llama_cpp 0.10.0 → 0.10.2

data/ext/llama_cpp/src/llama.h

@@ -39,6 +39,7 @@
 
 #define LLAMA_MAX_RNG_STATE (64*1024)
 
+#define LLAMA_FILE_MAGIC_GGLA 0x67676c61u // 'ggla'
 #define LLAMA_FILE_MAGIC_GGSN 0x6767736eu // 'ggsn'
 
 #define LLAMA_SESSION_MAGIC   LLAMA_FILE_MAGIC_GGSN
@@ -126,7 +127,7 @@ extern "C" {
         bool sorted;
     } llama_token_data_array;
 
-    typedef void (*llama_progress_callback)(float progress, void *ctx);
+    typedef bool (*llama_progress_callback)(float progress, void *ctx);
 
     // Input data for llama_decode
     // A llama_batch object can contain input about one or many sequences
@@ -179,7 +180,9 @@ extern "C" {
         int32_t main_gpu;     // the GPU that is used for scratch and small tensors
         const float * tensor_split; // how to split layers across multiple GPUs (size: LLAMA_MAX_DEVICES)
 
-        // called with a progress value between 0 and 1, pass NULL to disable
+        // Called with a progress value between 0.0 and 1.0. Pass NULL to disable.
+        // If the provided progress_callback returns true, model loading continues.
+        // If it returns false, model loading is immediately aborted.
         llama_progress_callback progress_callback;
 
         // context pointer passed to the progress callback
@@ -216,7 +219,7 @@ extern "C" {
 
         // Keep the booleans together to avoid misalignment during copy-by-value.
         bool mul_mat_q;   // if true, use experimental mul_mat_q kernels (DEPRECATED - always true)
-        bool logits_all;  // the llama_eval() call computes all logits, not just the last one
+        bool logits_all;  // the llama_eval() call computes all logits, not just the last one (DEPRECATED - set llama_batch.logits instead)
         bool embedding;   // embedding mode only
         bool offload_kqv; // whether to offload the KQV ops (including the KV cache) to GPU
     };
@@ -313,7 +316,9 @@ extern "C" {
 
     LLAMA_API const struct llama_model * llama_get_model(const struct llama_context * ctx);
 
-    LLAMA_API int llama_n_ctx      (const struct llama_context * ctx);
+    // TODO: become more consistent with returned int types across the API
+    LLAMA_API uint32_t llama_n_ctx      (const struct llama_context * ctx);
+    LLAMA_API uint32_t llama_n_batch    (const struct llama_context * ctx);
 
     LLAMA_API enum llama_vocab_type llama_vocab_type(const struct llama_model * model);
 

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 = '0.10.0'
+  VERSION = '0.10.2'
 
   # The version of llama.cpp bundled with llama_cpp.rb.
-  LLAMA_CPP_VERSION = 'b1620'
+  LLAMA_CPP_VERSION = 'b1686'
 end

data/sig/llama_cpp.rbs

@@ -256,6 +256,8 @@ module LLaMACpp
   class ModelQuantizeParams
     public
 
+    attr_reader params: ::LLaMACpp::ModelParams
+
     def n_thread: () -> Integer
     def n_thread=: (Integer) -> Integer
     def ftype: () -> Integer

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: llama_cpp
 version: !ruby/object:Gem::Version
-  version: 0.10.0
+  version: 0.10.2
 platform: ruby
 authors:
 - yoshoku
 autorequire:
 bindir: exe
 cert_chain: []
-date: 2023-12-09 00:00:00.000000000 Z
+date: 2023-12-23 00:00:00.000000000 Z
 dependencies: []
 description: llama_cpp.rb provides Ruby bindings for the llama.cpp.
 email: