llama_cpp 0.17.7 → 0.17.10

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: ad273fdefc8934e4fdb11f18f20379cd77e89ed9d7e1db84ea4d6d847a676540
-  data.tar.gz: ad91a3d8589e66c8ba3fb7e12a354d917fb0aa9596fd75ae773db79eace59cb5
+  metadata.gz: ec7b41006345cfe57e355d207718ae460a5a2c519d67027daddb7d40ab346754
+  data.tar.gz: efcfa7eef5374790cef788f719bede1ab3cd30e5dec79be0ba9844bfbcd433c2
 SHA512:
-  metadata.gz: 733d18dc4a3c7a72a7de1a4df331f5861d0d2321baf7d439c0f581ceec6ad7246b5dda45739b5154b9ec94c61a094915230506292b48de232f01840b29515f1f
-  data.tar.gz: 745e2d6f265f72871c258a52af907ecc616bdc2bf1f9dc8fc10d1d81597c930c0a999d4a518fe7f3a6d807df90fee8e559e1ef31e19f03f3d81cfa84d6105b9a
+  metadata.gz: c39150e77652060fcfee2c73b1334ff53ed40caf58679efa727a0d0d54bffe036b8795be9751cef8517b5d4b2300fb77fa5753a9c94b803988ec429b999591c2
+  data.tar.gz: 6054dee2f39ac2690de2836405ac793bf987538ef0780cffab2700544bc60ceaca8a4b109407c66dcf2dd9f38ffe6b98659d7ca4ddb5c9e93fa85d9884cea26e

data/CHANGELOG.md

@@ -1,4 +1,24 @@
-## [[0.17.7](https://github.com/yoshoku/llama_cpp.rb/compare/v0.17.5...v0.17.6)] - 2024-08-17
+## [[0.17.10](https://github.com/yoshoku/llama_cpp.rb/compare/v0.17.9...v0.17.10)] - 2024-09-07
+
+- Change supported llama.cpp version to b3676.
+  - Add `LLAMA_VOCAB_TYPE_RWKV` constant.
+  - Add `LLAMA_FTYPE_MOSTLY_TQ1_0` and `LLAMA_FTYPE_MOSTLY_TQ2_0` constants.
+  - Change type of n_threads and n_threads_batch from uint32_t to int32 in native extension codes.
+
+Implementation bindings for llama_attach_threadpool and llama_detach_threadpool have been skipped.
+
+## [[0.17.9](https://github.com/yoshoku/llama_cpp.rb/compare/v0.17.8...v0.17.9)] - 2024-08-31
+
+- Change supported llama.cpp version to b3639.
+  - There are no changes in the API.
+
+## [[0.17.8](https://github.com/yoshoku/llama_cpp.rb/compare/v0.17.7...v0.17.8)] - 2024-08-25
+
+- Change supported llama.cpp version to b3614.
+  - Add `LLAMA_VOCAB_PRE_TYPE_EXAONE` constant.
+  - Add `is_recurrent?` method to `Model`.
+
+## [[0.17.7](https://github.com/yoshoku/llama_cpp.rb/compare/v0.17.6...v0.17.7)] - 2024-08-17
 
 - Change supported llama.cpp version to b3590.
   - Add `LLAMA_VOCAB_PRE_TYPE_BLOOM` and `LLAMA_VOCAB_PRE_TYPE_GPT3_FINNISH` constants

data/ext/llama_cpp/llama_cpp.cpp

@@ -1555,6 +1555,7 @@ public:
     rb_define_method(rb_cLLaMAModel, "has_encoder?", RUBY_METHOD_FUNC(_llama_model_has_encoder), 0);
     rb_define_method(rb_cLLaMAModel, "has_decoder?", RUBY_METHOD_FUNC(_llama_model_has_decoder), 0);
     rb_define_method(rb_cLLaMAModel, "decoder_start_token", RUBY_METHOD_FUNC(_llama_model_decoder_start_token), 0);
+    rb_define_method(rb_cLLaMAModel, "is_recurrent?", RUBY_METHOD_FUNC(_llama_model_is_recurrent), 0);
     rb_define_method(rb_cLLaMAModel, "detokenize", RUBY_METHOD_FUNC(_llama_model_detokenize), -1);
   }
 
@@ -1914,6 +1915,11 @@ private:
     return INT2NUM(llama_model_decoder_start_token(ptr->model));
   }
 
+  static VALUE _llama_model_is_recurrent(VALUE self) {
+    LLaMAModelWrapper* ptr = get_llama_model(self);
+    return llama_model_is_recurrent(ptr->model) ? Qtrue : Qfalse;
+  }
+
   static VALUE _llama_model_detokenize(int argc, VALUE* argv, VALUE self) {
     VALUE kw_args = Qnil;
     ID kw_table[2] = { rb_intern("remove_special"), rb_intern("unparse_special") };
@@ -2506,7 +2512,7 @@ private:
       rb_raise(rb_eArgError, "LLaMA context is not initialized");
       return Qnil;
     }
-    llama_set_n_threads(ptr->ctx, NUM2UINT(n_threads), NUM2UINT(n_threads_batch));
+    llama_set_n_threads(ptr->ctx, NUM2INT(n_threads), NUM2INT(n_threads_batch));
     return Qnil;
   }
 
@@ -2552,7 +2558,7 @@ private:
       rb_raise(rb_eRuntimeError, "LLaMA context is not initialized");
       return Qnil;
     }
-    return UINT2NUM(llama_n_threads(ptr->ctx));
+    return INT2NUM(llama_n_threads(ptr->ctx));
   }
 
   static VALUE _llama_context_n_threads_batch(VALUE self) {
@@ -2561,7 +2567,7 @@ private:
       rb_raise(rb_eRuntimeError, "LLaMA context is not initialized");
       return Qnil;
     }
-    return UINT2NUM(llama_n_threads_batch(ptr->ctx));
+    return INT2NUM(llama_n_threads_batch(ptr->ctx));
   }
 
   static VALUE _llama_context_get_timings(VALUE self) {
@@ -3605,6 +3611,7 @@ extern "C" void Init_llama_cpp(void) {
   rb_define_const(rb_mLLaMACpp, "LLAMA_VOCAB_TYPE_BPE", INT2NUM(LLAMA_VOCAB_TYPE_BPE));
   rb_define_const(rb_mLLaMACpp, "LLAMA_VOCAB_TYPE_WPM", INT2NUM(LLAMA_VOCAB_TYPE_WPM));
   rb_define_const(rb_mLLaMACpp, "LLAMA_VOCAB_TYPE_UGM", INT2NUM(LLAMA_VOCAB_TYPE_UGM));
+  rb_define_const(rb_mLLaMACpp, "LLAMA_VOCAB_TYPE_RWKV", INT2NUM(LLAMA_VOCAB_TYPE_RWKV));
 
   rb_define_const(rb_mLLaMACpp, "LLAMA_VOCAB_PRE_TYPE_DEFAULT", INT2NUM(LLAMA_VOCAB_PRE_TYPE_DEFAULT));
   rb_define_const(rb_mLLaMACpp, "LLAMA_VOCAB_PRE_TYPE_LLAMA3", INT2NUM(LLAMA_VOCAB_PRE_TYPE_LLAMA3));
@@ -3631,6 +3638,7 @@ extern "C" void Init_llama_cpp(void) {
   rb_define_const(rb_mLLaMACpp, "LLAMA_VOCAB_PRE_TYPE_CODESHELL", INT2NUM(LLAMA_VOCAB_PRE_TYPE_CODESHELL));
   rb_define_const(rb_mLLaMACpp, "LLAMA_VOCAB_PRE_TYPE_BLOOM", INT2NUM(LLAMA_VOCAB_PRE_TYPE_BLOOM));
   rb_define_const(rb_mLLaMACpp, "LLAMA_VOCAB_PRE_TYPE_GPT3_FINNISH", INT2NUM(LLAMA_VOCAB_PRE_TYPE_GPT3_FINNISH));
+  rb_define_const(rb_mLLaMACpp, "LLAMA_VOCAB_PRE_TYPE_EXAONE", INT2NUM(LLAMA_VOCAB_PRE_TYPE_EXAONE));
 
   rb_define_const(rb_mLLaMACpp, "LLAMA_TOKEN_TYPE_UNDEFINED", INT2NUM(LLAMA_TOKEN_TYPE_UNDEFINED));
   rb_define_const(rb_mLLaMACpp, "LLAMA_TOKEN_TYPE_NORMAL", INT2NUM(LLAMA_TOKEN_TYPE_NORMAL));
@@ -3683,6 +3691,8 @@ extern "C" void Init_llama_cpp(void) {
   rb_define_const(rb_mLLaMACpp, "LLAMA_FTYPE_MOSTLY_Q4_0_4_4", INT2NUM(LLAMA_FTYPE_MOSTLY_Q4_0_4_4));
   rb_define_const(rb_mLLaMACpp, "LLAMA_FTYPE_MOSTLY_Q4_0_4_8", INT2NUM(LLAMA_FTYPE_MOSTLY_Q4_0_4_8));
   rb_define_const(rb_mLLaMACpp, "LLAMA_FTYPE_MOSTLY_Q4_0_8_8", INT2NUM(LLAMA_FTYPE_MOSTLY_Q4_0_8_8));
+  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_GUESSED", INT2NUM(LLAMA_FTYPE_GUESSED));
 

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 = '0.17.7'
+  VERSION = '0.17.10'
 
   # The supported version of llama.cpp.
-  LLAMA_CPP_VERSION = 'b3590'
+  LLAMA_CPP_VERSION = 'b3676'
 end

data/sig/llama_cpp.rbs

@@ -16,6 +16,7 @@ module LLaMACpp
   LLAMA_VOCAB_TYPE_BPE: Integer
   LLAMA_VOCAB_TYPE_WPM: Integer
   LLAMA_VOCAB_TYPE_UGM: Integer
+  LLAMA_VOCAB_TYPE_RWKV: Integer
 
   LLAMA_VOCAB_PRE_TYPE_DEFAULT: Integer
   LLAMA_VOCAB_PRE_TYPE_LLAMA3: Integer
@@ -42,6 +43,7 @@ module LLaMACpp
   LLAMA_VOCAB_PRE_TYPE_CODESHELL: Integer
   LLAMA_VOCAB_PRE_TYPE_BLOOM: Integer
   LLAMA_VOCAB_PRE_TYPE_GPT3_FINNISH: Integer
+  LLAMA_VOCAB_PRE_TYPE_EXAONE: Integer
 
   LLAMA_TOKEN_ATTR_UNDEFINED: Integer
   LLAMA_TOKEN_ATTR_UNKNOWN: Integer
@@ -86,6 +88,8 @@ module LLaMACpp
   LLAMA_FTYPE_MOSTLY_Q4_0_4_4: Integer
   LLAMA_FTYPE_MOSTLY_Q4_0_4_8: Integer
   LLAMA_FTYPE_MOSTLY_Q4_0_8_8: Integer
+  LLAMA_FTYPE_MOSTLY_TQ1_0: Integer
+  LLAMA_FTYPE_MOSTLY_TQ2_0: Integer
   LLAMA_FTYPE_GUESSED: Integer
 
   LLAMA_KV_OVERRIDE_TYPE_INT: Integer
@@ -197,6 +201,7 @@ module LLaMACpp
     def has_encoder?: () -> bool
     def has_decoder?: () -> bool
     def decoder_start_token: () -> Integer
+    def is_recurrent?: () -> bool
     def detokenize: (Array[Integer], ?remove_special: bool, ?unparse_special: bool) -> String
   end
 

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: llama_cpp
 version: !ruby/object:Gem::Version
-  version: 0.17.7
+  version: 0.17.10
 platform: ruby
 authors:
 - yoshoku
 autorequire:
 bindir: exe
 cert_chain: []
-date: 2024-08-17 00:00:00.000000000 Z
+date: 2024-09-07 00:00:00.000000000 Z
 dependencies: []
 description: llama_cpp.rb provides Ruby bindings for the llama.cpp.
 email: