llama_cpp 0.21.1 → 0.21.2

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 64aab047f44a9002c5c388f6a774448671d7be9618170310d2b7dd1091a99670
-  data.tar.gz: 63f0d908b99a45865a9b9d81ae595adb0d9a1f258976267da3ce5b5df747ae0d
+  metadata.gz: 5f3d80e415240f21df0ba7e314118712f6f2d2d8cb052eef41127e7de7ae7e51
+  data.tar.gz: 67fff55bf83ef4d97f24b0c3c450eb493464763c23595fcb5b60c248fd1f551e
 SHA512:
-  metadata.gz: 4f5344b36ff0c0bbd812f25f2b7a897893cb8ffb0700e3f8c23dfbd81cc8b1c4ee1893d5198c74df625fbc3c4c925f32e4c868c60b6ed13e7a5412eef3d9b2b6
-  data.tar.gz: dbc8d279a9284ba62786d3c5d267352445814289cb1f8a2268d2bb2ff72a4bc7a2534cb2c4582f945e13a68a645cec51101085203ff6dac1a2b7df6885412398
+  metadata.gz: f6a129952bc812c130a235743beeb0a5ebcafcbe627d7d13d625e8cf5b143e2cb41cd4d78cfacc98c00a1d0857dc68bb8a6b30d486da5f22e8bcbc6a63ac5489
+  data.tar.gz: c944f2a37727bd3e295dc83d9d9c5c5748a04d7f867aad86895c312a100fa65b7c3fa9f3973bf08adc47c26c2a0e73586f7aa92b92230bee72c86b10dc4693fc

data/CHANGELOG.md

@@ -1,10 +1,17 @@
+## [[0.21.2](https://github.com/yoshoku/llama_cpp.rb/compare/v0.21.1...v0.21.2)] - 2025-08-09
+
+- Change supported llama.cpp version to b6100.
+  - Add `LLAMA_FTYPE_MOSTLY_MXFP4_MOE` constant.
+  - Add `use_extra_bufts` accessor to `LlamaModelParams`.
+  - Add `llama_model_is_diffusion?` module function.
+
 ## [[0.21.1](https://github.com/yoshoku/llama_cpp.rb/compare/v0.21.0...v0.21.1)] - 2025-07-19
 
 - Change supported llama.cpp version to b5930.
   - Add `n_reused` reader to `LlamaPerfContextData`.
   - Add `llama_vocab_mask` module function.
   - Add `kv_unified` accessor to `LlamaContextParams`.
-  - Add `LLAMA_VOCAB_TYPE_PLAMO2` contant.
+  - Add `LLAMA_VOCAB_TYPE_PLAMO2` constant.
 
 ## [[0.21.0](https://github.com/yoshoku/llama_cpp.rb/compare/v0.20.4...v0.21.0)] - 2025-07-12
 

data/ext/llama_cpp/llama_cpp.c

@@ -530,6 +530,17 @@ static VALUE llama_model_params_set_check_tensors(VALUE self, VALUE check_tensor
   return check_tensors;
 }
 
+static VALUE llama_model_params_get_use_extra_bufts(VALUE self) {
+  struct llama_model_params* data = get_llama_model_params(self);
+  return data->use_extra_bufts ? Qtrue : Qfalse;
+}
+
+static VALUE llama_model_params_set_use_extra_bufts(VALUE self, VALUE use_extra_bufts) {
+  struct llama_model_params* data = get_llama_model_params(self);
+  data->use_extra_bufts = RTEST(use_extra_bufts) ? true : false;
+  return use_extra_bufts;
+}
+
 /* struct llama_context_params */
 static void llama_context_params_free(void *ptr) {
   if (ptr) {
@@ -1774,6 +1785,20 @@ static VALUE rb_llama_model_is_recurrent(VALUE self, VALUE model) {
   return llama_model_is_recurrent(model_wrapper->model) ? Qtrue : Qfalse;
 }
 
+/**
+ * @overload llama_model_is_diffusion?(model)
+ *  @param [LlamaModel] model
+ *  @return [Boolean]
+ */
+static VALUE rb_llama_model_is_diffusion(VALUE self, VALUE model) {
+  if (!rb_obj_is_kind_of(model, rb_cLlamaModel)) {
+    rb_raise(rb_eArgError, "model must be a LlamaModel");
+    return Qnil;
+  }
+  llama_model_wrapper* model_wrapper = get_llama_model_wrapper(model);
+  return llama_model_is_diffusion(model_wrapper->model) ? Qtrue : Qfalse;
+}
+
 /**
  * @overload llama_model_quantize(fname_inp, fname_out, params)
  *  @param [String] fname_inp
@@ -4251,6 +4276,7 @@ void Init_llama_cpp(void) {
   rb_define_const(rb_mLlamaCpp, "LLAMA_FTYPE_MOSTLY_BF16", INT2NUM(LLAMA_FTYPE_MOSTLY_BF16));
   rb_define_const(rb_mLlamaCpp, "LLAMA_FTYPE_MOSTLY_TQ1_0", INT2NUM(LLAMA_FTYPE_MOSTLY_TQ1_0));
   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_GUESSED", INT2NUM(LLAMA_FTYPE_GUESSED));
   /* llama_rope_scaling_type */
   /* Document-const: LlamaCpp::LLAMA_ROPE_SCALING_TYPE_UNSPECIFIED */
@@ -4496,6 +4522,17 @@ void Init_llama_cpp(void) {
    * @return [Boolean]
    */
   rb_define_method(rb_cLlamaModelParams, "check_tensors=", RUBY_METHOD_FUNC(llama_model_params_set_check_tensors), 1);
+  /**
+   * Document-method: use_extra_bufts
+   * @return [Boolean]
+   */
+  rb_define_method(rb_cLlamaModelParams, "use_extra_bufts", RUBY_METHOD_FUNC(llama_model_params_get_use_extra_bufts), 0);
+  /**
+   * Document-method: use_extra_bufts=
+   * @param [Boolean] use_extra_bufts
+   * @return [Boolean]
+   */
+  rb_define_method(rb_cLlamaModelParams, "use_extra_bufts=", RUBY_METHOD_FUNC(llama_model_params_set_use_extra_bufts), 1);
 
   /**
    * Document-class: LlamaCpp::LlamaContextParams
@@ -5106,6 +5143,9 @@ void Init_llama_cpp(void) {
   /* llama_model_is_recurrent */
   rb_define_module_function(rb_mLlamaCpp, "llama_model_is_recurrent?", rb_llama_model_is_recurrent, 1);
 
+  /* llama_model_is_diffusion */
+  rb_define_module_function(rb_mLlamaCpp, "llama_model_is_diffusion?", rb_llama_model_is_diffusion, 1);
+
   /* llama_model_quantize */
   rb_define_module_function(rb_mLlamaCpp, "llama_model_quantize", rb_llama_model_quantize, 3);
 

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 = '0.21.1'
+  VERSION = '0.21.2'
 
   # The supported version of llama.cpp.
-  LLAMA_CPP_VERSION = 'b5930'
+  LLAMA_CPP_VERSION = 'b6100'
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: llama_cpp
 version: !ruby/object:Gem::Version
-  version: 0.21.1
+  version: 0.21.2
 platform: ruby
 authors:
 - yoshoku
@@ -49,7 +49,7 @@ required_rubygems_version: !ruby/object:Gem::Requirement
     - !ruby/object:Gem::Version
       version: '0'
 requirements: []
-rubygems_version: 3.6.9
+rubygems_version: 3.7.0
 specification_version: 4
 summary: Ruby bindings for the llama.cpp.
 test_files: []