llama_cpp 0.22.0 → 0.23.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 44ea7eb462df4d37cd20d802e5f305d4ff05b605e3d442ca966458714a57881b
-  data.tar.gz: f3e73839dda61b43efd71c0d325384b78c6b47fc8535ab62d25a54c66aa6d09f
+  metadata.gz: bdc82f63eb7cca5133f24159dd648fa2722896a3b9cee9cafc11022e28646b5d
+  data.tar.gz: 0adad024f89582c57f1e541cb85b4220b171b64c4b12bb2f38d635efed4b6458
 SHA512:
-  metadata.gz: b2060e02b8bb154b5d34c242b54f331b840cc5e3f46bce6bcc307b52de4e6eb2613a5e725154b51c9e272576c00d004f72827acb035f62c280862fae94a9f5a5
-  data.tar.gz: 43a110ce026562b17b6f7da9b9bd013be7671bf457c3ddb84a3de362333a7b7f665031577755c7de984e0e84da39f1c474f96dc830040fa08185b5b405169697
+  metadata.gz: b2e9ce298ed0f5d2cb684c4362ed4edadd14af7b3c99ccb989308483b58b81eb659c64b8c91216eb27684f17f36170833a9bb3e3dd4c05ceacf9bc3f0603c159
+  data.tar.gz: aee0bcd5b2cecd91baf0473705ba4e43b74325a0a490f3d7a872a47e629248734cc76076dd18f584855db83671ae00896bf8185a046086cfbc9bcc741fc67064

data/CHANGELOG.md

@@ -1,3 +1,16 @@
+## [[0.23.0](https://github.com/yoshoku/llama_cpp.rb/compare/v0.22.1...v0.23.0)] - 2025-09-05
+
+- Change supported llama.cpp version to b6380.
+  - Add `llama_flash_attn_type_name` module function.
+  - Add `flash_attn_type` accessor to `LlamaContextParams`.
+  - Add `LLAMA_FLASH_ATTN_TYPE_AUTO`, `LLAMA_FLASH_ATTN_TYPE_DISABLED`, and `LLAMA_FLASH_ATTN_TYPE_ENABLED` constants.
+  - Remove `flash_attn` accessor from `LlamaContextParams`.
+
+## [[0.22.1](https://github.com/yoshoku/llama_cpp.rb/compare/v0.22.0...v0.22.1)] - 2025-08-30
+
+- Change supported llama.cpp version to b6310.
+  - Add `llama_adapter_meta_count` module function.
+
 ## [[0.22.0](https://github.com/yoshoku/llama_cpp.rb/compare/v0.21.2...v0.22.0)] - 2025-08-23
 
 - Change supported llama.cpp version to b6240.

data/ext/llama_cpp/llama_cpp.c

@@ -672,6 +672,17 @@ static VALUE llama_context_params_set_attention_type(VALUE self, VALUE attention
   return attention_type;
 }
 
+static VALUE llama_context_params_get_flash_attn_type(VALUE self) {
+  struct llama_context_params* data = get_llama_context_params(self);
+  return INT2NUM(data->flash_attn_type);
+}
+
+static VALUE llama_context_params_set_flash_attn_type(VALUE self, VALUE flash_attn_type) {
+  struct llama_context_params* data = get_llama_context_params(self);
+  data->flash_attn_type = (enum llama_flash_attn_type)NUM2INT(flash_attn_type);
+  return flash_attn_type;
+}
+
 static VALUE llama_context_params_get_rope_freq_base(VALUE self) {
   struct llama_context_params* data = get_llama_context_params(self);
   return DBL2NUM(data->rope_freq_base);
@@ -804,17 +815,6 @@ static VALUE llama_context_params_set_offload_kqv(VALUE self, VALUE offload_kqv)
   return offload_kqv;
 }
 
-static VALUE llama_context_params_get_flash_attn(VALUE self) {
-  struct llama_context_params* data = get_llama_context_params(self);
-  return data->flash_attn ? Qtrue : Qfalse;
-}
-
-static VALUE llama_context_params_set_flash_attn(VALUE self, VALUE flash_attn) {
-  struct llama_context_params* data = get_llama_context_params(self);
-  data->flash_attn = RTEST(flash_attn) ? true : false;
-  return flash_attn;
-}
-
 static VALUE llama_context_params_get_no_perf(VALUE self) {
   struct llama_context_params* data = get_llama_context_params(self);
   return data->no_perf ? Qtrue : Qfalse;
@@ -1852,6 +1852,20 @@ static VALUE rb_llama_adapter_lora_init(VALUE self, VALUE model, VALUE path_lora
   return TypedData_Wrap_Struct(rb_cLlamaAdapterLora, &llama_adapter_lora_wrapper_data_type, adapter_wrapper);
 }
 
+/**
+ * @overload llama_adapter_meta_count(adapter)
+ * @param [LlamaAdapterLora] adapter
+ * @return [Integer]
+ */
+static VALUE rb_llama_adapter_meta_count(VALUE self, VALUE adapter) {
+  if (!rb_obj_is_kind_of(adapter, rb_cLlamaAdapterLora)) {
+    rb_raise(rb_eArgError, "adapter must be a LlamaAdapterLora");
+    return Qnil;
+  }
+  llama_adapter_lora_wrapper* adapter_wrapper = get_llama_adapter_lora_wrapper(adapter);
+  return INT2NUM(llama_adapter_meta_count(adapter_wrapper->adapter));
+}
+
 /**
  * @overload llama_set_adapter_lora(context, adapter, scale)
  *  @param [LlamaContext] context
@@ -3858,6 +3872,20 @@ static VALUE rb_llama_perf_sampler_reset(VALUE self, VALUE chain) {
   return Qnil;
 }
 
+/**
+ * @overload llama_flash_attn_type_name(flash_attn_type)
+ *  @param [Integer] flash_attn_type
+ *  @return [String]
+ */
+static VALUE rb_llama_flash_attn_type_name(VALUE self, VALUE flash_attn_type) {
+  if (!RB_INTEGER_TYPE_P(flash_attn_type)) {
+    rb_raise(rb_eArgError, "flash_attn_type must be an Integer");
+    return Qnil;
+  }
+  const char* name = llama_flash_attn_type_name((enum llama_flash_attn_type)NUM2INT(flash_attn_type));
+  return rb_utf8_str_new_cstr(name);
+}
+
 /* MAIN */
 void Init_llama_cpp(void) {
   char tmp[12];
@@ -4005,12 +4033,19 @@ void Init_llama_cpp(void) {
   rb_define_const(rb_mLlamaCpp, "LLAMA_ATTENTION_TYPE_UNSPECIFIED", INT2NUM(LLAMA_ATTENTION_TYPE_UNSPECIFIED));
   rb_define_const(rb_mLlamaCpp, "LLAMA_ATTENTION_TYPE_CAUSAL", INT2NUM(LLAMA_ATTENTION_TYPE_CAUSAL));
   rb_define_const(rb_mLlamaCpp, "LLAMA_ATTENTION_TYPE_NON_CAUSAL", INT2NUM(LLAMA_ATTENTION_TYPE_NON_CAUSAL));
+  /* llama_flash_attn_type */
+  /* Document-const: LlamaCpp::LLAMA_FLASH_ATTN_TYPE_AUTO */
+  rb_define_const(rb_mLlamaCpp, "LLAMA_FLASH_ATTN_TYPE_AUTO", INT2NUM(LLAMA_FLASH_ATTN_TYPE_AUTO));
+  rb_define_const(rb_mLlamaCpp, "LLAMA_FLASH_ATTN_TYPE_DISABLED", INT2NUM(LLAMA_FLASH_ATTN_TYPE_DISABLED));
+  rb_define_const(rb_mLlamaCpp, "LLAMA_FLASH_ATTN_TYPE_ENABLED", INT2NUM(LLAMA_FLASH_ATTN_TYPE_ENABLED));
   /* llama_split_mode */
   /* Document-const: LlamaCpp::LLAMA_SPLIT_MODE_NONE */
   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_module_function(rb_mLlamaCpp, "llama_flash_attn_type_name", rb_llama_flash_attn_type_name, 1);
+
   /**
    * Document-class: LlamaCpp::LlamaTokenData
    * "struct llama_token_data" wrapper class
@@ -4345,6 +4380,17 @@ void Init_llama_cpp(void) {
    * @return [Integer]
    */
   rb_define_method(rb_cLlamaContextParams, "attention_type=", RUBY_METHOD_FUNC(llama_context_params_set_attention_type), 1);
+  /**
+   * Document-method: flash_attn_type
+   * @return [Integer]
+   */
+  rb_define_method(rb_cLlamaContextParams, "flash_attn_type", RUBY_METHOD_FUNC(llama_context_params_get_flash_attn_type), 0);
+  /**
+   * Document-method: flash_attn_type=
+   * @param [Integer] flash_attn_type
+   * @return [Integer]
+   */
+  rb_define_method(rb_cLlamaContextParams, "flash_attn_type=", RUBY_METHOD_FUNC(llama_context_params_set_flash_attn_type), 1);
   /**
    * Document-method: rope_freq_base
    * @return [Float]
@@ -4479,17 +4525,6 @@ void Init_llama_cpp(void) {
    * @return [Boolean]
    */
   rb_define_method(rb_cLlamaContextParams, "offload_kqv=", RUBY_METHOD_FUNC(llama_context_params_set_offload_kqv), 1);
-  /**
-   * Document-method: flash_attn
-   * @return [Boolean]
-   */
-  rb_define_method(rb_cLlamaContextParams, "flash_attn", RUBY_METHOD_FUNC(llama_context_params_get_flash_attn), 0);
-  /**
-   * Document-method: flash_attn=
-   * @param [Boolean] flash_attn
-   * @return [Boolean]
-   */
-  rb_define_method(rb_cLlamaContextParams, "flash_attn=", RUBY_METHOD_FUNC(llama_context_params_set_flash_attn), 1);
   /**
    * Document-method: no_perf
    * @return [Boolean]
@@ -4855,6 +4890,14 @@ void Init_llama_cpp(void) {
   /* llama_adapter_lora_init */
   rb_define_module_function(rb_mLlamaCpp, "llama_adapter_lora_init", rb_llama_adapter_lora_init, 2);
 
+  /* TODO: llama_adapter_meta_val_str */
+
+  /* llama_adapter_meta_count */
+  rb_define_module_function(rb_mLlamaCpp, "llama_adapter_meta_count", rb_llama_adapter_meta_count, 1);
+
+  /* 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);
 

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 = '0.22.0'
+  VERSION = '0.23.0'
 
   # The supported version of llama.cpp.
-  LLAMA_CPP_VERSION = 'b6240'
+  LLAMA_CPP_VERSION = 'b6380'
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: llama_cpp
 version: !ruby/object:Gem::Version
-  version: 0.22.0
+  version: 0.23.0
 platform: ruby
 authors:
 - yoshoku