llama_cpp 0.0.1 → 0.0.3

data/ext/llama_cpp/src/llama.h

@@ -6,7 +6,7 @@
 #include <stdbool.h>
 
 #ifdef LLAMA_SHARED
-#    ifdef _WIN32
+#    if defined(_WIN32) && !defined(__MINGW32__)
 #        ifdef LLAMA_BUILD
 #            define LLAMA_API __declspec(dllexport)
 #        else
@@ -20,7 +20,7 @@
 #endif
 
 #define LLAMA_FILE_VERSION 1
-#define LLAMA_FILE_MAGIC 0x67676d66 // 'ggmf' in hex
+#define LLAMA_FILE_MAGIC 0x67676a74 // 'ggjt' in hex
 #define LLAMA_FILE_MAGIC_UNVERSIONED 0x67676d6c // pre-versioned files
 
 #ifdef __cplusplus
@@ -83,6 +83,23 @@ extern "C" {
             const char * fname_out,
                    int   itype);
 
+    // Returns the KV cache that will contain the context for the
+    // ongoing prediction with the model.
+    LLAMA_API const uint8_t * llama_get_kv_cache(struct llama_context * ctx);
+
+    // Returns the size of the KV cache
+    LLAMA_API size_t llama_get_kv_cache_size(struct llama_context * ctx);
+
+    // Returns the number of tokens in the KV cache
+    LLAMA_API int llama_get_kv_cache_token_count(struct llama_context * ctx);
+
+    // Sets the KV cache containing the current context for the model
+    LLAMA_API void llama_set_kv_cache(
+            struct llama_context * ctx,
+                   const uint8_t * kv_cache,
+                          size_t   n_size,
+                             int   n_token_count);
+
     // Run the llama inference to obtain the logits and probabilities for the next token.
     // tokens + n_tokens is the provided batch of new tokens to process
     // n_past is the number of tokens to use from previous eval calls

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 = '0.0.1'
+  VERSION = '0.0.3'
 
   # The version of llama.cpp bundled with llama_cpp.rb.
-  LLAMA_CPP_VERSION = 'master-2a98bc1'
+  LLAMA_CPP_VERSION = 'master-698f7b5'
 end

data/lib/llama_cpp.rb

@@ -5,14 +5,18 @@ require_relative 'llama_cpp/llama_cpp'
 
 # llama_cpp.rb provides Ruby bindings for the llama.cpp.
 module LLaMACpp
+  # Class alias to match interface of whispercpp gem.
+  Params = ContextParams
+
   module_function
 
   # Generates sentences following the given prompt for operation check.
   #
   # @param context [LLaMACpp::Context]
   # @param prompt [String]
+  # @param n_threads [Integer]
   # @return [String]
-  def generate(context, prompt) # rubocop:disable Metrics/AbcSize, Metrics/MethodLength, Metrics/PerceivedComplexity
+  def generate(context, prompt, n_threads: 1) # rubocop:disable Metrics/AbcSize, Metrics/MethodLength, Metrics/PerceivedComplexity
     prompt.insert(0, ' ')
 
     embd_input = context.tokenize(text: prompt, add_bos: true)
@@ -36,7 +40,7 @@ module LLaMACpp
           embd.insert(0, last_n_tokens[(n_ctx - (n_left / 2) - embd.size)...-embd.size])
         end
 
-        context.eval(tokens: embd, n_past: n_past)
+        context.eval(tokens: embd, n_past: n_past, n_threads: n_threads)
       end
 
       n_past += embd.size

data/sig/llama_cpp.rbs

@@ -0,0 +1,52 @@
+module LLaMACpp
+  VERSION: String
+  LLAMA_CPP_VERSION: String
+  LLAMA_FILE_VERSION: String
+  LLAMA_FILE_MAGIC: String
+  LLAMA_FILE_MAGIC_UNVERSIONED: String
+
+  def self?.generate: (::LLaMACpp::Context, String, ?n_threads: Integer) -> String
+  def self?.print_system_info: () -> void
+  def self?.token_bos: () -> Integer
+  def self?.token_eos: () -> Integer
+
+  class Context
+    public
+
+    def initialize: (model_path: String, params: ::LLaMACpp::ContextParams) -> void
+    def embeddings: () -> Array[Float]
+    def eval: (tokens: Array[Integer], n_past: Integer, ?n_tokens: Integer, ?n_threads: Integer) -> Qnil
+    def logits: () -> Array[Float]
+    def n_ctx: () -> Integer
+    def n_embd: () -> Integer
+    def n_vocab: () -> Integer
+    def print_timings: () -> void
+    def reset_timings: () -> void
+    def sample_top_p_top_k: (top_k: Integer, top_p: Float, temp: Float, penalty: Float) -> Integer
+    def token_to_str: (Integer) -> String
+    def tokenize: (text: String, ?n_max_tokens: Integer, ?add_bos: bool) -> Array[Integer]
+  end
+
+  class ContextParams
+    public
+
+    def embedding: () -> bool
+    def embedding=: (bool) -> bool
+    def f16_kv: () -> bool
+    def f16_kv=: (bool) -> bool
+    def logits_all: () -> bool
+    def logits_all=: (bool) -> bool
+    def n_ctx: () -> Integer
+    def n_ctx=: (Integer) -> Integer
+    def n_parts: () -> Integer
+    def n_parts=: (Integer) -> Integer
+    def seed: () -> Integer
+    def seed=: (Integer) -> Integer
+    def use_mlock: () -> bool
+    def use_mlock=: (bool) -> bool
+    def vocab_only: () -> bool
+    def vocab_only=: (bool) -> bool
+  end
+
+  class Params = ContextParams
+end

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: llama_cpp
 version: !ruby/object:Gem::Version
-  version: 0.0.1
+  version: 0.0.3
 platform: ruby
 authors:
 - yoshoku
 autorequire:
 bindir: exe
 cert_chain: []
-date: 2023-04-02 00:00:00.000000000 Z
+date: 2023-04-08 00:00:00.000000000 Z
 dependencies: []
 description: llama_cpp.rb provides Ruby bindings for the llama.cpp.
 email:
@@ -32,6 +32,7 @@ files:
 - ext/llama_cpp/src/llama.h
 - lib/llama_cpp.rb
 - lib/llama_cpp/version.rb
+- sig/llama_cpp.rbs
 homepage: https://github.com/yoshoku/llama_cpp.rb
 licenses:
 - MIT