llama_cpp 0.19.1 → 0.19.3

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: bb93f6aa74a3f022cb15dde5df9b4e6deb4c228a5b0fbfc01ff2ebed9446f152
-  data.tar.gz: 0b94a7d328ed8ac1bdc637a9b79ff490f5e5af1df3c5893e88c91dcf4a73ed07
+  metadata.gz: 6f3cfec863591d3bd24156c27612d65dc565194c9272e1f69bc79ad77fe1b2a0
+  data.tar.gz: 29426a27e23f21ec63ebc4db0419bf9fe96e5e063ad0e42285b361e4f9ab5d50
 SHA512:
-  metadata.gz: 97f0be1659fabb158b328294b3d7b5df9ea7774f8d0063246ea5d006f39f6d9799d80236a712d6026b6d9df4acd5b20ced809d93145379dadc13a61e0813a689
-  data.tar.gz: 5072945b4b0152fe5dfa41807253a511274020fb5149988586e1932c65f0ce020876f200cf97b291ba2d804cac03aea731c5d57adb6ab1ef8d448a210104836b
+  metadata.gz: d52e67395a3cbde1108968837166a295f3a9a6d033ef230e1c9223f0f1dd92db4733b4b8bae37ca8f9ee648591dc082f20ceae234e86271cecb5f29e34507e14
+  data.tar.gz: aed2fd04608461b74c26d44c1f26591aada8699e10d4e23366cca28b798be227013f47aa2a2dcff3ccf2900f5f04fb7ed25ae6853f921d26b17b938713e3afd1

data/CHANGELOG.md

@@ -1,5 +1,16 @@
+## [[0.19.2](https://github.com/yoshoku/llama_cpp.rb/compare/v0.19.2...v0.19.3)] - 2025-04-12
 
-## [[0.19.0](https://github.com/yoshoku/llama_cpp.rb/compare/v0.19.0...v0.19.1)] - 2025-03-29
+- Change supported llama.cpp version to b5120
+  - Add `LLAMA_VOCAB_PRE_TYPE_LLAMA4` constant.
+
+## [[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.

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) {
@@ -3984,6 +4028,9 @@ void Init_llama_cpp(void) {
   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));
+  rb_define_const(rb_mLlamaCpp, "LLAMA_VOCAB_PRE_TYPE_LLAMA4", INT2NUM(LLAMA_VOCAB_PRE_TYPE_LLAMA4));
   /* llama_rope_type */
   /* Document-const: LlamaCpp::LLAMA_ROPE_TYPE_NONE */
   rb_define_const(rb_mLlamaCpp, "LLAMA_ROPE_TYPE_NONE", INT2NUM(LLAMA_ROPE_TYPE_NONE));
@@ -4186,6 +4233,18 @@ 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
@@ -4194,6 +4253,7 @@ 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]

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 = '0.19.1'
+  VERSION = '0.19.3'
 
   # The supported version of llama.cpp.
-  LLAMA_CPP_VERSION = 'b4980'
+  LLAMA_CPP_VERSION = 'b5120'
 end

metadata

@@ -1,13 +1,13 @@
 --- !ruby/object:Gem::Specification
 name: llama_cpp
 version: !ruby/object:Gem::Version
-  version: 0.19.1
+  version: 0.19.3
 platform: ruby
 authors:
 - yoshoku
 bindir: exe
 cert_chain: []
-date: 2025-03-29 00:00:00.000000000 Z
+date: 2025-04-12 00:00:00.000000000 Z
 dependencies: []
 description: llama_cpp.rb provides Ruby bindings for the llama.cpp.
 email: