llama_cpp 0.6.0 → 0.7.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 854493444a65cd1239649b991c8e6538c542c02a052932f6a69c56c984e28f58
-  data.tar.gz: 4e0b70de25eb2661b693af0d488efd25f570c3f62d4b9044fdd5c14fb5b9fac6
+  metadata.gz: 144a7130adb5ac32d31699bce809a6de6c3a6ecf8cfccca36ebdee436c28b645
+  data.tar.gz: d00b2c2db583e6e38d472033c7348f22e9614febdb633c4e454ca49e00d2fec6
 SHA512:
-  metadata.gz: b2524b8eb6e8568116f3c33eb57b764044083ee2ff2bbb7f15fc6301b024197ea8fca75968535b302a9e70449c9f9f28e0760cf4bfefb00ed8137c18e84137d5
-  data.tar.gz: faf26b552a8a862a97129b5bd25e05b3ae3edd2f8b118622b119634e4b004c05d200653c40085e4a28243c8994c517699baa35d3a8096ad8ac598fd637cf0565
+  metadata.gz: 2c30854fef304e0258250d9285bac8ab3ea014950d1638e88682029763a3e90eae36da1b3757b2441ff5a7a798401ee1e731bcfc014e7e651811726d7afea224
+  data.tar.gz: 10ea5bb5bf5d85a7e7030b514e2eb38650e9ce8a97ab339f63538b637d3c85293b406fea66c055a00f919c457a9a2af5c8f5710d0d31d702fe7e6f703b52933d

data/CHANGELOG.md

@@ -1,3 +1,11 @@
+## [[0.7.0](https://github.com/yoshoku/llama_cpp.rb/compare/v0.6.0...v0.7.0)] - 2023-10-07
+
+- Bump bundled llama.cpp from b1292 to b1334.
+- Refactor `generate` module function.
+
+**Breaking Changes**
+- Change to return UTF-8 String on `token_to_piece` and `desc` methods in `Model` and `text` method in `Context`.
+
 ## [[0.6.0](https://github.com/yoshoku/llama_cpp.rb/compare/v0.5.3...v0.6.0)] - 2023-09-30
 
 **Breaking Changes**

data/ext/llama_cpp/llama_cpp.cpp

@@ -1127,6 +1127,7 @@ public:
     rb_define_method(rb_cLLaMAModel, "n_vocab", RUBY_METHOD_FUNC(_llama_model_get_model_n_vocab), 0);
     rb_define_method(rb_cLLaMAModel, "n_ctx_train", RUBY_METHOD_FUNC(_llama_model_get_model_n_ctx_train), 0);
     rb_define_method(rb_cLLaMAModel, "n_embd", RUBY_METHOD_FUNC(_llama_model_get_model_n_embd), 0);
+    rb_define_method(rb_cLLaMAModel, "rope_freq_scale_train", RUBY_METHOD_FUNC(_llama_model_rope_freq_scale_train), 0);
     rb_define_method(rb_cLLaMAModel, "token_to_piece", RUBY_METHOD_FUNC(_llama_model_token_to_piece), 1);
     rb_define_method(rb_cLLaMAModel, "tokenize", RUBY_METHOD_FUNC(_llama_model_tokenize), -1);
     rb_define_method(rb_cLLaMAModel, "desc", RUBY_METHOD_FUNC(_llama_model_get_model_desc), 0);
@@ -1288,6 +1289,11 @@ private:
     return INT2NUM(llama_n_embd(ptr->model));
   }
 
+  static VALUE _llama_model_rope_freq_scale_train(VALUE self) {
+    LLaMAModelWrapper* ptr = get_llama_model(self);
+    return DBL2NUM(llama_rope_freq_scale_train(ptr->model));
+  }
+
   static VALUE _llama_model_token_to_piece(VALUE self, VALUE token_) {
     if (!RB_INTEGER_TYPE_P(token_)) {
       rb_raise(rb_eArgError, "token must be an integer");
@@ -1308,7 +1314,7 @@ private:
       result.resize(n_tokens);
     }
     std::string ret(result.data(), result.size());
-    return rb_str_new_cstr(ret.c_str());
+    return rb_utf8_str_new_cstr(ret.c_str());
   }
 
   static VALUE _llama_model_tokenize(int argc, VALUE* argv, VALUE self) {
@@ -1358,7 +1364,7 @@ private:
     LLaMAModelWrapper* ptr = get_llama_model(self);
     char buf[128];
     llama_model_desc(ptr->model, buf, sizeof(buf));
-    return rb_str_new_cstr(buf);
+    return rb_utf8_str_new_cstr(buf);
   }
 
   static VALUE _llama_model_get_model_size(VALUE self) {
@@ -1650,6 +1656,10 @@ public:
     rb_define_method(rb_cLLaMAContext, "token_bos", RUBY_METHOD_FUNC(_llama_context_token_bos), 0);
     rb_define_method(rb_cLLaMAContext, "token_eos", RUBY_METHOD_FUNC(_llama_context_token_eos), 0);
     rb_define_method(rb_cLLaMAContext, "token_nl", RUBY_METHOD_FUNC(_llama_context_token_nl), 0);
+    rb_define_method(rb_cLLaMAContext, "token_prefix", RUBY_METHOD_FUNC(_llama_context_token_prefix), 0);
+    rb_define_method(rb_cLLaMAContext, "token_middle", RUBY_METHOD_FUNC(_llama_context_token_middle), 0);
+    rb_define_method(rb_cLLaMAContext, "token_suffix", RUBY_METHOD_FUNC(_llama_context_token_suffix), 0);
+    rb_define_method(rb_cLLaMAContext, "token_eot", RUBY_METHOD_FUNC(_llama_context_token_eot), 0);
     rb_define_method(rb_cLLaMAContext, "n_ctx", RUBY_METHOD_FUNC(_llama_context_n_ctx), 0);
     rb_define_method(rb_cLLaMAContext, "timings", RUBY_METHOD_FUNC(_llama_context_get_timings), 0);
     rb_define_method(rb_cLLaMAContext, "print_timings", RUBY_METHOD_FUNC(_llama_context_print_timings), 0);
@@ -1905,7 +1915,7 @@ private:
     }
     const llama_token token = NUM2INT(token_);
     const char* text = llama_token_get_text(ptr->ctx, token);
-    return rb_str_new_cstr(text);
+    return rb_utf8_str_new_cstr(text);
   }
 
   static VALUE _llama_context_score(VALUE self, VALUE token_) {
@@ -1957,6 +1967,42 @@ private:
     return INT2NUM(llama_token_nl(ptr->ctx));
   }
 
+  static VALUE _llama_context_token_prefix(VALUE self) {
+    LLaMAContextWrapper* ptr = get_llama_context(self);
+    if (ptr->ctx == NULL) {
+      rb_raise(rb_eRuntimeError, "LLaMA context is not initialized");
+      return Qnil;
+    }
+    return INT2NUM(llama_token_prefix(ptr->ctx));
+  }
+
+  static VALUE _llama_context_token_middle(VALUE self) {
+    LLaMAContextWrapper* ptr = get_llama_context(self);
+    if (ptr->ctx == NULL) {
+      rb_raise(rb_eRuntimeError, "LLaMA context is not initialized");
+      return Qnil;
+    }
+    return INT2NUM(llama_token_middle(ptr->ctx));
+  }
+
+  static VALUE _llama_context_token_suffix(VALUE self) {
+    LLaMAContextWrapper* ptr = get_llama_context(self);
+    if (ptr->ctx == NULL) {
+      rb_raise(rb_eRuntimeError, "LLaMA context is not initialized");
+      return Qnil;
+    }
+    return INT2NUM(llama_token_suffix(ptr->ctx));
+  }
+
+  static VALUE _llama_context_token_eot(VALUE self) {
+    LLaMAContextWrapper* ptr = get_llama_context(self);
+    if (ptr->ctx == NULL) {
+      rb_raise(rb_eRuntimeError, "LLaMA context is not initialized");
+      return Qnil;
+    }
+    return INT2NUM(llama_token_eot(ptr->ctx));
+  }
+
   static VALUE _llama_context_n_ctx(VALUE self) {
     LLaMAContextWrapper* ptr = get_llama_context(self);
     if (ptr->ctx == NULL) {