94 @@ struct llama_model {
 
     ggml_tensor * get_rope_factors(const llama_cparams & cparams, int il) const;
 
-    // TODO: move this to new llm_arch_model_i interface
     llama_memory_i * create_memory(const llama_memory_params & params, const llama_cparams & cparams) const;
 
-    // TODO: move this to new llm_arch_model_i interface
     ggml_cgraph * build_graph(const llm_graph_params & params) const;
 
-private:
+    virtual void load_stats  (llama_model_loader & ml) = 0;
+    virtual void load_hparams(llama_model_loader & ml) = 0;
+    virtual void load_vocab  (llama_model_loader & ml) = 0;
+    virtual bool load_tensors(llama_model_loader & ml) = 0; // returns false if cancelled by progress_callback
+
+    // model must define these
+    virtual void load_arch_hparams(llama_model_loader & ml) = 0;
+    virtual void load_arch_tensors(llama_model_loader & ml) = 0;
+    virtual std::unique_ptr<llm_graph_context> build_arch_graph(const llm_graph_params & params) const = 0;
+
+protected:
     llama_model_params params;
 
     struct impl;
     std::unique_ptr<impl> pimpl;
 };
 
+llama_model * llama_model_create(llm_arch arch, const llama_model_params & params);
+llama_model * llama_model_create(llama_model_loader & ml, const llama_model_params & params);
+
+// model must inherit from this
+struct llama_model_base : public llama_model {
+    friend struct llama_model;
+
+    llama_model * model;
+    llama_model_loader * ml = nullptr;
+    const LLM_TN tn;
+
+    // llama_model_loader is not yet defined at this point, so we will set it after construction
+    const int TENSOR_DUPLICATED;
+    const int TENSOR_NOT_REQUIRED;
+    const int TENSOR_SKIP;
+    const int TENSOR_SKIP_IF_VIRTUAL;
+
+    explicit llama_model_base(const llama_model_params & params);
+    virtual ~llama_model_base() = default;
+
+    ggml_tensor * create_tensor(llama_model_loader & ml, const LLM_TN_IMPL & tn, const std::initializer_list<int64_t> & ne, int flags);
+
+    // convenience overload of create_tensor that doesn't require llama_model_loader
+    ggml_tensor * create_tensor(const LLM_TN_IMPL & tn, const std::initializer_list<int64_t> & ne, int flags);
+
+    // helper: try merged gate_up_exps first, fall back to separate gate and up
+    void create_tensor_gate_up_exps(llama_layer & layer, int bid, int64_t n_embd_,
+                int64_t n_ff_, int64_t n_expert_, int flags);
+
+    // helper: try to load merged qkv first, fall back to separate q, k, v
+    void create_tensor_qkv(llama_layer & layer, int bid,
+                int64_t n_embd_, int64_t n_embd_q_, int64_t n_embd_k_, int64_t n_embd_v_,
+                int flags);
+
+    void load_stats  (llama_model_loader & ml) override;
+    void load_hparams(llama_model_loader & ml) override;
+    void load_vocab  (llama_model_loader & ml) override;
+    bool load_tensors(llama_model_loader & ml) override;
+
+    // model must define these
+    void load_arch_hparams(llama_model_loader & ml) override = 0;
+    void load_arch_tensors(llama_model_loader & ml) override = 0;
+    std::unique_ptr<llm_graph_context> build_arch_graph(const llm_graph_params & params) const override = 0;
+};
+
 const char * llm_type_name(llm_type type);
 
+// convenience macro for loading local variables for load_tensors() in llama_model_base
+// note: cast to int64_t since we will use these for the tensor dimensions
+#define LLAMA_LOAD_LOCALS \
+    const int     n_layer        = hparams.n_layer;          GGML_UNUSED(n_layer); \
+    const int64_t n_head         = hparams.n_head();         GGML_UNUSED(n_head); \
+    const int64_t n_head_kv      = hparams.n_head_kv();      GGML_UNUSED(n_head_kv); \
+    const int64_t n_embd         = hparams.n_embd;           GGML_UNUSED(n_embd); \
+    const int64_t n_embd_k_gqa   = hparams.n_embd_k_gqa();   GGML_UNUSED(n_embd_k_gqa); \
+    const int64_t n_embd_v_gqa   = hparams.n_embd_v_gqa();   GGML_UNUSED(n_embd_v_gqa); \
+    const int64_t n_embd_head_k  = hparams.n_embd_head_k();  GGML_UNUSED(n_embd_head_k); \
+    const int64_t n_embd_head_v  = hparams.n_embd_head_v();  GGML_UNUSED(n_embd_head_v); \
+    const int64_t n_ff           = hparams.n_ff();           GGML_UNUSED(n_ff); \
+    const int64_t n_embd_gqa     = n_embd_v_gqa;             GGML_UNUSED(n_embd_gqa); \
+    const int64_t n_vocab        = vocab.n_tokens();         GGML_UNUSED(n_vocab); \
+    const int64_t n_token_types  = vocab.n_token_types();    GGML_UNUSED(n_token_types); \
+    const int64_t n_rot          = hparams.n_rot();          GGML_UNUSED(n_rot); \
+    const int64_t n_expert       = hparams.n_expert;         GGML_UNUSED(n_expert); \
+    const int64_t n_expert_used  = hparams.n_expert_used;    GGML_UNUSED(n_expert_used); \
+    const int64_t n_ctx_train    = hparams.n_ctx_train;      GGML_UNUSED(n_ctx_train);
+
 // For internal test use
 // TODO: remove
 const std::vector<std::pair<std::string, ggml_tensor *>> & llama_internal_get_tensor_map(const llama_model * model);

src/llama-quant.cpp

@@ -683,9 +683,9 @@ static ggml_type llama_tensor_get_type(quantize_state_impl & qs, const llama_mod
                         LLAMA_LOG_WARN("%s: %-36s - applying manual override: %s -> %s\n",
                                        __func__, tensor_name.c_str(), ggml_type_name(new_type), ggml_type_name(qtype));
                         new_type = qtype;
-                        manual = true;
-                        break;
                     }
+                    manual = true;
+                    break;
                 }
             }
         }
@@ -882,13 +882,18 @@ static void llama_model_quantize_impl(const std::string & fname_inp, const std::
         fname_inp, splits, /*file*/ nullptr, use_mmap, /*use_direct_io*/ false, /*check_tensors*/ true, /*no_alloc*/ false, kv_overrides, nullptr);
     ml.init_mappings(false); // no prefetching
 
-    llama_model model(llama_model_default_params());
+    auto mparams = llama_model_default_params();
+    std::unique_ptr<llama_model> model_ptr(llama_model_create(ml, mparams));
 
-    model.load_arch   (ml);
-    model.load_hparams(ml);
-    model.load_stats  (ml);
+    auto * model = dynamic_cast<llama_model_base *>(model_ptr.get());
+    if (model == nullptr) {
+        GGML_ABORT("fatal error: model does not implement llama_model_base");
+    }
 
-    quantize_state_impl qs(model, params);
+    model->load_hparams(ml);
+    model->load_stats  (ml);
+
+    quantize_state_impl qs(*model, params);
 
     if (params->only_copy) {
         ftype = ml.ftype;
@@ -1023,7 +1028,7 @@ static void llama_model_quantize_impl(const std::string & fname_inp, const std::
         }
         gguf_add_tensor(ctx_outs[i_split].get(), tensor);
 
-        metadata[i].allows_quantization = tensor_allows_quantization(params, model.arch, tensor);
+        metadata[i].allows_quantization = tensor_allows_quantization(params, model->arch, tensor);
 
         if (metadata[i].allows_quantization) {
             metadata[i].target_type = llama_tensor_get_type(qs, params, tensor, default_type, metadata[i]);
@@ -1331,9 +1336,9 @@ void llama_quant_free(quantize_state_impl * qs) {
 
 llama_model * llama_quant_model_from_metadata(const llama_quant_model_desc * desc) {
     struct llama_model_params mparams = llama_model_default_params();
-    auto * model = new llama_model(mparams);
-
-    model->arch = llm_arch_from_string(desc->architecture);
+    auto arch = llm_arch_from_string(desc->architecture);
+    auto * model = llama_model_create(arch, mparams);
+    model->arch = arch;
 
     // infer llm_type: only LLM_TYPE_70B matters for quantization logic
     if (model->arch == LLM_ARCH_LLAMA && desc->n_layer == 80 && desc->n_head != desc->n_head_kv) {

src/llama.cpp

@@ -89,6 +89,10 @@ void llama_backend_init(void) {
         struct ggml_context * ctx = ggml_init(params);
         ggml_free(ctx);
     }
+
+    if (!ggml_backend_reg_count()) {
+        ggml_backend_load_all();
+    }
 }
 
 void llama_numa_init(enum ggml_numa_strategy numa) {
@@ -111,113 +115,8 @@ int64_t llama_time_us(void) {
     return ggml_time_us();
 }
 
-// Returns 0 on success, -1 on error, and -2 on cancellation via llama_progress_callback
-static int llama_model_load(struct gguf_context * metadata, llama_model_set_tensor_data_t set_tensor_data, void * set_tensor_data_ud,
-        const std::string & fname, std::vector<std::string> & splits, FILE * file, llama_model & model, llama_model_params & params) {
-    // loading time will be recalculated after the first eval, so
-    // we take page faults deferred by mmap() into consideration
-    model.t_load_us = 0;
-    time_meas tm(model.t_load_us);
-
-    model.t_start_us = tm.t_start_us;
-
-    try {
-        llama_model_loader ml(metadata, set_tensor_data, set_tensor_data_ud, fname, splits, file, params.use_mmap, params.use_direct_io,
-            params.check_tensors, params.no_alloc, params.kv_overrides, params.tensor_buft_overrides);
-
-        ml.print_info();
-
-        model.hparams.vocab_only = params.vocab_only;
-        model.hparams.no_alloc   = params.no_alloc;
-
-        try {
-            model.load_arch(ml);
-        } catch(const std::exception & e) {
-            throw std::runtime_error("error loading model architecture: " + std::string(e.what()));
-        }
-        try {
-            model.load_hparams(ml);
-        } catch(const std::exception & e) {
-            throw std::runtime_error("error loading model hyperparameters: " + std::string(e.what()));
-        }
-        if (model.arch == LLM_ARCH_CLIP) {
-            throw std::runtime_error("CLIP cannot be used as main model, use it with --mmproj instead");
-        }
-        try {
-            model.load_vocab(ml);
-        } catch(const std::exception & e) {
-            throw std::runtime_error("error loading model vocabulary: " + std::string(e.what()));
-        }
-
-        model.load_stats(ml);
-        model.print_info();
-
-        if (params.vocab_only) {
-            LLAMA_LOG_INFO("%s: vocab only - skipping tensors\n", __func__);
-            return 0;
-        }
-
-        if (!model.load_tensors(ml)) {
-            return -2;
-        }
-    } catch (const std::exception & err) {
-        LLAMA_LOG_ERROR("%s: error loading model: %s\n", __func__, err.what());
-        return -1;
-    }
-
-    return 0;
-}
-
-static struct llama_model * llama_model_load_from_file_impl(
-        struct gguf_context * metadata,
-        llama_model_set_tensor_data_t set_tensor_data,
-        void * set_tensor_data_ud,
-        const std::string & path_model,
-        std::vector<std::string> & splits,
-        FILE * file,
-        struct llama_model_params params) {
-    {
-        int n_sources_defined = 0;
-        if (metadata != nullptr) {
-            n_sources_defined++;
-        }
-        if (!path_model.empty()) {
-            n_sources_defined++;
-        }
-        if (file != nullptr) {
-            n_sources_defined++;
-        }
-        if (n_sources_defined != 1) {
-            LLAMA_LOG_ERROR("%s: exactly one out metadata, path_model, and file must be defined\n", __func__);
-            return nullptr;
-        }
-    }
-    ggml_time_init();
-
-    if (!params.vocab_only && ggml_backend_reg_count() == 0) {
-        LLAMA_LOG_ERROR("%s: no backends are loaded. hint: use ggml_backend_load() or ggml_backend_load_all() to load a backend before calling this function\n", __func__);
-        return nullptr;
-    }
-
-    unsigned cur_percentage = 0;
-    if (params.progress_callback == NULL) {
-        params.progress_callback_user_data = &cur_percentage;
-        params.progress_callback = [](float progress, void * ctx) {
-            unsigned * cur_percentage_p = (unsigned *) ctx;
-            unsigned percentage = (unsigned) (100 * progress);
-            while (percentage > *cur_percentage_p) {
-                *cur_percentage_p = percentage;
-                LLAMA_LOG_CONT(".");
-                if (percentage >= 100) {
-                    LLAMA_LOG_CONT("\n");
-                }
-            }
-            return true;
-        };
-    }
-
-    llama_model * model = new llama_model(params);
-
+// returns true on success
+static bool llama_prepare_model_devices(const llama_model_params & params, llama_model * model) {
     // create list of devices to use with this model
     if (params.devices) {
         if (params.split_mode == LLAMA_SPLIT_MODE_TENSOR) {
@@ -227,7 +126,7 @@ static struct llama_model * llama_model_load_from_file_impl(
             }
             if (n_devs == 0) {
                 LLAMA_LOG_ERROR("%s: LLAMA_SPLIT_MODE_TENSOR needs >= 1 devices\n", __func__);
-                return nullptr;
+                return false;
             }
             LLAMA_LOG_INFO("%s: creating a Meta device with %zu devices\n", __func__, n_devs);
             for (size_t i = 0; i < n_devs; ++i) {
@@ -265,7 +164,7 @@ static struct llama_model * llama_model_load_from_file_impl(
             }
             if (devs.empty()) {
                 LLAMA_LOG_ERROR("%s: LLAMA_SPLIT_MODE_TENSOR needs >= 1 devices\n", __func__);
-                return nullptr;
+                return false;
             }
 
             LLAMA_LOG_INFO("%s: creating a Meta device for tensor parallelism from %zu devices:\n", __func__, devs.size());
@@ -347,8 +246,7 @@ static struct llama_model * llama_model_load_from_file_impl(
         } else {
             if (params.main_gpu >= (int)model->devices.size()) {
                 LLAMA_LOG_ERROR("%s: invalid value for main_gpu: %d (available devices: %zu)\n", __func__, params.main_gpu, model->devices.size());
-                llama_model_free(model);
-                return nullptr;
+                return false;
             }
             llama_device main_gpu = model->devices[params.main_gpu];
             model->devices.clear();
@@ -365,7 +263,121 @@ static struct llama_model * llama_model_load_from_file_impl(
                 props.memory_free/1024/1024);
     }
 
-    const int status = llama_model_load(metadata, set_tensor_data, set_tensor_data_ud, path_model, splits, file, *model, params);
+    return true;
+}
+
+// Returns 0 on success, -1 on error, and -2 on cancellation via llama_progress_callback
+static std::pair<int, llama_model *> llama_model_load(struct gguf_context * metadata, llama_model_set_tensor_data_t set_tensor_data, void * set_tensor_data_ud,
+        const std::string & fname, std::vector<std::string> & splits, FILE * file, llama_model_params & params) {
+    try {
+        llama_model_loader ml(metadata, set_tensor_data, set_tensor_data_ud, fname, splits, file, params.use_mmap, params.use_direct_io,
+            params.check_tensors, params.no_alloc, params.kv_overrides, params.tensor_buft_overrides);
+
+        ml.print_info();
+        std::unique_ptr<llama_model> model_ptr(llama_model_create(ml, params));
+
+        bool ok = llama_prepare_model_devices(params, model_ptr.get());
+        if (!ok) {
+            return {-1, nullptr};
+        }
+
+        auto * model = dynamic_cast<llama_model_base *>(model_ptr.get());
+        if (model == nullptr) {
+            GGML_ABORT("fatal error: model does not implement llama_model_base");
+        }
+
+        // loading time will be recalculated after the first eval, so
+        // we take page faults deferred by mmap() into consideration
+        model->t_load_us = 0;
+        time_meas tm(model->t_load_us);
+
+        model->t_start_us = tm.t_start_us;
+
+        model->hparams.vocab_only = params.vocab_only;
+        model->hparams.no_alloc   = params.no_alloc;
+
+        try {
+            model->load_hparams(ml);
+        } catch(const std::exception & e) {
+            throw std::runtime_error("error loading model hyperparameters: " + std::string(e.what()));
+        }
+        if (model->arch == LLM_ARCH_CLIP) {
+            throw std::runtime_error("CLIP cannot be used as main model, use it with --mmproj instead");
+        }
+        try {
+            model->load_vocab(ml);
+        } catch(const std::exception & e) {
+            throw std::runtime_error("error loading model vocabulary: " + std::string(e.what()));
+        }
+
+        model->load_stats(ml);
+        model->print_info();
+
+        if (params.vocab_only) {
+            LLAMA_LOG_INFO("%s: vocab only - skipping tensors\n", __func__);
+            return {0, model_ptr.release()};
+        }
+
+        if (!model->load_tensors(ml)) {
+            return {-2, nullptr};
+        }
+
+        return {0, model_ptr.release()};
+    } catch (const std::exception & err) {
+        LLAMA_LOG_ERROR("%s: error loading model: %s\n", __func__, err.what());
+        return {-1, nullptr};
+    }
+}
+
+static struct llama_model * llama_model_load_from_file_impl(
+        struct gguf_context * metadata,
+        llama_model_set_tensor_data_t set_tensor_data,
+        void * set_tensor_data_ud,
+        const std::string & path_model,
+        std::vector<std::string> & splits,
+        FILE * file,
+        struct llama_model_params params) {
+    {
+        int n_sources_defined = 0;
+        if (metadata != nullptr) {
+            n_sources_defined++;
+        }
+        if (!path_model.empty()) {
+            n_sources_defined++;
+        }
+        if (file != nullptr) {
+            n_sources_defined++;
+        }
+        if (n_sources_defined != 1) {
+            LLAMA_LOG_ERROR("%s: exactly one out metadata, path_model, and file must be defined\n", __func__);
+            return nullptr;
+        }
+    }
+    ggml_time_init();
+
+    if (!params.vocab_only && ggml_backend_reg_count() == 0) {
+        LLAMA_LOG_ERROR("%s: no backends are loaded. hint: use ggml_backend_load() or ggml_backend_load_all() to load a backend before calling this function\n", __func__);
+        return nullptr;
+    }
+
+    unsigned cur_percentage = 0;
+    if (params.progress_callback == NULL) {
+        params.progress_callback_user_data = &cur_percentage;
+        params.progress_callback = [](float progress, void * ctx) {
+            unsigned * cur_percentage_p = (unsigned *) ctx;
+            unsigned percentage = (unsigned) (100 * progress);
+            while (percentage > *cur_percentage_p) {
+                *cur_percentage_p = percentage;
+                LLAMA_LOG_CONT(".");
+                if (percentage >= 100) {
+                    LLAMA_LOG_CONT("\n");
+                }
+            }
+            return true;
+        };
+    }
+
+    const auto [status, model] = llama_model_load(metadata, set_tensor_data, set_tensor_data_ud, path_model, splits, file, params);
     GGML_ASSERT(status <= 0);
     if (status < 0) {
         if (status == -1) {
@@ -374,7 +386,9 @@ static struct llama_model * llama_model_load_from_file_impl(
             LLAMA_LOG_INFO("%s: cancelled model load\n", __func__);
         }
 
-        llama_model_free(model);
+        if (model) {
+            llama_model_free(model);
+        }
         return nullptr;
     }
 

src/models/afmoe.cpp

@@ -1,6 +1,112 @@
 #include "models.h"
 
-llm_build_afmoe::llm_build_afmoe(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
+void llama_model_afmoe::load_arch_hparams(llama_model_loader & ml) {
+    ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
+    ml.get_key(LLM_KV_LEADING_DENSE_BLOCK_COUNT,   hparams.n_layer_dense_lead, false);
+    ml.get_key(LLM_KV_EXPERT_FEED_FORWARD_LENGTH,  hparams.n_ff_exp);
+    ml.get_key(LLM_KV_EXPERT_SHARED_COUNT,         hparams.n_expert_shared);
+    ml.get_key(LLM_KV_EXPERT_GATING_FUNC,          hparams.expert_gating_func, false);
+    ml.get_key(LLM_KV_EXPERT_WEIGHTS_SCALE,        hparams.expert_weights_scale, false);
+    ml.get_key(LLM_KV_EXPERT_WEIGHTS_NORM,         hparams.expert_weights_norm, false);
+    ml.get_key(LLM_KV_ATTENTION_SLIDING_WINDOW,    hparams.n_swa, false);
+
+    // Set up interleaved sliding window attention (ISWA)
+    // Pattern: 3 sliding - 1 full (global_attn_every_n_layers = 4)
+    if (hparams.n_swa > 0) {
+        hparams.swa_type = LLAMA_SWA_TYPE_STANDARD;
+        uint32_t swa_period = 4;
+        ml.get_key_or_arr(LLM_KV_ATTENTION_SLIDING_WINDOW_PATTERN, swa_period, false);
+        hparams.set_swa_pattern(swa_period);
+
+        hparams.rope_freq_base_train_swa  = hparams.rope_freq_base_train;
+        hparams.rope_freq_scale_train_swa = hparams.rope_freq_scale_train;
+        ml.get_key(LLM_KV_ROPE_FREQ_BASE_SWA, hparams.rope_freq_base_train_swa, false);
+    } else {
+        hparams.swa_type = LLAMA_SWA_TYPE_NONE;
+    }
+
+    // Default to sigmoid if not set
+    if (hparams.expert_gating_func == LLAMA_EXPERT_GATING_FUNC_TYPE_NONE) {
+        hparams.expert_gating_func = LLAMA_EXPERT_GATING_FUNC_TYPE_SIGMOID;
+    }
+
+    switch (hparams.n_layer) {
+        case 56: type = LLM_TYPE_6B; break;
+        case 32: type = LLM_TYPE_26B; break;
+        default: type = LLM_TYPE_UNKNOWN;
+    }
+}
+
+void llama_model_afmoe::load_arch_tensors(llama_model_loader &) {
+    LLAMA_LOAD_LOCALS;
+    const int64_t n_expert_shared = hparams.n_expert_shared;
+
+    tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+
+    // output
+    output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
+    output      = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, TENSOR_NOT_REQUIRED);
+
+    // if output is NULL, init from the input tok embed
+    if (output == NULL) {
+        output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, TENSOR_DUPLICATED);
+    }
+
+    const int64_t n_ff_exp = hparams.n_ff_exp;
+
+    for (int i = 0; i < n_layer; ++i) {
+        auto & layer = layers[i];
+
+        // dual attention normalization
+        layer.attn_norm      = create_tensor(tn(LLM_TENSOR_ATTN_NORM,      "weight", i), {n_embd}, 0);
+        layer.attn_post_norm = create_tensor(tn(LLM_TENSOR_ATTN_POST_NORM, "weight", i), {n_embd}, 0);
+
+        // attention projections
+        create_tensor_qkv(layer, i, n_embd, n_embd_head_k * n_head, n_embd_k_gqa, n_embd_v_gqa, 0);
+        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd_head_k * n_head, n_embd}, 0);
+
+        // Q/K normalization
+        layer.attn_q_norm = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "weight", i), {n_embd_head_k}, 0);
+        layer.attn_k_norm = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "weight", i), {n_embd_head_k}, 0);
+
+        // attention gating
+        layer.wqkv_gate = create_tensor(tn(LLM_TENSOR_ATTN_GATE, "weight", i), {n_embd, n_embd_head_k * n_head}, 0);
+
+        // dual ffn normalization
+        layer.ffn_norm      = create_tensor(tn(LLM_TENSOR_FFN_NORM,      "weight", i), {n_embd}, 0);
+        layer.ffn_post_norm = create_tensor(tn(LLM_TENSOR_FFN_POST_NORM, "weight", i), {n_embd}, 0);
+
+        if (static_cast<uint32_t>(i) >= hparams.n_layer_dense_lead) {
+            // MoE layers
+            layer.ffn_gate_inp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP, "weight", i), {n_embd, n_expert}, 0);
+            layer.ffn_exp_probs_b = create_tensor(tn(LLM_TENSOR_FFN_EXP_PROBS_B, "bias", i), {n_expert}, 0);
+
+            // grouped expert weights
+            layer.ffn_gate_exps = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), {n_embd, n_ff_exp, n_expert}, 0);
+            layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), {n_ff_exp, n_embd, n_expert}, 0);
+            layer.ffn_up_exps   = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS,   "weight", i), {n_embd, n_ff_exp, n_expert}, 0);
+
+            // shared expert
+            if (n_expert_shared > 0) {
+                const int64_t n_ff_shexp = n_ff_exp * n_expert_shared;
+                layer.ffn_gate_shexp = create_tensor(tn(LLM_TENSOR_FFN_GATE_SHEXP, "weight", i), {n_embd, n_ff_shexp}, 0);
+                layer.ffn_down_shexp = create_tensor(tn(LLM_TENSOR_FFN_DOWN_SHEXP, "weight", i), {n_ff_shexp, n_embd}, 0);
+                layer.ffn_up_shexp   = create_tensor(tn(LLM_TENSOR_FFN_UP_SHEXP,   "weight", i), {n_embd, n_ff_shexp}, 0);
+            }
+        } else {
+            // Dense layers
+            layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0);
+            layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0);
+            layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd, n_ff}, 0);
+        }
+    }
+}
+
+std::unique_ptr<llm_graph_context> llama_model_afmoe::build_arch_graph(const llm_graph_params & params) const {
+    return std::make_unique<graph>(*this, params);
+}
+
+llama_model_afmoe::graph::graph(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
     const int64_t n_embd_head = hparams.n_embd_head_v();
     GGML_ASSERT(n_embd_head == hparams.n_embd_head_k());
 

src/models/apertus.cpp

@@ -1,6 +1,62 @@
 #include "models.h"
 
-llm_build_apertus::llm_build_apertus(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
+void llama_model_apertus::load_arch_hparams(llama_model_loader & ml) {
+    ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
+    ml.get_key_or_arr(LLM_KV_XIELU_ALPHA_N,        hparams.xielu_alpha_n, hparams.n_layer);
+    ml.get_key_or_arr(LLM_KV_XIELU_ALPHA_P,        hparams.xielu_alpha_p, hparams.n_layer);
+    ml.get_key_or_arr(LLM_KV_XIELU_BETA,           hparams.xielu_beta,    hparams.n_layer);
+    ml.get_key_or_arr(LLM_KV_XIELU_EPS,            hparams.xielu_eps,     hparams.n_layer);
+
+    switch (hparams.n_layer) {
+        case 32: type = LLM_TYPE_8B; break;
+        default: type = LLM_TYPE_UNKNOWN;
+    }
+}
+
+void llama_model_apertus::load_arch_tensors(llama_model_loader &) {
+    LLAMA_LOAD_LOCALS;
+
+    tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), { n_embd, n_vocab }, 0);
+
+    // output
+    output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), { n_embd }, 0);
+    output      = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), { n_embd, n_vocab }, 0);
+
+    for (int i = 0; i < n_layer; ++i) {
+        auto & layer = layers[i];
+
+        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), { n_embd }, 0);
+
+        if (hparams.rope_scaling_type_train == LLAMA_ROPE_SCALING_TYPE_LONGROPE) {
+            layer.rope_long  = create_tensor(tn(LLM_TENSOR_ROPE_FACTORS_LONG,  "weight", i), { n_rot/2 }, TENSOR_NOT_REQUIRED | (i != 0 ? TENSOR_DUPLICATED : 0));
+            layer.rope_short = create_tensor(tn(LLM_TENSOR_ROPE_FACTORS_SHORT, "weight", i), { n_rot/2 }, TENSOR_NOT_REQUIRED | (i != 0 ? TENSOR_DUPLICATED : 0));
+        } else {
+            layer.rope_freqs = create_tensor(tn(LLM_TENSOR_ROPE_FREQS, "weight", i), { n_rot/2 }, TENSOR_NOT_REQUIRED | (i != 0 ? TENSOR_DUPLICATED : 0));
+        }
+
+        create_tensor_qkv(layer, i, n_embd, n_embd_head_k * n_head, n_embd_gqa, n_embd_gqa, 0);
+        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), { n_embd_head_k * n_head, n_embd }, 0);
+
+        // optional bias tensors
+        layer.wo_b = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i), { n_embd }, TENSOR_NOT_REQUIRED);
+
+        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), { n_embd }, 0);
+        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd }, 0);
+        layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), { n_embd, n_ff }, 0);
+
+        // Q and K layernorms for Apertus
+        layer.attn_q_norm   = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "weight", i), { n_embd_head_k }, 0);
+        layer.attn_q_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "bias",   i), { n_embd_head_k }, TENSOR_NOT_REQUIRED);
+        layer.attn_k_norm   = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "weight", i), { n_embd_head_k }, 0);
+        layer.attn_k_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "bias",   i), { n_embd_head_k }, TENSOR_NOT_REQUIRED);
+    }
+}
+
+std::unique_ptr<llm_graph_context> llama_model_apertus::build_arch_graph(const llm_graph_params & params) const {
+    return std::make_unique<graph>(*this, params);
+}
+
+llama_model_apertus::graph::graph(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
     const int64_t n_embd_head = hparams.n_embd_head_v();
 
     GGML_ASSERT(n_embd_head == hparams.n_embd_head_k());

src/models/arcee.cpp

@@ -1,6 +1,51 @@
 #include "models.h"
 
-llm_build_arcee::llm_build_arcee(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
+void llama_model_arcee::load_arch_hparams(llama_model_loader & ml) {
+    ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
+
+    // Arcee uses the same structure as Llama
+    switch (hparams.n_layer) {
+        case 36: type = LLM_TYPE_4B; break;
+        default: type = LLM_TYPE_UNKNOWN;
+    }
+}
+
+void llama_model_arcee::load_arch_tensors(llama_model_loader &) {
+    LLAMA_LOAD_LOCALS;
+
+    tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+
+    // output
+    output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
+    output      = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, TENSOR_NOT_REQUIRED);
+
+    // if output is NULL, init from the input tok embed
+    if (output == NULL) {
+        output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, TENSOR_DUPLICATED);
+    }
+
+    for (int i = 0; i < n_layer; ++i) {
+        auto & layer = layers[i];
+
+        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
+
+        create_tensor_qkv(layer, i, n_embd, n_embd_head_k * n_head, n_embd_k_gqa, n_embd_v_gqa, 0);
+        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd_head_k * n_head, n_embd}, 0);
+
+        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
+
+        layer.rope_freqs = create_tensor(tn(LLM_TENSOR_ROPE_FREQS, "weight", i), {n_rot/2}, TENSOR_NOT_REQUIRED | (i != 0 ? TENSOR_DUPLICATED : 0));
+
+        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {  n_ff, n_embd}, 0);
+        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd,   n_ff}, 0);
+    }
+}
+
+std::unique_ptr<llm_graph_context> llama_model_arcee::build_arch_graph(const llm_graph_params & params) const {
+    return std::make_unique<graph>(*this, params);
+}
+
+llama_model_arcee::graph::graph(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
     const int64_t n_embd_head = hparams.n_embd_head_v();
 
     GGML_ASSERT(n_embd_head == hparams.n_embd_head_k());

src/models/arctic.cpp

@@ -1,6 +1,59 @@
 #include "models.h"
 
-llm_build_arctic::llm_build_arctic(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
+void llama_model_arctic::load_arch_hparams(llama_model_loader & ml) {
+    ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
+
+    if (hparams.n_expert == 128) {
+        switch (hparams.n_layer) {
+            case 35: type = LLM_TYPE_10B_128x3_66B; break;
+            default: type = LLM_TYPE_UNKNOWN;
+        }
+    } else {
+        type = LLM_TYPE_UNKNOWN;
+    }
+}
+
+void llama_model_arctic::load_arch_tensors(llama_model_loader &) {
+    LLAMA_LOAD_LOCALS;
+
+    tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+
+    // output
+    output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
+    output      = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, TENSOR_NOT_REQUIRED);
+
+    // if output is NULL, init from the input tok embed
+    if (output == NULL) {
+        output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, TENSOR_DUPLICATED);
+    }
+
+    for (int i = 0; i < n_layer; ++i) {
+        auto & layer = layers[i];
+
+        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
+
+        create_tensor_qkv(layer, i, n_embd, n_embd, n_embd_gqa, n_embd_gqa, 0);
+        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
+
+        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
+
+        layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_embd}, 0);
+        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_embd, n_embd}, 0);
+        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd, n_embd}, 0);
+
+        layer.ffn_gate_inp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP, "weight", i), {n_embd, n_expert}, 0);
+        layer.ffn_norm_exps = create_tensor(tn(LLM_TENSOR_FFN_NORM_EXPS, "weight", i), {n_embd}, 0);
+        layer.ffn_gate_exps = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), {n_embd,   n_ff, n_expert}, false);
+        layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), {  n_ff, n_embd, n_expert}, 0);
+        layer.ffn_up_exps   = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS,   "weight", i), {n_embd,   n_ff, n_expert}, 0);
+    }
+}
+
+std::unique_ptr<llm_graph_context> llama_model_arctic::build_arch_graph(const llm_graph_params & params) const {
+    return std::make_unique<graph>(*this, params);
+}
+
+llama_model_arctic::graph::graph(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
     const int64_t n_embd_head = hparams.n_embd_head_v();
 
     GGML_ASSERT(n_embd_head == hparams.n_embd_head_k());

src/models/arwkv7.cpp

@@ -1,7 +1,123 @@
 #include "models.h"
 
+void llama_model_arwkv7::load_arch_hparams(llama_model_loader & ml) {
+    ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS,                hparams.f_norm_eps, false);
+    ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS,            hparams.f_norm_rms_eps, false);
+    ml.get_key(LLM_KV_WKV_HEAD_SIZE,                          hparams.wkv_head_size);
+    ml.get_key(LLM_KV_ATTENTION_DECAY_LORA_RANK,              hparams.n_lora_decay);
+    ml.get_key(LLM_KV_ATTENTION_ICLR_LORA_RANK,               hparams.n_lora_iclr);
+    ml.get_key(LLM_KV_ATTENTION_VALUE_RESIDUAL_MIX_LORA_RANK, hparams.n_lora_value_res_mix);
+    ml.get_key(LLM_KV_ATTENTION_GATE_LORA_RANK,               hparams.n_lora_gate, false);
+    ml.get_key(LLM_KV_TOKEN_SHIFT_COUNT,                      hparams.token_shift_count, false);
 
-llm_build_arwkv7::llm_build_arwkv7(const llama_model & model, const llm_graph_params & params) : llm_build_rwkv7_base(model, params) {
+    switch (hparams.n_layer) {
+        case 12:
+            switch (hparams.n_embd) {
+                case 768: type = LLM_TYPE_190M; break;
+                default: type = LLM_TYPE_UNKNOWN;
+            } break;
+        case 24:
+            switch (hparams.n_embd) {
+                case 1024: type = LLM_TYPE_450M; break;
+                case 2048: type = LLM_TYPE_1_5B; break;
+                default: type = LLM_TYPE_UNKNOWN;
+            } break;
+        case 28:
+            switch (hparams.n_embd) {
+                case 1536: type = LLM_TYPE_1_5B; break;
+                case 3584: type = LLM_TYPE_7B; break;
+                default: type = LLM_TYPE_UNKNOWN;
+            } break;
+        case 32:
+            switch (hparams.n_embd) {
+                case 2560: type = LLM_TYPE_2_9B; break;
+                case 4096: type = LLM_TYPE_7B; break;
+                default: type = LLM_TYPE_UNKNOWN;
+            } break;
+        case 61:
+            switch (hparams.n_embd) {
+                case 4096: type = LLM_TYPE_14B; break;
+                default: type = LLM_TYPE_UNKNOWN;
+            } break;
+        default: type = LLM_TYPE_UNKNOWN;
+    }
+}
+
+void llama_model_arwkv7::load_arch_tensors(llama_model_loader &) {
+    LLAMA_LOAD_LOCALS;
+
+    tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+
+    // output
+    output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
+    output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, 0);
+
+    const int n_lora_decay = hparams.n_lora_decay;
+    const int n_lora_iclr = hparams.n_lora_iclr;
+    const int n_lora_value_res_mix = hparams.n_lora_value_res_mix;
+    const int n_lora_gate = hparams.n_lora_gate;
+    const int attn_hidden_size = n_embd;
+
+    for (int i = 0; i < n_layer; ++i) {
+        auto & layer = layers[i];
+
+        layer.attn_norm   = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
+
+        layer.time_mix_w0 = create_tensor(tn(LLM_TENSOR_TIME_MIX_W0, "weight", i), {n_embd}, 0);
+        layer.time_mix_w1 = create_tensor(tn(LLM_TENSOR_TIME_MIX_W1, "weight", i), {n_embd, n_lora_decay}, 0);
+        layer.time_mix_w2 = create_tensor(tn(LLM_TENSOR_TIME_MIX_W2, "weight", i), {n_lora_decay, n_embd}, 0);
+
+        layer.time_mix_a0 = create_tensor(tn(LLM_TENSOR_TIME_MIX_A0, "weight", i), {n_embd}, 0);
+        layer.time_mix_a1 = create_tensor(tn(LLM_TENSOR_TIME_MIX_A1, "weight", i), {n_embd, n_lora_iclr}, 0);
+        layer.time_mix_a2 = create_tensor(tn(LLM_TENSOR_TIME_MIX_A2, "weight", i), {n_lora_iclr, n_embd}, 0);
+
+        if (i == 0) {
+            // actually not used
+            layer.time_mix_v0 = create_tensor(tn(LLM_TENSOR_TIME_MIX_V0, "weight", i), {n_embd}, 0);
+            layer.time_mix_v1 = create_tensor(tn(LLM_TENSOR_TIME_MIX_V1, "weight", i), {n_embd, n_lora_iclr}, 0);
+            layer.time_mix_v2 = create_tensor(tn(LLM_TENSOR_TIME_MIX_V2, "weight", i), {n_lora_iclr, n_embd}, 0);
+        } else {
+            layer.time_mix_v0 = create_tensor(tn(LLM_TENSOR_TIME_MIX_V0, "weight", i), {n_embd}, 0);
+            layer.time_mix_v1 = create_tensor(tn(LLM_TENSOR_TIME_MIX_V1, "weight", i), {n_embd, n_lora_value_res_mix}, 0);
+            layer.time_mix_v2 = create_tensor(tn(LLM_TENSOR_TIME_MIX_V2, "weight", i), {n_lora_value_res_mix, n_embd}, 0);
+        }
+
+        layer.time_mix_g1 = create_tensor(tn(LLM_TENSOR_TIME_MIX_G1, "weight", i), {n_embd, n_lora_gate}, TENSOR_NOT_REQUIRED);
+        layer.time_mix_g2 = create_tensor(tn(LLM_TENSOR_TIME_MIX_G2, "weight", i), {n_lora_gate, n_embd}, TENSOR_NOT_REQUIRED);
+
+        try {
+            layer.time_mix_lerp_fused = create_tensor(tn(LLM_TENSOR_TIME_MIX_LERP_FUSED, "weight", i), {n_embd, 1, 1, 6}, 0);
+        } catch(std::runtime_error & e) {
+            // ARWKV models may not have gate tensors
+            layer.time_mix_lerp_fused = create_tensor(tn(LLM_TENSOR_TIME_MIX_LERP_FUSED, "weight", i), {n_embd, 1, 1, 5}, 0);
+        }
+
+        layer.time_mix_k_k = create_tensor(tn(LLM_TENSOR_TIME_MIX_K_K, "weight", i), {attn_hidden_size}, 0);
+        layer.time_mix_k_a = create_tensor(tn(LLM_TENSOR_TIME_MIX_K_A, "weight", i), {attn_hidden_size}, 0);
+        layer.time_mix_r_k = create_tensor(tn(LLM_TENSOR_TIME_MIX_R_K, "weight", i), {attn_hidden_size}, 0);
+
+        layer.time_mix_key = create_tensor(tn(LLM_TENSOR_TIME_MIX_KEY, "weight", i), {attn_hidden_size, n_embd}, 0);
+        layer.time_mix_value = create_tensor(tn(LLM_TENSOR_TIME_MIX_VALUE, "weight", i), {attn_hidden_size, n_embd}, 0);
+        layer.time_mix_receptance = create_tensor(tn(LLM_TENSOR_TIME_MIX_RECEPTANCE, "weight", i), {attn_hidden_size, n_embd}, 0);
+
+        layer.time_mix_ln = create_tensor(tn(LLM_TENSOR_TIME_MIX_LN, "weight", i), {n_embd}, TENSOR_NOT_REQUIRED);
+        layer.time_mix_ln_b = create_tensor(tn(LLM_TENSOR_TIME_MIX_LN, "bias", i), {n_embd}, TENSOR_NOT_REQUIRED);
+        layer.time_mix_output = create_tensor(tn(LLM_TENSOR_TIME_MIX_OUTPUT, "weight", i), {n_embd, attn_hidden_size}, 0);
+
+        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
+
+        layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd,   n_ff}, 0);
+        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {  n_ff, n_embd}, 0);
+        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd,   n_ff}, 0);
+    }
+
+}
+
+std::unique_ptr<llm_graph_context> llama_model_arwkv7::build_arch_graph(const llm_graph_params & params) const {
+    return std::make_unique<graph>(*this, params);
+}
+
+llama_model_arwkv7::graph::graph(const llama_model & model, const llm_graph_params & params) : llm_build_rwkv7_base(model, params) {
     GGML_ASSERT(n_embd == hparams.n_embd_r());
 
     ggml_tensor * cur;

src/models/baichuan.cpp

@@ -1,6 +1,49 @@
 #include "models.h"
 
-llm_build_baichuan::llm_build_baichuan(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
+void llama_model_baichuan::load_arch_hparams(llama_model_loader & ml) {
+    ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
+    switch (hparams.n_layer) {
+        case 32: type = LLM_TYPE_7B; break;
+        case 40: type = LLM_TYPE_13B; break;
+        default: type = LLM_TYPE_UNKNOWN;
+    }
+
+    if (type == LLM_TYPE_13B) {
+        // TODO: become GGUF KV parameter
+        hparams.f_max_alibi_bias = 8.0f;
+    }
+}
+
+void llama_model_baichuan::load_arch_tensors(llama_model_loader &) {
+    LLAMA_LOAD_LOCALS;
+
+    tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+    {
+        output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
+        output      = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, 0);
+    }
+
+    for (int i = 0; i < n_layer; ++i) {
+        auto & layer = layers[i];
+
+        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
+
+        create_tensor_qkv(layer, i, n_embd, n_embd, n_embd_gqa, n_embd_gqa, 0);
+        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
+
+        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
+
+        layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd,   n_ff}, 0);
+        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {  n_ff, n_embd}, 0);
+        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd,   n_ff}, 0);
+    }
+}
+
+std::unique_ptr<llm_graph_context> llama_model_baichuan::build_arch_graph(const llm_graph_params & params) const {
+    return std::make_unique<graph>(*this, params);
+}
+
+llama_model_baichuan::graph::graph(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
     const int64_t n_embd_head = hparams.n_embd_head_v();
 
     GGML_ASSERT(n_embd_head == hparams.n_embd_head_k());

src/models/bailingmoe.cpp

@@ -1,6 +1,65 @@
 #include "models.h"
 
-llm_build_bailingmoe::llm_build_bailingmoe(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
+void llama_model_bailingmoe::load_arch_hparams(llama_model_loader & ml) {
+    ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
+    ml.get_key(LLM_KV_LEADING_DENSE_BLOCK_COUNT,   hparams.n_layer_dense_lead, false);
+    ml.get_key(LLM_KV_EXPERT_FEED_FORWARD_LENGTH,  hparams.n_ff_exp);
+    ml.get_key(LLM_KV_EXPERT_SHARED_COUNT,         hparams.n_expert_shared);
+    ml.get_key(LLM_KV_EXPERT_WEIGHTS_SCALE,        hparams.expert_weights_scale, false);
+    ml.get_key(LLM_KV_EXPERT_WEIGHTS_NORM,         hparams.expert_weights_norm, false);
+
+    switch (hparams.n_layer) {
+        case 28: type = LLM_TYPE_16B; break;
+        case 88: type = LLM_TYPE_290B; break;
+        default: type = LLM_TYPE_UNKNOWN;
+    }
+}
+
+void llama_model_bailingmoe::load_arch_tensors(llama_model_loader &) {
+    LLAMA_LOAD_LOCALS;
+    const int64_t n_expert_shared = hparams.n_expert_shared;
+
+    const int64_t n_ff_exp            = hparams.n_ff_exp;
+
+    tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+
+    // output
+    output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
+    output      = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, 0);
+
+    for (int i = 0; i < n_layer; ++i) {
+        auto & layer = layers[i];
+
+        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
+
+        create_tensor_qkv(layer, i, n_embd, n_head * n_rot, n_head_kv * n_rot, n_head_kv * n_rot, 0);
+        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_head * n_rot, n_embd}, 0);
+        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
+
+        layer.ffn_gate_inp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP, "weight", i), {n_embd, n_expert}, 0);
+
+        if (n_expert == 0) {
+            throw std::runtime_error("n_expert must be > 0");
+        }
+        if (n_expert_used == 0) {
+            throw std::runtime_error("n_expert_used must be > 0");
+        }
+
+        layer.ffn_gate_exps = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), {  n_embd, n_ff_exp, n_expert}, 0);
+        layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), {n_ff_exp,   n_embd, n_expert}, 0);
+        layer.ffn_up_exps   = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS,   "weight", i), {  n_embd, n_ff_exp, n_expert}, 0);
+
+        layer.ffn_gate_shexp = create_tensor(tn(LLM_TENSOR_FFN_GATE_SHEXP, "weight", i), {n_embd, n_ff_exp * n_expert_shared}, 0);
+        layer.ffn_down_shexp = create_tensor(tn(LLM_TENSOR_FFN_DOWN_SHEXP, "weight", i), {        n_ff_exp * n_expert_shared, n_embd}, 0);
+        layer.ffn_up_shexp   = create_tensor(tn(LLM_TENSOR_FFN_UP_SHEXP,   "weight", i), {n_embd, n_ff_exp * n_expert_shared}, 0);
+    }
+}
+
+std::unique_ptr<llm_graph_context> llama_model_bailingmoe::build_arch_graph(const llm_graph_params & params) const {
+    return std::make_unique<graph>(*this, params);
+}
+
+llama_model_bailingmoe::graph::graph(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
     ggml_tensor * cur;
     ggml_tensor * inpL;
 

src/models/bailingmoe2.cpp

@@ -1,6 +1,100 @@
 #include "models.h"
 
-llm_build_bailingmoe2::llm_build_bailingmoe2(const llama_model & model, const llm_graph_params & params) :
+void llama_model_bailingmoe2::load_arch_hparams(llama_model_loader & ml) {
+    ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS,       hparams.f_norm_rms_eps);
+    ml.get_key(LLM_KV_LEADING_DENSE_BLOCK_COUNT,         hparams.n_layer_dense_lead, false);
+    ml.get_key(LLM_KV_EXPERT_FEED_FORWARD_LENGTH,        hparams.n_ff_exp);
+    ml.get_key(LLM_KV_EXPERT_SHARED_FEED_FORWARD_LENGTH, hparams.n_ff_shexp, false);
+    ml.get_key(LLM_KV_EXPERT_SHARED_COUNT,               hparams.n_expert_shared);
+    ml.get_key(LLM_KV_EXPERT_WEIGHTS_SCALE,              hparams.expert_weights_scale, false);
+    ml.get_key(LLM_KV_EXPERT_WEIGHTS_NORM,               hparams.expert_weights_norm, false);
+    ml.get_key(LLM_KV_EXPERT_GATING_FUNC,                hparams.expert_gating_func);
+    ml.get_key(LLM_KV_NEXTN_PREDICT_LAYERS,              hparams.nextn_predict_layers, false);
+    GGML_ASSERT(hparams.nextn_predict_layers < hparams.n_layer && "nextn_predict_layers must be < n_layer");
+
+    // TODO: when MTP is implemented, this should probably be updated if needed
+    hparams.n_layer_kv_from_start = hparams.n_layer - hparams.nextn_predict_layers;
+
+    switch (hparams.n_layer) {
+        case 20: type = LLM_TYPE_16B_A1B; break;
+        case 21: type = LLM_TYPE_16B_A1B; break;
+        case 32: type = LLM_TYPE_100B_A6B; break;
+        case 33: type = LLM_TYPE_100B_A6B; break;
+        default: type = LLM_TYPE_UNKNOWN;
+    }
+}
+
+void llama_model_bailingmoe2::load_arch_tensors(llama_model_loader &) {
+    LLAMA_LOAD_LOCALS;
+    const int64_t n_expert_shared = hparams.n_expert_shared;
+
+    const int64_t n_ff_exp        = hparams.n_ff_exp;
+
+    tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+
+    // output
+    output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
+    output      = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, 0);
+
+    GGML_ASSERT(n_expert > 0 && "n_expert must be > 0 for bailingmoe2");
+    GGML_ASSERT(n_expert_used > 0 && "n_expert_used must be > 0 for bailingmoe2");
+
+    for (int i = 0; i < n_layer; ++i) {
+        int flags = 0;
+        if (hparams.nextn_predict_layers > 0 && static_cast<uint32_t>(i) >= n_layer - hparams.nextn_predict_layers) {
+            // skip all tensors in the NextN layers
+            flags |= TENSOR_SKIP;
+        }
+
+        auto & layer = layers[i];
+
+        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, flags);
+
+        layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, flags);
+        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd_head_k * n_head, n_embd}, flags);
+
+        layer.attn_q_norm = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "weight", i), {n_embd_head_k}, flags);
+        layer.attn_k_norm = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "weight", i), {n_embd_head_k}, flags);
+
+        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, flags);
+
+        if (static_cast<uint32_t>(i) >= hparams.n_layer_dense_lead) { // MoE layers
+            const int64_t n_ff_shexp = (hparams.n_ff_shexp ? hparams.n_ff_shexp : n_ff_exp) * n_expert_shared;
+
+            layer.ffn_gate_inp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP, "weight", i), {n_embd, n_expert}, flags);
+            layer.ffn_exp_probs_b = create_tensor(tn(LLM_TENSOR_FFN_EXP_PROBS_B, "bias", i), {n_expert}, TENSOR_NOT_REQUIRED | flags);
+
+            layer.ffn_gate_exps = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), {  n_embd, n_ff_exp, n_expert}, flags);
+            layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), {n_ff_exp,   n_embd, n_expert}, flags);
+            layer.ffn_up_exps   = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS,   "weight", i), {  n_embd, n_ff_exp, n_expert}, flags);
+
+            layer.ffn_gate_shexp = create_tensor(tn(LLM_TENSOR_FFN_GATE_SHEXP, "weight", i), {n_embd, n_ff_shexp}, flags);
+            layer.ffn_down_shexp = create_tensor(tn(LLM_TENSOR_FFN_DOWN_SHEXP, "weight", i), {n_ff_shexp, n_embd}, flags);
+            layer.ffn_up_shexp   = create_tensor(tn(LLM_TENSOR_FFN_UP_SHEXP,   "weight", i), {n_embd, n_ff_shexp}, flags);
+        } else { // Dense layers
+            layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd,   n_ff}, flags);
+            layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {  n_ff, n_embd}, flags);
+            layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd,   n_ff}, flags);
+        }
+
+        // NextN/MTP tensors (preserved but unused) - conditionally load for last nextn_predict_layers
+        if (hparams.nextn_predict_layers > 0 && static_cast<uint32_t>(i) >= n_layer - hparams.nextn_predict_layers) {
+            layer.nextn.eh_proj          = create_tensor(tn(LLM_TENSOR_NEXTN_EH_PROJ, "weight", i), { 2 * n_embd, n_embd }, flags);
+            layer.nextn.embed_tokens     = create_tensor(tn(LLM_TENSOR_NEXTN_EMBED_TOKENS, "weight", i), { n_embd, n_vocab }, TENSOR_NOT_REQUIRED | flags);
+            layer.nextn.enorm            = create_tensor(tn(LLM_TENSOR_NEXTN_ENORM, "weight", i), { n_embd }, flags);
+            layer.nextn.hnorm            = create_tensor(tn(LLM_TENSOR_NEXTN_HNORM, "weight", i), { n_embd }, flags);
+            layer.nextn.shared_head_head = create_tensor(tn(LLM_TENSOR_NEXTN_SHARED_HEAD_HEAD, "weight", i), { n_embd, n_vocab }, TENSOR_NOT_REQUIRED | flags);
+            layer.nextn.shared_head_norm = create_tensor(tn(LLM_TENSOR_NEXTN_SHARED_HEAD_NORM, "weight", i), { n_embd }, TENSOR_NOT_REQUIRED | flags);
+            layer.layer_out_norm         = create_tensor(tn(LLM_TENSOR_LAYER_OUT_NORM, "weight", i), {n_embd}, flags);
+        }
+    }
+}
+
+std::unique_ptr<llm_graph_context> llama_model_bailingmoe2::build_arch_graph(const llm_graph_params & params) const {
+    return std::make_unique<graph>(*this, params);
+}
+
+llama_model_bailingmoe2::graph::graph(const llama_model & model, const llm_graph_params & params) :
     llm_graph_context(params) {
     const int64_t n_embd_head = hparams.n_embd_head_v();
 

src/models/bert.cpp

@@ -1,6 +1,83 @@
 #include "models.h"
 
-llm_build_bert::llm_build_bert(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
+void llama_model_bert::load_arch_hparams(llama_model_loader & ml) {
+    ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS,    hparams.f_norm_eps);
+
+    switch (hparams.n_layer) {
+        case 3:
+            type = LLM_TYPE_17M; break; // bge-micro
+        case 6:
+            type = LLM_TYPE_22M; break; // MiniLM-L6
+        case 12:
+            switch (hparams.n_embd) {
+                case 384: type = LLM_TYPE_33M; break; // MiniLM-L12, bge-small
+                case 768: type = LLM_TYPE_109M; break; // bge-base
+                default: type = LLM_TYPE_UNKNOWN;
+            } break;
+        case 24:
+            type = LLM_TYPE_335M; break; // bge-large
+        default: type = LLM_TYPE_UNKNOWN;
+    }
+}
+
+void llama_model_bert::load_arch_tensors(llama_model_loader &) {
+    LLAMA_LOAD_LOCALS;
+
+    if (n_token_types == 0) {
+        throw std::runtime_error(arch_name() + " model needs to define token type count");
+    }
+    tok_embd     = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD,  "weight"), {n_embd, n_vocab}, 0);
+    type_embd    = create_tensor(tn(LLM_TENSOR_TOKEN_TYPES, "weight"), {n_embd, n_token_types}, TENSOR_NOT_REQUIRED);
+
+    if (arch == LLM_ARCH_BERT) {
+        pos_embd = create_tensor(tn(LLM_TENSOR_POS_EMBD,    "weight"), {n_embd, n_ctx_train}, 0);
+
+        cls   = create_tensor(tn(LLM_TENSOR_CLS, "weight"), {n_embd, n_embd}, TENSOR_NOT_REQUIRED);
+        cls_b = create_tensor(tn(LLM_TENSOR_CLS, "bias"),   {n_embd},         TENSOR_NOT_REQUIRED);
+
+        cls_out   = create_tensor(tn(LLM_TENSOR_CLS_OUT, "weight"), {n_embd, hparams.n_cls_out}, TENSOR_NOT_REQUIRED);
+        cls_out_b = create_tensor(tn(LLM_TENSOR_CLS_OUT, "bias"),   {hparams.n_cls_out},         TENSOR_NOT_REQUIRED);
+    }
+
+    tok_norm   = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD_NORM, "weight", 0), {n_embd}, 0);
+    tok_norm_b = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD_NORM, "bias",   0), {n_embd}, 0);
+
+    for (int i = 0; i < n_layer; ++i) {
+        auto & layer = layers[i];
+
+        create_tensor_qkv(layer, i, n_embd, n_embd, n_embd_gqa, n_embd_gqa, 0);
+
+        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
+        layer.wo_b = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i), {n_embd}, TENSOR_NOT_REQUIRED);
+
+        layer.attn_out_norm   = create_tensor(tn(LLM_TENSOR_ATTN_OUT_NORM, "weight", i), {n_embd}, 0);
+        layer.attn_out_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_OUT_NORM, "bias", i),   {n_embd}, 0);
+
+        if (hparams.moe_every_n_layers > 0 && i % hparams.moe_every_n_layers == 1) {
+            layer.ffn_up_exps   = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS,   "weight", i), {  n_embd, n_ff,   n_expert}, 0);
+            layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), {  n_ff,   n_embd, n_expert}, 0);
+            layer.ffn_gate_inp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP,   "weight", i), {n_embd, n_expert}, 0);
+        } else {
+            layer.ffn_up     = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd, n_ff}, 0);
+            layer.ffn_up_b   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "bias", i),   {n_ff}, TENSOR_NOT_REQUIRED);
+            layer.ffn_down   = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0);
+            layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i),   {n_embd}, TENSOR_NOT_REQUIRED);
+
+            if (arch == LLM_ARCH_NOMIC_BERT) {
+                layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0);
+            }
+        }
+
+        layer.layer_out_norm   = create_tensor(tn(LLM_TENSOR_LAYER_OUT_NORM, "weight", i), {n_embd}, 0);
+        layer.layer_out_norm_b = create_tensor(tn(LLM_TENSOR_LAYER_OUT_NORM, "bias", i),   {n_embd}, 0);
+    }
+}
+
+std::unique_ptr<llm_graph_context> llama_model_bert::build_arch_graph(const llm_graph_params & params) const {
+    return std::make_unique<graph>(*this, params);
+}
+
+llama_model_bert::graph::graph(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
     const int64_t n_embd_head = hparams.n_embd_head_v();
 
     GGML_ASSERT(n_embd_head == hparams.n_embd_head_k());