llama_cpp 0.24.2 → 0.25.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 1130bd5d4bd478e4aed2e67d836fe66aa0bf166dac85e28557e05814b75d48b2
-  data.tar.gz: fb276cff62ba89f3726b526c7efea7d6b76ff4164b3885cd70c07c36ae2a4ec7
+  metadata.gz: 953b205d7cedadd2f1db35fc301c6b94b1db87e0121317f6c154c204e09e9d56
+  data.tar.gz: bbbc1eef7f7312e667fd238b1e5ef353861beaa445c348595470879eccd12280
 SHA512:
-  metadata.gz: f45b85cc4dfebd8a0afb0592a1ddee159656a749033cc5abf395f88c19742a705263018b72acb234142d2821b3b5ba2e1a09ff1884347b0a73bea5b1a6b0c3bc
-  data.tar.gz: 9b000d84f97eaa7e4f6b775ce03899a60aae832440fb53b28115d576441df330d42bd3761984fe2765c3b9fed60121caeaed7c697288140a58fb193467e9d082
+  metadata.gz: a38097de3f8e5a8acf862bf28db94140efc8f5e31d2be8e9fcbbf461ad79c77614a65acb873d25a706380056295b09cc14f94d90661e4c05e81a7c84a7cc461d
+  data.tar.gz: 2225bfdc2526274a6c3b60ec7c6ad0d88c62a291c847d511dd43772b7cc4c978cecbf69dd0fa50d10fafc3a915b95c44d8eebfdb14815777ace34318f1b28066

data/CHANGELOG.md

@@ -1,3 +1,19 @@
+## [[0.25.0](https://github.com/yoshoku/llama_cpp.rb/compare/v0.24.3...v0.25.0)] - 2026-04-25
+
+- Change supported llama.cpp version to b8920.
+  - Remove `LLAMA_PARAMS_FIT_STATUS_SUCCESS` constant value.
+  - Remove `LLAMA_PARAMS_FIT_STATUS_FAILURE` constant value.
+  - Remove `LLAMA_PARAMS_FIT_STATUS_ERROR` constant value.
+  - Remove `llama_memory_breakdown_print` module function.
+  - Add `LLAMA_FTYPE_MOSTLY_Q1_0` constant value.
+  - Add `LLAMA_SPLIT_MODE_TENSOR` constant value.
+
+## [[0.24.2](https://github.com/yoshoku/llama_cpp.rb/compare/v0.24.2...v0.24.3)] - 2026-04-06
+
+- Change supported llama.cpp version to b8640.
+  - Add `LlamaModelImatrixData` class to `LlamaCpp`.
+  - Add `LlamaModelTensorOverride` class to `LlamaCpp`.
+
 ## [[0.24.2](https://github.com/yoshoku/llama_cpp.rb/compare/v0.24.1...v0.24.2)] - 2026-03-15
 
 - Change supported llama.cpp version to b8340.

data/ext/llama_cpp/llama_cpp.c

@@ -7,6 +7,8 @@ VALUE rb_cLlamaContext;
 VALUE rb_cLlamaModelTensorBuftOverride;
 VALUE rb_cLlamaModelParams;
 VALUE rb_cLlamaContextParams;
+VALUE rb_cLlamaModelTensorOverride;
+VALUE rb_cLlamaModelImatrixData;
 VALUE rb_cLlamaModelQuantizeParams;
 VALUE rb_cLlamaLogitBias;
 VALUE rb_cLlamaAdapterLora;
@@ -892,6 +894,100 @@ static VALUE llama_context_params_set_kv_unified(VALUE self, VALUE kv_unified) {
   return kv_unified;
 }
 
+/* struct llama_model_tensor_override */
+static void llama_model_tensor_override_free(void *ptr) {
+  if (ptr) {
+    ruby_xfree(ptr);
+  }
+}
+
+static size_t llama_model_tensor_override_size(const void *ptr) {
+  return sizeof(*((struct llama_model_tensor_override*)ptr));
+}
+
+static rb_data_type_t llama_model_tensor_override_type = {
+  "LlamaModelTensorOverride",
+  { NULL,
+    llama_model_tensor_override_free,
+    llama_model_tensor_override_size },
+  NULL,
+  NULL,
+  RUBY_TYPED_FREE_IMMEDIATELY
+};
+
+static VALUE llama_model_tensor_override_alloc(VALUE self) {
+  struct llama_model_tensor_override* data = (struct llama_model_tensor_override*)ruby_xmalloc(sizeof(struct llama_model_tensor_override));
+  return TypedData_Wrap_Struct(self, &llama_model_tensor_override_type, data);
+}
+
+static struct llama_model_tensor_override* get_llama_model_tensor_override(VALUE self) {
+  struct llama_model_tensor_override* data = NULL;
+  TypedData_Get_Struct(self, struct llama_model_tensor_override, &llama_model_tensor_override_type, data);
+  return data;
+}
+
+static VALUE llama_model_tensor_override_get_pattern(VALUE self) {
+  struct llama_model_tensor_override* data = get_llama_model_tensor_override(self);
+  const char* pattern = data->pattern;
+  return rb_utf8_str_new_cstr(pattern);
+}
+
+static VALUE llama_model_tensor_override_get_type(VALUE self) {
+  struct llama_model_tensor_override* data = get_llama_model_tensor_override(self);
+  return INT2NUM(data->type);
+}
+
+/* struct llama_model_imatrix_data */
+static void llama_model_imatrix_data_free(void *ptr) {
+  if (ptr) {
+    ruby_xfree(ptr);
+  }
+}
+
+static size_t llama_model_imatrix_data_size(const void *ptr) {
+  return sizeof(*((struct llama_model_imatrix_data*)ptr));
+}
+
+static rb_data_type_t llama_model_imatrix_data_type = {
+  "LlamaModelImatrixData",
+  { NULL,
+    llama_model_imatrix_data_free,
+    llama_model_imatrix_data_size },
+  NULL,
+  NULL,
+  RUBY_TYPED_FREE_IMMEDIATELY
+};
+
+static VALUE llama_model_imatrix_data_alloc(VALUE self) {
+  struct llama_model_imatrix_data* data = (struct llama_model_imatrix_data*)ruby_xmalloc(sizeof(struct llama_model_imatrix_data));
+  return TypedData_Wrap_Struct(self, &llama_model_imatrix_data_type, data);
+}
+
+static struct llama_model_imatrix_data* get_llama_model_imatrix_data(VALUE self) {
+  struct llama_model_imatrix_data* data = NULL;
+  TypedData_Get_Struct(self, struct llama_model_imatrix_data, &llama_model_imatrix_data_type, data);
+  return data;
+}
+
+static VALUE llama_model_imatrix_data_get_name(VALUE self) {
+  struct llama_model_imatrix_data* data = get_llama_model_imatrix_data(self);
+  return rb_utf8_str_new_cstr(data->name);
+}
+
+static VALUE llama_model_imatrix_data_get_size(VALUE self) {
+  struct llama_model_imatrix_data* data = get_llama_model_imatrix_data(self);
+  return SIZET2NUM(data->size);
+}
+
+static VALUE llama_model_imatrix_data_get_data(VALUE self) {
+  struct llama_model_imatrix_data* data = get_llama_model_imatrix_data(self);
+  VALUE ary = rb_ary_new2(data->size);
+  for (size_t i = 0; i < data->size; i++) {
+    rb_ary_store(ary, i, DBL2NUM(data->data[i]));
+  }
+  return ary;
+}
+
 /* llama_model_quantize_params */
 static void llama_model_quantize_params_free(void *ptr) {
   if (ptr) {
@@ -3279,22 +3375,6 @@ static VALUE rb_llama_sampler_reset(VALUE self, VALUE sampler) {
   return Qnil;
 }
 
-/**
- * @overload llama_memory_breakdown_print(context)
- *  @param [LlamaContext] context
- *  @return [NilClass]
- */
-static VALUE rb_llama_memory_breakdown_print(VALUE self, VALUE ctx) {
-  if (!rb_obj_is_kind_of(ctx, rb_cLlamaContext)) {
-    rb_raise(rb_eArgError, "ctx must be a LlamaContext");
-    return Qnil;
-  }
-  llama_context_wrapper* context_wrapper = get_llama_context_wrapper(ctx);
-  llama_memory_breakdown_print(context_wrapper->context);
-  RB_GC_GUARD(ctx);
-  return Qnil;
-}
-
 /**
  * @overload llama_sampler_clone(sampler)
  *  @param [LlamaSampler] sampler
@@ -4196,6 +4276,7 @@ void Init_llama_cpp(void) {
   rb_define_const(rb_mLlamaCpp, "LLAMA_FTYPE_MOSTLY_TQ2_0", INT2NUM(LLAMA_FTYPE_MOSTLY_TQ2_0));
   rb_define_const(rb_mLlamaCpp, "LLAMA_FTYPE_MOSTLY_MXFP4_MOE", INT2NUM(LLAMA_FTYPE_MOSTLY_MXFP4_MOE));
   rb_define_const(rb_mLlamaCpp, "LLAMA_FTYPE_MOSTLY_NVFP4", INT2NUM(LLAMA_FTYPE_MOSTLY_NVFP4));
+  rb_define_const(rb_mLlamaCpp, "LLAMA_FTYPE_MOSTLY_Q1_0", INT2NUM(LLAMA_FTYPE_MOSTLY_Q1_0));
   rb_define_const(rb_mLlamaCpp, "LLAMA_FTYPE_GUESSED", INT2NUM(LLAMA_FTYPE_GUESSED));
   /* llama_rope_scaling_type */
   /* Document-const: LlamaCpp::LLAMA_ROPE_SCALING_TYPE_UNSPECIFIED */
@@ -4228,6 +4309,7 @@ void Init_llama_cpp(void) {
   rb_define_const(rb_mLlamaCpp, "LLAMA_SPLIT_MODE_NONE", INT2NUM(LLAMA_SPLIT_MODE_NONE));
   rb_define_const(rb_mLlamaCpp, "LLAMA_SPLIT_MODE_LAYER", INT2NUM(LLAMA_SPLIT_MODE_LAYER));
   rb_define_const(rb_mLlamaCpp, "LLAMA_SPLIT_MODE_ROW", INT2NUM(LLAMA_SPLIT_MODE_ROW));
+  rb_define_const(rb_mLlamaCpp, "LLAMA_SPLIT_MODE_TENSOR", INT2NUM(LLAMA_SPLIT_MODE_TENSOR));
 
   rb_define_module_function(rb_mLlamaCpp, "llama_flash_attn_type_name", rb_llama_flash_attn_type_name, 1);
 
@@ -4811,6 +4893,45 @@ void Init_llama_cpp(void) {
   /* TODO: ggml_abort_callback abort_callback */
   /* TODO: void* abort_callback_data */
 
+  /**
+   * Document-class: LlamaCpp::LlamaModelTensorOverride
+   * "struct llama_model_tensor_override" wrapper class
+   */
+  rb_cLlamaModelTensorOverride = rb_define_class_under(rb_mLlamaCpp, "LlamaModelTensorOverride", rb_cObject);
+  rb_define_alloc_func(rb_cLlamaModelTensorOverride, llama_model_tensor_override_alloc);
+  /**
+   * Document-method: pattern
+   * @return [String]
+   */
+  rb_define_method(rb_cLlamaModelTensorOverride, "pattern", RUBY_METHOD_FUNC(llama_model_tensor_override_get_pattern), 0);
+  /**
+   * Document-method: type
+   * @return [Integer]
+   */
+  rb_define_method(rb_cLlamaModelTensorOverride, "type", RUBY_METHOD_FUNC(llama_model_tensor_override_get_type), 0);
+
+  /**
+   * Document-class: LlamaCpp::LlamaModelImatrixData
+   * "struct llama_model_i_matrix_data" wrapper class
+   */
+  rb_cLlamaModelImatrixData = rb_define_class_under(rb_mLlamaCpp, "LlamaModelImatrixData", rb_cObject);
+  rb_define_alloc_func(rb_cLlamaModelImatrixData, llama_model_imatrix_data_alloc);
+  /**
+   * Document-method: name
+   * @return [String]
+   */
+  rb_define_method(rb_cLlamaModelImatrixData, "name", RUBY_METHOD_FUNC(llama_model_imatrix_data_get_name), 0);
+  /**
+   * Document-method: size
+   * @return [Integer]
+   */
+  rb_define_method(rb_cLlamaModelImatrixData, "size", RUBY_METHOD_FUNC(llama_model_imatrix_data_get_size), 0);
+  /**
+   * Document-method: data
+   * @return [Array<Float>]
+   */
+  rb_define_method(rb_cLlamaModelImatrixData, "data", RUBY_METHOD_FUNC(llama_model_imatrix_data_get_data), 0);
+
   /**
    * Document-class: LlamaCpp::LlamaModelQuantizeParams
    * "struct llama_model_quantize_params" wrapper class
@@ -4927,10 +5048,10 @@ void Init_llama_cpp(void) {
    * @return [Boolean]
    */
   rb_define_method(rb_cLlamaModelQuantizeParams, "dry_run=", RUBY_METHOD_FUNC(llama_model_quantize_params_set_dry_run), 1);
-  /* TODO: void* imatrix */
-  /* TODO: void* kv_overrides */
-  /* TODO: void* tensor_types */
-  /* TODO: void* prune_layers */
+  /* TODO: const struct llama_model_imatrix_data* imatrix */
+  /* TODO: const struct llama_model_kv_override* kv_overrides */
+  /* TODO: const struct llama_model_tensor_override* tt_overrides */
+  /* TODO: const int32_t* prune_layers */
 
   /**
    * Document-class: LlamaCpp::LlamaLogitBias
@@ -5022,12 +5143,6 @@ void Init_llama_cpp(void) {
   /* llama_free */
   rb_define_module_function(rb_mLlamaCpp, "llama_free", rb_llama_free, 1);
 
-  /* llama_params_fit_status */
-  /* Document-const: LlamaCpp::LLAMA_PARAMS_FIT_STATUS_SUCCESS */
-  rb_define_const(rb_mLlamaCpp, "LLAMA_PARAMS_FIT_STATUS_SUCCESS", INT2NUM(LLAMA_PARAMS_FIT_STATUS_SUCCESS));
-  rb_define_const(rb_mLlamaCpp, "LLAMA_PARAMS_FIT_STATUS_FAILURE", INT2NUM(LLAMA_PARAMS_FIT_STATUS_FAILURE));
-  rb_define_const(rb_mLlamaCpp, "LLAMA_PARAMS_FIT_STATUS_ERROR", INT2NUM(LLAMA_PARAMS_FIT_STATUS_ERROR));
-
   /* TODO: llama_params_fit */
 
   /* llama_time_us */
@@ -5564,9 +5679,6 @@ void Init_llama_cpp(void) {
   /* llama_perf_sampler_reset */
   rb_define_module_function(rb_mLlamaCpp, "llama_perf_sampler_reset", rb_llama_perf_sampler_reset, 1);
 
-  /* llama_memory_breakdown_print */
-  rb_define_module_function(rb_mLlamaCpp, "llama_memory_breakdown_print", rb_llama_memory_breakdown_print, 1);
-
   /* TODO: typedef bool (*llama_opt_param_filter) */
   /* TODO: bool llama_opt_param_filter_all */
   /* TODO: struct llama_opt_params */

data/lib/llama_cpp/version.rb

@@ -3,8 +3,8 @@
 # llama_cpp.rb provides Ruby bindings for the llama.cpp.
 module LlamaCpp
   # The version of llama_cpp.rb you install.
-  VERSION = '0.24.2'
+  VERSION = '0.25.0'
 
   # The supported version of llama.cpp.
-  LLAMA_CPP_VERSION = 'b8340'
+  LLAMA_CPP_VERSION = 'b8920'
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: llama_cpp
 version: !ruby/object:Gem::Version
-  version: 0.24.2
+  version: 0.25.0
 platform: ruby
 authors:
 - yoshoku
@@ -33,7 +33,7 @@ metadata:
   homepage_uri: https://github.com/yoshoku/llama_cpp.rb
   source_code_uri: https://github.com/yoshoku/llama_cpp.rb
   changelog_uri: https://github.com/yoshoku/llama_cpp.rb/blob/main/CHANGELOG.md
-  documentation_uri: https://gemdocs.org/gems/llama_cpp/0.24.2/
+  documentation_uri: https://gemdocs.org/gems/llama_cpp/0.25.0/
   rubygems_mfa_required: 'true'
 rdoc_options: []
 require_paths:
@@ -49,7 +49,7 @@ required_rubygems_version: !ruby/object:Gem::Requirement
     - !ruby/object:Gem::Version
       version: '0'
 requirements: []
-rubygems_version: 4.0.3
+rubygems_version: 4.0.6
 specification_version: 4
 summary: Ruby bindings for the llama.cpp.
 test_files: []