llama_cpp 0.19.0 → 0.19.2

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 60a718430e278569f1e5eeb08cfa9f56b1f7b1f804234633dbfba9a504959ae9
-  data.tar.gz: f33f971717366760f84c40ca003015f3fadd18cde969aa511031ffcd4517de32
+  metadata.gz: b9d8c3d491c1da2a69c57d881b9a712078868e9d11f9fdbda3cc9510a4d19c77
+  data.tar.gz: 4e4db26e6f428f9d9eac6ba2097a4453baa61515eff0249a12a1f57b51d89eb3
 SHA512:
-  metadata.gz: c85e04506be2a1bf0d4c3d99aa5b9fcf9f2ca6f4eb8673558a8d8755a056cff700769f16e429a048c2c3abdbac5fcb02818f92ae78966fdbe2b8f813638948a3
-  data.tar.gz: a9ecfe4fa2b2314d57390e87bbed7734e9499d0ceea80c9e444e1877617f55f0da6fa06f77afddd0107723315bb43b0f66e85557612816cba66e29735f9da261
+  metadata.gz: dfa8562de2134396f10e8b085c4935a58bdfbc37948d29e668592788739fcdbef565bc14faed2e3d7580df2d99592b20c36964fad0b66c33e707efab13477d6e
+  data.tar.gz: ee7d052d9daa1acaf3ae7ecc9b9e87df30d4743781ca145553c32d62b6e7b849b85d50644da8b8e10829b9396f4389989f9252319f8abb272ed102170bbdf39e

data/CHANGELOG.md

@@ -1,4 +1,17 @@
 
+## [[0.19.2](https://github.com/yoshoku/llama_cpp.rb/compare/v0.19.1...v0.19.2)] - 2025-04-05
+
+- Change supported llama.cpp version to b5050
+  - Add `LLAMA_VOCAB_PRE_TYPE_TRILLION` constant.
+  - Add `LLAMA_VOCAB_PRE_TYPE_BAILINGMOE` constant.
+  - Add `LlamaModelTensorBuftOverride` class.
+
+## [[0.19.1](https://github.com/yoshoku/llama_cpp.rb/compare/v0.19.0...v0.19.1)] - 2025-03-29
+
+- Change supported llama.cpp version to b4980
+  - Add `LLAMA_VOCAB_PRE_TYPE_SUPERBPE` constant.
+  - Add `llama_set_warmup` module function.
+
 ## [[0.19.0](https://github.com/yoshoku/llama_cpp.rb/compare/v0.18.2...v0.19.0)] - 2025-03-16
 
 **Breaking Changes**

data/ext/llama_cpp/llama_cpp.c

@@ -4,6 +4,7 @@ VALUE rb_mLlamaCpp;
 VALUE rb_cLlamaVocab;
 VALUE rb_cLlamaModel;
 VALUE rb_cLlamaContext;
+VALUE rb_cLlamaModelTensorBuftOverride;
 VALUE rb_cLlamaModelParams;
 VALUE rb_cLlamaContextParams;
 VALUE rb_cLlamaModelQuantizeParams;
@@ -361,6 +362,49 @@ static VALUE llama_model_kv_override_get_val_str(VALUE self) {
   return rb_utf8_str_new_cstr(data->val_str);
 }
 
+/* struct llama_model_tensor_buft_override */
+static void llama_model_tensor_buft_override_free(void *ptr) {
+  if (ptr) {
+    ruby_xfree(ptr);
+  }
+}
+
+static size_t llama_model_tensor_buft_override_size(const void *ptr) {
+  return sizeof(*((struct llama_model_tensor_buft_override*)ptr));
+}
+
+static rb_data_type_t llama_model_tensor_buft_override_type = {
+  "LlamaModelTensorBuftOverride",
+  { NULL,
+    llama_model_tensor_buft_override_free,
+    llama_model_tensor_buft_override_size },
+  NULL,
+  NULL,
+  RUBY_TYPED_FREE_IMMEDIATELY
+};
+
+static VALUE llama_model_tensor_buft_override_alloc(VALUE self) {
+  struct llama_model_tensor_buft_override* data = (struct llama_model_tensor_buft_override*)ruby_xmalloc(sizeof(struct llama_model_tensor_buft_override));
+  return TypedData_Wrap_Struct(self, &llama_model_tensor_buft_override_type, data);
+}
+
+static struct llama_model_tensor_buft_override* get_llama_model_tensor_buft_override(VALUE self) {
+  struct llama_model_tensor_buft_override* data = NULL;
+  TypedData_Get_Struct(self, struct llama_model_tensor_buft_override, &llama_model_tensor_buft_override_type, data);
+  return data;
+}
+
+static VALUE llama_model_tensor_buft_override_get_pattern(VALUE self) {
+  struct llama_model_tensor_buft_override* data = get_llama_model_tensor_buft_override(self);
+  const char* pattern = data->pattern;
+  return rb_utf8_str_new_cstr(pattern);
+}
+
+// static VALUE llama_model_tensor_buft_override_get_buft(VALUE self) {
+//   struct llama_model_tensor_buft_override* data = get_llama_model_tensor_buft_override(self);
+//   return TypedData_Wrap_Struct(rb_cGgmlBackendBufferTypeT, &llama_model_wrapper_data_type, data->buft);
+// }
+
 /* struct llama_model_params */
 static void llama_model_params_free(void *ptr) {
   if (ptr) {
@@ -2518,6 +2562,24 @@ static VALUE rb_llama_set_causal_attn(VALUE self, VALUE ctx, VALUE causal_attn)
   return Qnil;
 }
 
+/**
+ * @overload llama_set_warmup(context, warmup)
+ *  @param [LlamaContext] context
+ *  @param [Boolean] warmup
+ *  @return [NilClass]
+ */
+static VALUE rb_llama_set_warmup(VALUE self, VALUE ctx, VALUE warmup) {
+  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);
+  const bool warmup_ = RTEST(warmup) ? true : false;
+  llama_set_warmup(context_wrapper->context, warmup_);
+  RB_GC_GUARD(ctx);
+  return Qnil;
+}
+
 /**
  * @overload llama_synchronize(context)
  *  @param [LlamaContext] context
@@ -3965,6 +4027,9 @@ void Init_llama_cpp(void) {
   rb_define_const(rb_mLlamaCpp, "LLAMA_VOCAB_PRE_TYPE_MINERVA", INT2NUM(LLAMA_VOCAB_PRE_TYPE_MINERVA));
   rb_define_const(rb_mLlamaCpp, "LLAMA_VOCAB_PRE_TYPE_DEEPSEEK3_LLM", INT2NUM(LLAMA_VOCAB_PRE_TYPE_DEEPSEEK3_LLM));
   rb_define_const(rb_mLlamaCpp, "LLAMA_VOCAB_PRE_TYPE_GPT4O", INT2NUM(LLAMA_VOCAB_PRE_TYPE_GPT4O));
+  rb_define_const(rb_mLlamaCpp, "LLAMA_VOCAB_PRE_TYPE_SUPERBPE", INT2NUM(LLAMA_VOCAB_PRE_TYPE_SUPERBPE));
+  rb_define_const(rb_mLlamaCpp, "LLAMA_VOCAB_PRE_TYPE_TRILLION", INT2NUM(LLAMA_VOCAB_PRE_TYPE_TRILLION));
+  rb_define_const(rb_mLlamaCpp, "LLAMA_VOCAB_PRE_TYPE_BAILINGMOE", INT2NUM(LLAMA_VOCAB_PRE_TYPE_BAILINGMOE));
   /* llama_rope_type */
   /* Document-const: LlamaCpp::LLAMA_ROPE_TYPE_NONE */
   rb_define_const(rb_mLlamaCpp, "LLAMA_ROPE_TYPE_NONE", INT2NUM(LLAMA_ROPE_TYPE_NONE));
@@ -4167,6 +4232,18 @@ void Init_llama_cpp(void) {
    */
   rb_define_method(rb_cLlamaModelKvOverride, "val_str", RUBY_METHOD_FUNC(llama_model_kv_override_get_val_str), 0);
 
+  /**
+   * Document-class: LlamaCpp::LlamaModelTensorBuftOverride
+   * "struct llama_model_tensor_buf_override" wrapper class
+   */
+  rb_cLlamaModelTensorBuftOverride = rb_define_class_under(rb_mLlamaCpp, "LlamaModelTensorBuftOverride", rb_cObject);
+  rb_define_alloc_func(rb_cLlamaModelTensorBuftOverride, llama_model_tensor_buft_override_alloc);
+  /**
+   * Document-method: pattern
+   * @return [String]
+   */
+  rb_define_method(rb_cLlamaModelTensorBuftOverride, "pattern", RUBY_METHOD_FUNC(llama_model_tensor_buft_override_get_pattern), 0);
+  /* TODO: ggml_backend_buffer_type_t buff */
 
   /**
    * Document-class: LlamaCpp::LlamaModelParams
@@ -4175,6 +4252,7 @@ void Init_llama_cpp(void) {
   rb_cLlamaModelParams = rb_define_class_under(rb_mLlamaCpp, "LlamaModelParams", rb_cObject);
   rb_define_alloc_func(rb_cLlamaModelParams, llama_model_params_alloc);
   /* TODO: ggml_backend_dev_t* devices */
+  /* TODO: const struct llama_model_tensor_buft_override * tensor_buft_overrides */
   /**
    * Document-method: n_gpu_layers
    * @return [Integer]
@@ -4995,6 +5073,9 @@ void Init_llama_cpp(void) {
   /* llama_set_causal_attn */
   rb_define_module_function(rb_mLlamaCpp, "llama_set_causal_attn", rb_llama_set_causal_attn, 2);
 
+  /* llama_set_warmup */
+  rb_define_module_function(rb_mLlamaCpp, "llama_set_warmup", rb_llama_set_warmup, 2);
+
   /* llama_set_abort_callback */
 
   /* llama_synchronize */

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 = '0.19.0'
+  VERSION = '0.19.2'
 
   # The supported version of llama.cpp.
-  LLAMA_CPP_VERSION = 'b4885'
+  LLAMA_CPP_VERSION = 'b5050'
 end

metadata

@@ -1,13 +1,13 @@
 --- !ruby/object:Gem::Specification
 name: llama_cpp
 version: !ruby/object:Gem::Version
-  version: 0.19.0
+  version: 0.19.2
 platform: ruby
 authors:
 - yoshoku
 bindir: exe
 cert_chain: []
-date: 2025-03-16 00:00:00.000000000 Z
+date: 2025-04-05 00:00:00.000000000 Z
 dependencies: []
 description: llama_cpp.rb provides Ruby bindings for the llama.cpp.
 email:
@@ -49,7 +49,7 @@ required_rubygems_version: !ruby/object:Gem::Requirement
     - !ruby/object:Gem::Version
       version: '0'
 requirements: []
-rubygems_version: 3.6.2
+rubygems_version: 3.6.6
 specification_version: 4
 summary: Ruby bindings for the llama.cpp.
 test_files: []