llama_cpp 0.22.1 → 0.23.1

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: d9d289500d478dbaea942656eeaf4c076dd81fdcbf5fe670c323c5ce431945da
-  data.tar.gz: 5dff9d66db034b7f275add566e760adf84b10b578ff9a2fb32a684fd17735f8a
+  metadata.gz: 839ef41e6f1588768629f776034abce6ea4c668b5de753ea8d9ac5c4c0d93ddd
+  data.tar.gz: c18f5ac34f673247eea8eb63e3e10aed14cd95abe9f8fcb90d7931a32b94482d
 SHA512:
-  metadata.gz: 6b3109cee58c80d79ba90aa3ec33e0517c8bb54a1592ff4020ca18e16bf78c0bed5389ac1c5dd1221e757c87c8e3fb1af226c45d5b3d2a5b1ee6bc7afd13e242
-  data.tar.gz: 139cb66ec6cd2adbd2b178ad7f84581698905870bb549105f9683e14a80d04d8210405762e98937fed685352a4434d89f8fe43d8cbf500b1396d091c7d6366ba
+  metadata.gz: 1045e721b28f804e6536461f15c0de640fc1201d02c97c8ca807c383e9730d779795fbc745d0404dab51dffa59f90e28a90335dff7513bd1909294e4b3382cd9
+  data.tar.gz: 9c8db351db13d57153c1b939eccf73cd355dae7b7281faaa55e91a41e5762db854e685cd69c25c4c1d9e06a9379dd41d07cb18d1fe82154e9bbf7070617a779e

data/CHANGELOG.md

@@ -1,3 +1,16 @@
+## [[0.23.1](https://github.com/yoshoku/llama_cpp.rb/compare/v0.23.0...v0.23.1)] - 2025-09-13
+
+- Change supported llama.cpp version to b6440.
+  - Add `llama_adapter_get_alora_n_invocation_tokens` module function.
+
+## [[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.

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;
@@ -1952,6 +1952,20 @@ static VALUE rb_llama_adapter_lora_free(VALUE self, VALUE adapter) {
   return Qnil;
 }
 
+/**
+ * @overload llama_adapter_get_alora_n_invocation_tokens(adapter)
+ *  @param [LlamaAdapterLora] adapter
+ *  @return [Integer]
+ */
+static VALUE rb_llama_adapter_get_alora_n_invocation_tokens(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 UINT2NUM(llama_adapter_get_alora_n_invocation_tokens(adapter_wrapper->adapter));
+}
+
 /* llama_memory_t wrapper */
 typedef struct {
   llama_memory_t memory;
@@ -3872,6 +3886,20 @@ 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];
@@ -4019,12 +4047,19 @@ 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
@@ -4359,6 +4394,17 @@ 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]
@@ -4493,17 +4539,6 @@ 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]
@@ -4889,6 +4924,11 @@ void Init_llama_cpp(void) {
   /* llama_adapter_lora_free */
   rb_define_module_function(rb_mLlamaCpp, "llama_adapter_lora_free", rb_llama_adapter_lora_free, 1);
 
+  /* llama_adapter_get_alora_n_invocation_tokens */
+  rb_define_module_function(rb_mLlamaCpp, "llama_adapter_get_alora_n_invocation_tokens", rb_llama_adapter_get_alora_n_invocation_tokens, 1);
+
+  /* TODO: llama_adapter_get_alora_invocation_tokens */
+
   /* TODO: llama_apply_adapter_cvec */
 
   /**

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 = '0.22.1'
+  VERSION = '0.23.1'
 
   # The supported version of llama.cpp.
-  LLAMA_CPP_VERSION = 'b6310'
+  LLAMA_CPP_VERSION = 'b6440'
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: llama_cpp
 version: !ruby/object:Gem::Version
-  version: 0.22.1
+  version: 0.23.1
 platform: ruby
 authors:
 - yoshoku