llama_cpp 0.23.11 → 0.24.2

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 5fc81dd7e098dace7f301394170fa517ec4d214bc41d76dbbf4b79c162ebff85
-  data.tar.gz: 8641cedea81065a2d7ced2e8db028bf8209de42312e5b3431749c19f605d5134
+  metadata.gz: 1130bd5d4bd478e4aed2e67d836fe66aa0bf166dac85e28557e05814b75d48b2
+  data.tar.gz: fb276cff62ba89f3726b526c7efea7d6b76ff4164b3885cd70c07c36ae2a4ec7
 SHA512:
-  metadata.gz: 3cb0176fc18bb430ee7d00177a911ebe204f975ef7d2db88a79c696a1e4b3fde2ae74cc9fa34648294a552b4f026238b00572e35c9bb20b892c14e8108286557
-  data.tar.gz: f1431d0adb6348e78b62e96c96fa017dda979c282a5d5c61a164f287a6a85b3650bb83cf1206c40d0eb3387ddd9c6cab2325d6d2c12824e0ca5a7df4d9576602
+  metadata.gz: f45b85cc4dfebd8a0afb0592a1ddee159656a749033cc5abf395f88c19742a705263018b72acb234142d2821b3b5ba2e1a09ff1884347b0a73bea5b1a6b0c3bc
+  data.tar.gz: 9b000d84f97eaa7e4f6b775ce03899a60aae832440fb53b28115d576441df330d42bd3761984fe2765c3b9fed60121caeaed7c697288140a58fb193467e9d082

data/CHANGELOG.md

@@ -1,3 +1,21 @@
+## [[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.
+  - Add `LLAMA_FTYPE_MOSTLY_NVFP4` constant value.
+
+## [[0.24.1](https://github.com/yoshoku/llama_cpp.rb/compare/v0.24.0...v0.24.1)] - 2026-03-01
+
+- Change supported llama.cpp version to b8170.
+  - Add `dry_run` accessor to `LlamaModelQuantizeParams`.
+
+## [[0.24.0](https://github.com/yoshoku/llama_cpp.rb/compare/v0.23.11...v0.24.0)] - 2026-02-22
+
+- Change supported llama.cpp version to b8110.
+  - Add `llama_set_adapters_lora` module function to `LlamaCpp`.
+  - Remove `llama_set_adapter_lora` module function.
+  - Remove `llama_clear_adapter_lora` module function.
+  - Remove `llama_rm_adapter_lora` module function.
+
 ## [[0.23.11](https://github.com/yoshoku/llama_cpp.rb/compare/v0.23.10...v0.23.11)] - 2026-01-24
 
 - Change supported llama.cpp version to b7790.

data/ext/llama_cpp/llama_cpp.c

@@ -1024,6 +1024,17 @@ static VALUE llama_model_quantize_params_set_keep_split(VALUE self, VALUE keep_s
   return keep_split;
 }
 
+static VALUE llama_model_quantize_params_get_dry_run(VALUE self) {
+  llama_model_quantize_params* data = get_llama_model_quantize_params(self);
+  return data->dry_run ? Qtrue : Qfalse;
+}
+
+static VALUE llama_model_quantize_params_set_dry_run(VALUE self, VALUE dry_run) {
+  llama_model_quantize_params* data = get_llama_model_quantize_params(self);
+  data->dry_run = RTEST(dry_run) ? true : false;
+  return dry_run;
+}
+
 /* llama_logit_bias */
 static void llama_logit_bias_free(void *ptr) {
   if (ptr) {
@@ -1992,70 +2003,62 @@ static VALUE rb_llama_adapter_meta_count(VALUE self, VALUE adapter) {
 }
 
 /**
- * @overload llama_set_adapter_lora(context, adapter, scale)
+ * @overload llama_set_adapters_lora(context, adapters, scales)
  *  @param [LlamaContext] context
- *  @param [LlamaAdapterLora] adapter
- *  @param [Float] scale
+ *  @param [Array<LlamaAdapterLora>] adapters
+ *  @param [Array<Float>] scales
  *  @return [Integer]
  */
-static VALUE rb_llama_set_adapter_lora(VALUE self, VALUE ctx, VALUE adapter, VALUE scale) {
+static VALUE rb_llama_set_adapters_lora(VALUE self, VALUE ctx, VALUE adapters, VALUE scales) {
   if (!rb_obj_is_kind_of(ctx, rb_cLlamaContext)) {
     rb_raise(rb_eArgError, "ctx must be a LlamaContext");
     return Qnil;
   }
-  if (!rb_obj_is_kind_of(adapter, rb_cLlamaAdapterLora)) {
-    rb_raise(rb_eArgError, "adapter must be a LlamaAdapterLora");
+  if (!RB_TYPE_P(adapters, T_ARRAY)) {
+    rb_raise(rb_eArgError, "adapters must be an Array");
     return Qnil;
   }
-  if (!RB_FLOAT_TYPE_P(scale)) {
-    rb_raise(rb_eArgError, "scale must be a Float");
+  if (!RB_TYPE_P(scales, T_ARRAY)) {
+    rb_raise(rb_eArgError, "scales must be an Array");
     return Qnil;
   }
-  llama_adapter_lora_wrapper* adapter_wrapper = get_llama_adapter_lora_wrapper(adapter);
-  llama_context_wrapper* context_wrapper = get_llama_context_wrapper(ctx);
-  const int32_t res = llama_set_adapter_lora(context_wrapper->context, adapter_wrapper->adapter, (float)NUM2DBL(scale));
-  RB_GC_GUARD(ctx);
-  RB_GC_GUARD(adapter);
-  return NUM2INT(res);
-}
-
-/**
- * @overload llama_rm_adapter_lora(context, adapter)
- *  @param [LlamaContext] context
- *  @param [LlamaAdapterLora] adapter
- *  @return [Integer]
- */
-static VALUE rb_llama_rm_adapter_lora(VALUE self, VALUE ctx, VALUE adapter) {
-  if (!rb_obj_is_kind_of(ctx, rb_cLlamaContext)) {
-    rb_raise(rb_eArgError, "ctx must be a LlamaContext");
+  long n_adapters = RARRAY_LEN(adapters);
+  long n_scales = RARRAY_LEN(scales);
+  if (n_adapters != n_scales) {
+    rb_raise(rb_eArgError, "adapters and scales must have the same length");
     return Qnil;
   }
-  if (!rb_obj_is_kind_of(adapter, rb_cLlamaAdapterLora)) {
-    rb_raise(rb_eArgError, "adapter must be a LlamaAdapterLora");
-    return Qnil;
+  for (long i = 0; i < n_adapters; i++) {
+    VALUE adapter = rb_ary_entry(adapters, i);
+    if (!rb_obj_is_kind_of(adapter, rb_cLlamaAdapterLora)) {
+      rb_raise(rb_eArgError, "adapters must be an Array of LlamaAdapterLora");
+      return Qnil;
+    }
   }
-  llama_context_wrapper* context_wrapper = get_llama_context_wrapper(ctx);
-  llama_adapter_lora_wrapper* adapter_wrapper = get_llama_adapter_lora_wrapper(adapter);
-  const int32_t res = llama_rm_adapter_lora(context_wrapper->context, adapter_wrapper->adapter);
-  RB_GC_GUARD(ctx);
-  RB_GC_GUARD(adapter);
-  return NUM2INT(res);
-}
-
-/**
- * @overload llama_clear_adapter_lora(context)
- *  @param [LlamaContext] context
- *  @return [NilClass]
- */
-static VALUE rb_llama_clear_adapter_lora(VALUE self, VALUE ctx) {
-  if (!rb_obj_is_kind_of(ctx, rb_cLlamaContext)) {
-    rb_raise(rb_eArgError, "ctx must be a LlamaContext");
-    return Qnil;
+  for (long i = 0; i < n_scales; i++) {
+    VALUE scale = rb_ary_entry(scales, i);
+    if (!RB_FLOAT_TYPE_P(scale)) {
+      rb_raise(rb_eArgError, "scales must be an Array of Float");
+      return Qnil;
+    }
+  }
+  struct llama_adapter_lora** adapters_ = ALLOCA_N(struct llama_adapter_lora*, n_adapters);
+  for (long i = 0; i < n_adapters; i++) {
+    VALUE adapter = rb_ary_entry(adapters, i);
+    llama_adapter_lora_wrapper* adapter_wrapper = get_llama_adapter_lora_wrapper(adapter);
+    adapters_[i] = adapter_wrapper->adapter;
+  }
+  float* scales_ = ALLOCA_N(float, n_scales);
+  for (long i = 0; i < n_scales; i++) {
+    VALUE scale = rb_ary_entry(scales, i);
+    scales_[i] = (float)NUM2DBL(scale);
   }
   llama_context_wrapper* context_wrapper = get_llama_context_wrapper(ctx);
-  llama_clear_adapter_lora(context_wrapper->context);
+  const int32_t res = llama_set_adapters_lora(context_wrapper->context, adapters_, n_adapters, scales_);
   RB_GC_GUARD(ctx);
-  return Qnil;
+  RB_GC_GUARD(adapters);
+  RB_GC_GUARD(scales);
+  return NUM2INT(res);
 }
 
 /**
@@ -4192,6 +4195,7 @@ void Init_llama_cpp(void) {
   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_MOSTLY_NVFP4", INT2NUM(LLAMA_FTYPE_MOSTLY_NVFP4));
   rb_define_const(rb_mLlamaCpp, "LLAMA_FTYPE_GUESSED", INT2NUM(LLAMA_FTYPE_GUESSED));
   /* llama_rope_scaling_type */
   /* Document-const: LlamaCpp::LLAMA_ROPE_SCALING_TYPE_UNSPECIFIED */
@@ -4912,6 +4916,17 @@ void Init_llama_cpp(void) {
    * @return [Boolean]
    */
   rb_define_method(rb_cLlamaModelQuantizeParams, "keep_split=", RUBY_METHOD_FUNC(llama_model_quantize_params_set_keep_split), 1);
+  /**
+   * Document-method: dry_run
+   * @return [Boolean]
+   */
+  rb_define_method(rb_cLlamaModelQuantizeParams, "dry_run", RUBY_METHOD_FUNC(llama_model_quantize_params_get_dry_run), 0);
+  /**
+   * Document-method: dry_run=
+   * @param [Boolean] dry_run
+   * @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 */
@@ -4987,6 +5002,7 @@ void Init_llama_cpp(void) {
 
   /* TODO: llama_attach_threadpool */
   /* TODO: llama_detach_threadpool */
+  /* TODO: llama_model_init_from_user */
 
   /* llama_model_load_from_file */
   rb_define_module_function(rb_mLlamaCpp, "llama_model_load_from_file", rb_llama_model_load_from_file, 2);
@@ -5159,14 +5175,8 @@ void Init_llama_cpp(void) {
   /* TODO: llama_adapter_meta_key_by_index */
   /* TODO: llama_adapter_meta_val_str_by_index */
 
-  /* llama_set_adapter_lora */
-  rb_define_module_function(rb_mLlamaCpp, "llama_set_adapter_lora", rb_llama_set_adapter_lora, 3);
-
-  /* llama_rm_adapter_lora */
-  rb_define_module_function(rb_mLlamaCpp, "llama_rm_adapter_lora", rb_llama_rm_adapter_lora, 2);
-
-  /* llama_clear_adapter_lora */
-  rb_define_module_function(rb_mLlamaCpp, "llama_clear_adapter_lora", rb_llama_clear_adapter_lora, 1);
+  /* llama_set_adapters_lora */
+  rb_define_module_function(rb_mLlamaCpp, "llama_set_adapters_lora", rb_llama_set_adapters_lora, 3);
 
   /* llama_adapter_lora_free */
   rb_define_module_function(rb_mLlamaCpp, "llama_adapter_lora_free", rb_llama_adapter_lora_free, 1);

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 = '0.23.11'
+  VERSION = '0.24.2'
 
   # The supported version of llama.cpp.
-  LLAMA_CPP_VERSION = 'b7790'
+  LLAMA_CPP_VERSION = 'b8340'
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: llama_cpp
 version: !ruby/object:Gem::Version
-  version: 0.23.11
+  version: 0.24.2
 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/0.23.11/
+  documentation_uri: https://gemdocs.org/gems/llama_cpp/0.24.2/
   rubygems_mfa_required: 'true'
 rdoc_options: []
 require_paths: