llama_cpp 0.25.2 → 0.25.4

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 2c2a769491201813ec3b65cc286f283342ca492cf8a893b24ba46a4536b7669d
-  data.tar.gz: 95799ab5fa55ccdd691d65f81e5c48cd0702163c4eff2f589c2384e43d64af45
+  metadata.gz: 7918f80b77eb32fa218c3cd9845594eb6f882e066e600caa2af31a979f916610
+  data.tar.gz: 2a631621308ba8206d5c27c8dad8d01b415865d99c1f632e197c4bea6ae43941
 SHA512:
-  metadata.gz: a58b1879d867586b9b60a44b468babb69996359fb91bf8cf27f02a3c1b8a30538311f973a762f172d62e47dbeef8e2652c44d0b88cbbed345bba5532a5210d77
-  data.tar.gz: 5ca17aa628b3c1d2a95bb1e21101c0531bbd7216efffe07fc3ac184cf7512fef54192e85762f74d363c090d1e91a2d3497e13f030beaf89d3f162f2dee15ac2b
+  metadata.gz: 7429e8a0d7255b8d000fa6d2be3d346573743f8238d7fe3d29c4f6a0cdafeb17ef7f5cd6faf1dbad9272e402af1a85e924ebaec1992e2a1077cb5796b2e8521f
+  data.tar.gz: d3eaeb4ac110955a9475aca37ac16c660e5e7780275c6861450ced9e1463478c9f7c30a8f6cc61aa22bca2cfff8a4bae1e10fd3c93e3170bb7d5e2ea1206c210

data/CHANGELOG.md

@@ -1,12 +1,30 @@
+## [[0.25.4](https://github.com/yoshoku/llama_cpp.rb/compare/v0.25.3...v0.25.4)] - 2026-06-13
+
+- Change supported llama.cpp version to b9610.
+  - Add `n_outputs_max` accessor to `LlamaContextParams`.
+
+## [[0.25.3](https://github.com/yoshoku/llama_cpp.rb/compare/v0.25.2...v0.25.3)] - 2026-05-24
+
+- Add `llama_model_chat_template` module function to `LlamaCpp`.
+- Add `llama_model_meta_val_str` module function to `LlamaCpp`.
+- Add `llama_model_meta_key_by_index` module function to `LlamaCpp`.
+- Add `llama_model_meta_val_str_by_index` module function to `LlamaCpp`.
+- Change supported llama.cpp version to b9290.
+  - Add `LLAMA_CONTEXT_TYPE_DEFAULT` constant value.
+  - Add `LLAMA_CONTEXT_TYPE_MTP` constant value.
+  - Add `ctx_type` accessor to `LlamaContextParams`.
+  - Add `n_rs_seq` accessor to `LlamaContextParams`.
+  - Add `n_rs_seq` module function to `LlamaCpp`.
+
 ## [[0.25.2](https://github.com/yoshoku/llama_cpp.rb/compare/v0.25.1...v0.25.2)] - 2026-05-16
 
 - Change supported llama.cpp version to b9150.
-  - ADD `LLAMA_STATE_SEQ_FLAGS_NONE` constant value.
+  - Add `LLAMA_STATE_SEQ_FLAGS_NONE` constant value.
 
 ## [[0.25.1](https://github.com/yoshoku/llama_cpp.rb/compare/v0.25.0...v0.25.1)] - 2026-05-09
 
 - Change supported llama.cpp version to b9070.
-  - ADD `LLAMA_STATE_SEQ_FLAGS_ON_DEVICE` constant value.
+  - Add `LLAMA_STATE_SEQ_FLAGS_ON_DEVICE` constant value.
 
 ## [[0.25.0](https://github.com/yoshoku/llama_cpp.rb/compare/v0.24.3...v0.25.0)] - 2026-04-25
 

data/ext/llama_cpp/llama_cpp.c

@@ -652,6 +652,28 @@ static VALUE llama_context_params_set_n_seq_max(VALUE self, VALUE n_seq_max) {
   return n_seq_max;
 }
 
+static VALUE llama_context_params_get_n_rs_seq(VALUE self) {
+  struct llama_context_params* data = get_llama_context_params(self);
+  return UINT2NUM(data->n_rs_seq);
+}
+
+static VALUE llama_context_params_set_n_rs_seq(VALUE self, VALUE n_rs_seq) {
+  struct llama_context_params* data = get_llama_context_params(self);
+  data->n_rs_seq = NUM2UINT(n_rs_seq);
+  return n_rs_seq;
+}
+
+static VALUE llama_context_params_get_n_outputs_max(VALUE self) {
+  struct llama_context_params* data = get_llama_context_params(self);
+  return UINT2NUM(data->n_outputs_max);
+}
+
+static VALUE llama_context_params_set_n_outputs_max(VALUE self, VALUE n_outputs_max) {
+  struct llama_context_params* data = get_llama_context_params(self);
+  data->n_outputs_max = NUM2UINT(n_outputs_max);
+  return n_outputs_max;
+}
+
 static VALUE llama_context_params_get_n_threads(VALUE self) {
   struct llama_context_params* data = get_llama_context_params(self);
   return INT2NUM(data->n_threads);
@@ -674,6 +696,17 @@ static VALUE llama_context_params_set_n_threads_batch(VALUE self, VALUE n_thread
   return n_threads_batch;
 }
 
+static VALUE llama_context_params_get_ctx_type(VALUE self) {
+  struct llama_context_params* data = get_llama_context_params(self);
+  return INT2NUM(data->ctx_type);
+}
+
+static VALUE llama_context_params_set_ctx_type(VALUE self, VALUE ctx_type) {
+  struct llama_context_params* data = get_llama_context_params(self);
+  data->ctx_type = (enum llama_context_type)NUM2INT(ctx_type);
+  return ctx_type;
+}
+
 static VALUE llama_context_params_get_rope_scaling_type(VALUE self) {
   struct llama_context_params* data = get_llama_context_params(self);
   return INT2NUM(data->rope_scaling_type);
@@ -1602,6 +1635,20 @@ static VALUE rb_llama_n_seq_max(VALUE self, VALUE ctx) {
   return UINT2NUM(llama_n_seq_max(context_wrapper->context));
 }
 
+/**
+ * @overload llama_n_rs_seq(context)
+ *  @param [LlamaContext] context
+ *  @return [Integer]
+ */
+static VALUE rb_llama_n_rs_seq(VALUE self, VALUE ctx) {
+  if (!rb_obj_is_kind_of(ctx, rb_cLlamaContext)) {
+    rb_raise(rb_eArgError, "ctx must be a LlamaContext");
+    return Qnil;
+  }
+  llama_context_wrapper* context_wrapper = get_llama_context_wrapper(ctx);
+  return UINT2NUM(llama_n_rs_seq(context_wrapper->context));
+}
+
 /**
  * @overload llama_get_model(context)
  *  @param [LlamaContext] context
@@ -1887,6 +1934,116 @@ static VALUE rb_llama_model_meta_key_str(VALUE self, VALUE key) {
   return rb_utf8_str_new_cstr(key_str);
 }
 
+/**
+ * @overload llama_model_meta_val_str(model, key)
+ *  @param [LlamaModel] model
+ *  @param [String] key
+ *  @return [String, nil] nil if the key is not found
+ */
+static VALUE rb_llama_model_meta_val_str(VALUE self, VALUE model, VALUE key) {
+  if (!rb_obj_is_kind_of(model, rb_cLlamaModel)) {
+    rb_raise(rb_eArgError, "model must be a LlamaModel");
+    return Qnil;
+  }
+  if (!RB_TYPE_P(key, T_STRING)) {
+    rb_raise(rb_eArgError, "key must be a String");
+    return Qnil;
+  }
+  llama_model_wrapper* model_wrapper = get_llama_model_wrapper(model);
+  const char* key_ = StringValueCStr(key);
+  char stack_buf[1024];
+  int32_t n = llama_model_meta_val_str(model_wrapper->model, key_, stack_buf, sizeof(stack_buf));
+  if (n < 0) {
+    RB_GC_GUARD(model);
+    RB_GC_GUARD(key);
+    return Qnil;
+  }
+  VALUE result;
+  if ((size_t)n < sizeof(stack_buf)) {
+    result = rb_utf8_str_new(stack_buf, n);
+  } else {
+    char* heap_buf = (char*)ruby_xmalloc((size_t)n + 1);
+    llama_model_meta_val_str(model_wrapper->model, key_, heap_buf, (size_t)n + 1);
+    result = rb_utf8_str_new(heap_buf, n);
+    ruby_xfree(heap_buf);
+  }
+  RB_GC_GUARD(model);
+  RB_GC_GUARD(key);
+  return result;
+}
+
+/**
+ * @overload llama_model_meta_key_by_index(model, idx)
+ *  @param [LlamaModel] model
+ *  @param [Integer] idx
+ *  @return [String, nil] nil if the index is out of range
+ */
+static VALUE rb_llama_model_meta_key_by_index(VALUE self, VALUE model, VALUE idx) {
+  if (!rb_obj_is_kind_of(model, rb_cLlamaModel)) {
+    rb_raise(rb_eArgError, "model must be a LlamaModel");
+    return Qnil;
+  }
+  if (!RB_INTEGER_TYPE_P(idx)) {
+    rb_raise(rb_eArgError, "i must be an Integer");
+    return Qnil;
+  }
+  llama_model_wrapper* model_wrapper = get_llama_model_wrapper(model);
+  int32_t idx_ = NUM2INT(idx);
+  char stack_buf[1024];
+  int32_t n = llama_model_meta_key_by_index(model_wrapper->model, idx_, stack_buf, sizeof(stack_buf));
+  if (n < 0) {
+    RB_GC_GUARD(model);
+    return Qnil;
+  }
+  VALUE result;
+  if ((size_t)n < sizeof(stack_buf)) {
+    result = rb_utf8_str_new(stack_buf, n);
+  } else {
+    char* heap_buf = (char*)ruby_xmalloc((size_t)n + 1);
+    llama_model_meta_key_by_index(model_wrapper->model, idx_, heap_buf, (size_t)n + 1);
+    result = rb_utf8_str_new(heap_buf, n);
+    ruby_xfree(heap_buf);
+  }
+  RB_GC_GUARD(model);
+  return result;
+}
+
+/**
+ * @overload llama_model_meta_val_str_by_index(model, idx)
+ *  @param [LlamaModel] model
+ *  @param [Integer] idx
+ *  @return [String, nil] nil if the index is out of range
+ */
+static VALUE rb_llama_model_meta_val_str_by_index(VALUE self, VALUE model, VALUE idx) {
+  if (!rb_obj_is_kind_of(model, rb_cLlamaModel)) {
+    rb_raise(rb_eArgError, "model must be a LlamaModel");
+    return Qnil;
+  }
+  if (!RB_INTEGER_TYPE_P(idx)) {
+    rb_raise(rb_eArgError, "i must be an Integer");
+    return Qnil;
+  }
+  llama_model_wrapper* model_wrapper = get_llama_model_wrapper(model);
+  int32_t idx_ = NUM2INT(idx);
+  char stack_buf[1024];
+  int32_t n = llama_model_meta_val_str_by_index(model_wrapper->model, idx_, stack_buf, sizeof(stack_buf));
+  if (n < 0) {
+    RB_GC_GUARD(model);
+    return Qnil;
+  }
+  VALUE result;
+  if ((size_t)n < sizeof(stack_buf)) {
+    result = rb_utf8_str_new(stack_buf, n);
+  } else {
+    char* heap_buf = (char*)ruby_xmalloc((size_t)n + 1);
+    llama_model_meta_val_str_by_index(model_wrapper->model, idx_, heap_buf, (size_t)n + 1);
+    result = rb_utf8_str_new(heap_buf, n);
+    ruby_xfree(heap_buf);
+  }
+  RB_GC_GUARD(model);
+  return result;
+}
+
 /**
  * @overload llama_model_desc(model)
  *  @param [LlamaModel] model
@@ -1918,19 +2075,28 @@ static VALUE rb_llama_model_size(VALUE self, VALUE model) {
   return ULONG2NUM(llama_model_size(model_wrapper->model));
 }
 
-/* llama_model_chat_template */
-/*
-static VALUE rb_llama_model_chat_template(VALUE self, VALUE model) {
+/**
+ * @overload llama_model_chat_template(model, name)
+ *  @param [LlamaModel] model
+ *  @param [String, nil] name pass nil to get the default chat template
+ *  @return [String, nil] nil if no chat template is available
+ */
+static VALUE rb_llama_model_chat_template(VALUE self, VALUE model, VALUE name) {
   if (!rb_obj_is_kind_of(model, rb_cLlamaModel)) {
     rb_raise(rb_eArgError, "model must be a LlamaModel");
     return Qnil;
   }
+  if (!NIL_P(name) && !RB_TYPE_P(name, T_STRING)) {
+    rb_raise(rb_eArgError, "name must be a String or nil");
+    return Qnil;
+  }
   llama_model_wrapper* model_wrapper = get_llama_model_wrapper(model);
-  const char* templ = llama_model_chat_template(model_wrapper->model)
+  const char* name_ = NIL_P(name) ? NULL : StringValueCStr(name);
+  const char* templ = llama_model_chat_template(model_wrapper->model, name_);
   RB_GC_GUARD(model);
-  return rb_utf8_str_new_cstr(templ);
+  RB_GC_GUARD(name);
+  return templ == NULL ? Qnil : rb_utf8_str_new_cstr(templ);
 }
-*/
 
 /**
  * @overload llama_model_n_params(model)
@@ -4312,6 +4478,10 @@ void Init_llama_cpp(void) {
   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_const(rb_mLlamaCpp, "LLAMA_SPLIT_MODE_TENSOR", INT2NUM(LLAMA_SPLIT_MODE_TENSOR));
+  /* llama_context_type */
+  /* Document-const: LlamaCpp::LLAMA_CONTEXT_TYPE_DEFAULT */
+  rb_define_const(rb_mLlamaCpp, "LLAMA_CONTEXT_TYPE_DEFAULT", INT2NUM(LLAMA_CONTEXT_TYPE_DEFAULT));
+  rb_define_const(rb_mLlamaCpp, "LLAMA_CONTEXT_TYPE_MTP", INT2NUM(LLAMA_CONTEXT_TYPE_MTP));
 
   rb_define_module_function(rb_mLlamaCpp, "llama_flash_attn_type_name", rb_llama_flash_attn_type_name, 1);
 
@@ -4644,6 +4814,28 @@ void Init_llama_cpp(void) {
    * @return [Integer]
    */
   rb_define_method(rb_cLlamaContextParams, "n_seq_max=", RUBY_METHOD_FUNC(llama_context_params_set_n_seq_max), 1);
+  /**
+   * Document-method: n_rs_seq
+   * @return [Integer]
+   */
+  rb_define_method(rb_cLlamaContextParams, "n_rs_seq", RUBY_METHOD_FUNC(llama_context_params_get_n_rs_seq), 0);
+  /**
+   * Document-method: n_rs_seq=
+   * @param [Integer] n_rs_seq
+   * @return [Integer]
+   */
+  rb_define_method(rb_cLlamaContextParams, "n_rs_seq=", RUBY_METHOD_FUNC(llama_context_params_set_n_rs_seq), 1);
+  /**
+   * Document-method: n_outputs_max
+   * @return [Integer]
+   */
+  rb_define_method(rb_cLlamaContextParams, "n_outputs_max", RUBY_METHOD_FUNC(llama_context_params_get_n_outputs_max), 0);
+  /**
+   * Document-method: n_outputs_max=
+   * @param [Integer] n_outputs_max
+   * @return [Integer]
+   */
+  rb_define_method(rb_cLlamaContextParams, "n_outputs_max=", RUBY_METHOD_FUNC(llama_context_params_set_n_outputs_max), 1);
   /**
    * Document-method: n_threads
    * @return [Integer]
@@ -4666,6 +4858,17 @@ void Init_llama_cpp(void) {
    * @return [Integer]
    */
   rb_define_method(rb_cLlamaContextParams, "n_threads_batch=", RUBY_METHOD_FUNC(llama_context_params_set_n_threads_batch), 1);
+  /**
+   * Document-method: ctx_type
+   * @return [Integer]
+   */
+  rb_define_method(rb_cLlamaContextParams, "ctx_type", RUBY_METHOD_FUNC(llama_context_params_get_ctx_type), 0);
+  /**
+   * Document-method: ctx_type=
+   * @param [Integer] ctx_type
+   * @return [Integer]
+   */
+  rb_define_method(rb_cLlamaContextParams, "ctx_type=", RUBY_METHOD_FUNC(llama_context_params_set_ctx_type), 1);
   /**
    * Document-method: rope_scaling_type
    * @return [Integer]
@@ -4885,6 +5088,7 @@ void Init_llama_cpp(void) {
 
   /* TODO: struct llama_sampler_seq_config * samplers */
   /* TODO: size_t n_samplers */
+  /* TODO: struct llama_context * ctx_other */
 
   /**
    * Document-method: kv_unified=
@@ -5186,6 +5390,9 @@ void Init_llama_cpp(void) {
   /* llama_n_seq_max */
   rb_define_module_function(rb_mLlamaCpp, "llama_n_seq_max", rb_llama_n_seq_max, 1);
 
+  /* llama_n_rs_seq */
+  rb_define_module_function(rb_mLlamaCpp, "llama_n_rs_seq", rb_llama_n_rs_seq, 1);
+
   /* TODO: llama_get_model */
   rb_define_module_function(rb_mLlamaCpp, "llama_get_model", rb_llama_get_model, 1);
 
@@ -5244,9 +5451,12 @@ void Init_llama_cpp(void) {
   rb_define_module_function(rb_mLlamaCpp, "llama_model_meta_count", rb_llama_model_meta_count, 1);
   /* llama_model_meta_key_str */
   rb_define_module_function(rb_mLlamaCpp, "llama_model_meta_key_str", rb_llama_model_meta_key_str, 1);
-  /* TODO: llama_model_meta_val_str */
-  /* TODO: llama_model_meta_key_by_index */
-  /* TODO: llama_model_meta_val_str_by_index */
+  /* llama_model_meta_val_str */
+  rb_define_module_function(rb_mLlamaCpp, "llama_model_meta_val_str", rb_llama_model_meta_val_str, 2);
+  /* llama_model_meta_key_by_index */
+  rb_define_module_function(rb_mLlamaCpp, "llama_model_meta_key_by_index", rb_llama_model_meta_key_by_index, 2);
+  /* llama_model_meta_val_str_by_index */
+  rb_define_module_function(rb_mLlamaCpp, "llama_model_meta_val_str_by_index", rb_llama_model_meta_val_str_by_index, 2);
 
   /* llama_model_desc */
   rb_define_module_function(rb_mLlamaCpp, "llama_model_desc", rb_llama_model_desc, 1);
@@ -5254,8 +5464,8 @@ void Init_llama_cpp(void) {
   /* llama_model_size */
   rb_define_module_function(rb_mLlamaCpp, "llama_model_size", rb_llama_model_size, 1);
 
-  /* TODO: llama_model_chat_template */
-  /* rb_define_module_function(rb_mLlamaCpp, "llama_model_chat_template", rb_llama_model_chat_template, 1); */
+  /* llama_model_chat_template */
+  rb_define_module_function(rb_mLlamaCpp, "llama_model_chat_template", rb_llama_model_chat_template, 2);
 
   /* llama_model_n_params */
   rb_define_module_function(rb_mLlamaCpp, "llama_model_n_params", rb_llama_model_n_params, 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 = '0.25.2'
+  VERSION = '0.25.4'
 
   # The supported version of llama.cpp.
-  LLAMA_CPP_VERSION = 'b9150'
+  LLAMA_CPP_VERSION = 'b9610'
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: llama_cpp
 version: !ruby/object:Gem::Version
-  version: 0.25.2
+  version: 0.25.4
 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/0.25.2/
+  documentation_uri: https://gemdocs.org/gems/llama_cpp/0.25.4/
   rubygems_mfa_required: 'true'
 rdoc_options: []
 require_paths:
@@ -49,7 +49,7 @@ required_rubygems_version: !ruby/object:Gem::Requirement
     - !ruby/object:Gem::Version
       version: '0'
 requirements: []
-rubygems_version: 4.0.10
+rubygems_version: 4.0.13
 specification_version: 4
 summary: Ruby bindings for the llama.cpp.
 test_files: []