llama_cpp 0.3.5 → 0.3.6

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 991df3df6b16ec98a203a6c6565794988eec04697ccb963faab976f436e1bfcc
-  data.tar.gz: dafd3b8274640eb79353e11056f497a02392fef332a46dcc1717878c836f62bd
+  metadata.gz: 545786d4c9308ffe0f7e214a12427beaea0b26bec915ff84b16eed25ef1932a4
+  data.tar.gz: aaa0d4fc1710b13a26163306c8b51e423233c2f7e4b3d6127f94c9b6c4846f9c
 SHA512:
-  metadata.gz: 0a54fdf18c5be5273f01d4d991b59975a8e8b6a0a8f54087fb90df3f1a8a7ebad557d01fb119f3d61cafa8f6c59fb81641624779fda27b732eea2868cb4642e8
-  data.tar.gz: 6574064078070502e36ad933bd5efb2f479c94f47a1260286fe485e48d35a7b3985274943c6163189326891349b47b1d9815d77c600b015fe111cc3842179392
+  metadata.gz: 12b3ac122fd7ea59b51e2d6ff905ed1a71cf8a8b3650a269d4a3793ae32a0149f6836a792c8f216d0fdb0c39aeb3b47914e73ffc74b574bbe686660e6be84ea1
+  data.tar.gz: 5056b95552f3434692a6c19653810d77bb28ddf9b28abd78712ccfb4ee4f7d836a5d54e283513fcfc617cc79ffa7bb9257d4ac2b6d96ec89158bf94acd4cec86

data/CHANGELOG.md

@@ -1,3 +1,8 @@
+## [[0.3.6](https://github.com/yoshoku/llama_cpp.rb/compare/v0.3.5...v0.3.6)] - 2023-08-04
+
+- Bump bundled llama.cpp from master-1a94186 to master-468ea24.
+  - Add `mul_mat_q` option to ContextParams.
+
 ## [[0.3.5](https://github.com/yoshoku/llama_cpp.rb/compare/v0.3.4...v0.3.5)] - 2023-07-29
 
 - Bump bundled llama.cpp from master-d924522 to master-1a94186.

data/README.md

@@ -12,11 +12,27 @@ This gem is still under development and may undergo many changes in the future.
 
 Install the gem and add to the application's Gemfile by executing:
 
-    $ bundle add llama_cpp
+```sh
+$ bundle add llama_cpp
+```
 
 If bundler is not being used to manage dependencies, install the gem by executing:
 
-    $ gem install llama_cpp
+```sh
+$ gem install llama_cpp
+```
+
+There are several installation options for improving execution performance:
+
+```sh
+# use OpenBLAS
+$ gem install llama_cpp -- --with-openblas
+
+# use Metal on macOS
+$ gem install llama_cpp -- --with-metal
+```
+
+Those options are defined in [extconf.rb](https://github.com/yoshoku/llama_cpp.rb/blob/main/ext/llama_cpp/extconf.rb) by with_config method.
 
 ## Usage
 

data/ext/llama_cpp/extconf.rb

@@ -5,7 +5,7 @@ require 'fileutils'
 
 abort 'libstdc++ is not found.' unless have_library('stdc++')
 
-$srcs = %w[ggml.c llama.cpp llama_cpp.cpp]
+$srcs = %w[ggml.c ggml-alloc.c llama.cpp llama_cpp.cpp]
 $srcs << 'ggml-opencl.cpp' if with_config('clblast')
 $srcs << 'ggml-mpi.c' if with_config('mpi')
 $CFLAGS << ' -w -DNDEBUG'

data/ext/llama_cpp/llama_cpp.cpp

@@ -414,6 +414,8 @@ public:
     rb_define_method(rb_cLLaMAContextParams, "rope_freq_scale", RUBY_METHOD_FUNC(_llama_context_params_get_rope_freq_scale), 0);
     rb_define_method(rb_cLLaMAContextParams, "low_vram=", RUBY_METHOD_FUNC(_llama_context_params_set_low_vram), 1);
     rb_define_method(rb_cLLaMAContextParams, "low_vram", RUBY_METHOD_FUNC(_llama_context_params_get_low_vram), 0);
+    rb_define_method(rb_cLLaMAContextParams, "mul_mat_q=", RUBY_METHOD_FUNC(_llama_context_params_set_mul_mat_q), 1);
+    rb_define_method(rb_cLLaMAContextParams, "mul_mat_q", RUBY_METHOD_FUNC(_llama_context_params_get_mul_mat_q), 0);
     rb_define_method(rb_cLLaMAContextParams, "seed=", RUBY_METHOD_FUNC(_llama_context_params_set_seed), 1);
     rb_define_method(rb_cLLaMAContextParams, "seed", RUBY_METHOD_FUNC(_llama_context_params_get_seed), 0);
     rb_define_method(rb_cLLaMAContextParams, "f16_kv=", RUBY_METHOD_FUNC(_llama_context_params_set_f16_kv), 1);
@@ -527,7 +529,7 @@ private:
   // low_vram
   static VALUE _llama_context_params_set_low_vram(VALUE self, VALUE low_vram) {
     LLaMAContextParamsWrapper* ptr = get_llama_context_params(self);
-    ptr->params.low_vram = low_vram == Qtrue ? true : false;
+    ptr->params.low_vram = RTEST(low_vram) ? true : false;
     return ptr->params.low_vram ? Qtrue : Qfalse;
   }
 
@@ -536,6 +538,18 @@ private:
     return ptr->params.low_vram ? Qtrue : Qfalse;
   }
 
+  // mul_mat_q
+  static VALUE _llama_context_params_set_mul_mat_q(VALUE self, VALUE mul_mat_q) {
+    LLaMAContextParamsWrapper* ptr = get_llama_context_params(self);
+    ptr->params.mul_mat_q = RTEST(mul_mat_q) ? true : false;
+    return ptr->params.mul_mat_q ? Qtrue : Qfalse;
+  }
+
+  static VALUE _llama_context_params_get_mul_mat_q(VALUE self) {
+    LLaMAContextParamsWrapper* ptr = get_llama_context_params(self);
+    return ptr->params.mul_mat_q ? Qtrue : Qfalse;
+  }
+
   // seed
   static VALUE _llama_context_params_set_seed(VALUE self, VALUE seed) {
     LLaMAContextParamsWrapper* ptr = get_llama_context_params(self);
@@ -555,7 +569,7 @@ private:
   // f16_kv
   static VALUE _llama_context_params_set_f16_kv(VALUE self, VALUE f16_kv) {
     LLaMAContextParamsWrapper* ptr = get_llama_context_params(self);
-    ptr->params.f16_kv = f16_kv == Qtrue ? true : false;
+    ptr->params.f16_kv = RTEST(f16_kv) ? true : false;
     return ptr->params.f16_kv ? Qtrue : Qfalse;
   }
 
@@ -567,7 +581,7 @@ private:
   // logits_all
   static VALUE _llama_context_params_set_logits_all(VALUE self, VALUE logits_all) {
     LLaMAContextParamsWrapper* ptr = get_llama_context_params(self);
-    ptr->params.logits_all = logits_all == Qtrue ? true : false;
+    ptr->params.logits_all = RTEST(logits_all) ? true : false;
     return ptr->params.logits_all ? Qtrue : Qfalse;
   }
 
@@ -579,7 +593,7 @@ private:
   // vocab_only
   static VALUE _llama_context_params_set_vocab_only(VALUE self, VALUE vocab_only) {
     LLaMAContextParamsWrapper* ptr = get_llama_context_params(self);
-    ptr->params.vocab_only = vocab_only == Qtrue ? true : false;
+    ptr->params.vocab_only = RTEST(vocab_only) ? true : false;
     return ptr->params.vocab_only ? Qtrue : Qfalse;
   }
 
@@ -591,7 +605,7 @@ private:
   // use_mmap
   static VALUE _llama_context_params_set_use_mmap(VALUE self, VALUE use_mmap) {
     LLaMAContextParamsWrapper* ptr = get_llama_context_params(self);
-    ptr->params.use_mmap = use_mmap == Qtrue ? true : false;
+    ptr->params.use_mmap = RTEST(use_mmap) ? true : false;
     return ptr->params.use_mmap ? Qtrue : Qfalse;
   }
 
@@ -603,7 +617,7 @@ private:
   // use_mlock
   static VALUE _llama_context_params_set_use_mlock(VALUE self, VALUE use_mlock) {
     LLaMAContextParamsWrapper* ptr = get_llama_context_params(self);
-    ptr->params.use_mlock = use_mlock == Qtrue ? true : false;
+    ptr->params.use_mlock = RTEST(use_mlock) ? true : false;
     return ptr->params.use_mlock ? Qtrue : Qfalse;
   }
 
@@ -615,7 +629,7 @@ private:
   // embedding
   static VALUE _llama_context_params_set_embedding(VALUE self, VALUE embedding) {
     LLaMAContextParamsWrapper* ptr = get_llama_context_params(self);
-    ptr->params.embedding = embedding == Qtrue ? true : false;
+    ptr->params.embedding = RTEST(embedding) ? true : false;
     return ptr->params.embedding ? Qtrue : Qfalse;
   }
 
@@ -2408,7 +2422,7 @@ static VALUE rb_llama_llama_backend_init(int argc, VALUE* argv, VALUE self) {
   rb_scan_args(argc, argv, ":", &kw_args);
   rb_get_kwargs(kw_args, kw_table, 0, 1, kw_values);
 
-  const bool numa = kw_values[0] == Qundef ? false : (RTEST ? true : false);
+  const bool numa = kw_values[0] == Qundef ? false : (RTEST(kw_values[0]) ? true : false);
   llama_backend_init(numa);
 
   return Qnil;