llama_cpp 0.12.2 → 0.12.4

data/vendor/tmp/llama.cpp/llama.h

@@ -2,12 +2,8 @@
 #define LLAMA_H
 
 #include "ggml.h"
-#ifdef GGML_USE_CUBLAS
-#include "ggml-cuda.h"
-#define LLAMA_MAX_DEVICES GGML_CUDA_MAX_DEVICES
-#else
-#define LLAMA_MAX_DEVICES 1
-#endif // GGML_USE_CUBLAS
+#include "ggml-backend.h"
+
 #include <stddef.h>
 #include <stdint.h>
 #include <stdio.h>
@@ -45,11 +41,6 @@
 #define LLAMA_SESSION_MAGIC   LLAMA_FILE_MAGIC_GGSN
 #define LLAMA_SESSION_VERSION 4
 
-#if defined(GGML_USE_CUBLAS) || defined(GGML_USE_CLBLAST) || defined(GGML_USE_METAL)
-// Defined when llama.cpp is compiled with support for offloading model layers to GPU.
-#define LLAMA_SUPPORTS_GPU_OFFLOAD
-#endif
-
 #ifdef __cplusplus
 extern "C" {
 #endif
@@ -106,6 +97,8 @@ extern "C" {
         LLAMA_FTYPE_MOSTLY_IQ2_XXS       = 19, // except 1d tensors
         LLAMA_FTYPE_MOSTLY_IQ2_XS        = 20, // except 1d tensors
         LLAMA_FTYPE_MOSTLY_Q2_K_S        = 21, // except 1d tensors
+        LLAMA_FTYPE_MOSTLY_Q3_K_XS       = 22, // except 1d tensors
+        LLAMA_FTYPE_MOSTLY_IQ3_XXS       = 23, // except 1d tensors
 
         LLAMA_FTYPE_GUESSED = 1024, // not specified in the model file
     };
@@ -194,7 +187,7 @@ extern "C" {
         // LLAMA_SPLIT_LAYER: ignored
         int32_t main_gpu;
 
-        // proportion of the model (layers or rows) to offload to each GPU, size: LLAMA_MAX_DEVICES
+        // proportion of the model (layers or rows) to offload to each GPU, size: llama_max_devices()
         const float * tensor_split;
 
         // Called with a progress value between 0.0 and 1.0. Pass NULL to disable.
@@ -231,6 +224,9 @@ extern "C" {
         float    yarn_beta_slow;   // YaRN high correction dim
         uint32_t yarn_orig_ctx;    // YaRN original context size
 
+        ggml_backend_sched_eval_callback cb_eval;
+        void * cb_eval_user_data;
+
         enum ggml_type type_k; // data type for K cache
         enum ggml_type type_v; // data type for V cache
 
@@ -328,9 +324,14 @@ extern "C" {
 
     LLAMA_API int64_t llama_time_us(void);
 
-    LLAMA_API int32_t  llama_max_devices(void);
-    LLAMA_API bool llama_mmap_supported (void);
-    LLAMA_API bool llama_mlock_supported(void);
+    LLAMA_API size_t llama_max_devices(void);
+
+    LLAMA_API bool llama_supports_mmap       (void);
+    LLAMA_API bool llama_supports_mlock      (void);
+    LLAMA_API bool llama_supports_gpu_offload(void);
+
+    LLAMA_API DEPRECATED(bool llama_mmap_supported (void), "use llama_supports_mmap() instead");
+    LLAMA_API DEPRECATED(bool llama_mlock_supported(void), "use llama_supports_mlock() instead");
 
     LLAMA_API const struct llama_model * llama_get_model(const struct llama_context * ctx);
 
@@ -770,6 +771,14 @@ extern "C" {
                            float   p,
                           size_t   min_keep);
 
+    /// @details Dynamic temperature implementation described in the paper https://arxiv.org/abs/2309.02772.
+    LLAMA_API void llama_sample_entropy(
+            struct llama_context * ctx,
+          llama_token_data_array * candidates_p,
+                           float   min_temp,
+                           float   max_temp,
+                           float   exponent_val);
+
     LLAMA_API void llama_sample_temp(
             struct llama_context * ctx,
           llama_token_data_array * candidates,

data/vendor/tmp/llama.cpp/unicode.h

@@ -2,8 +2,9 @@
 
 #include <cassert>
 #include <stdexcept>
-#include <vector>
+#include <string>
 #include <unordered_map>
+#include <vector>
 
 static const std::vector<std::pair<uint32_t, uint32_t>> digit_ranges = {
 {0x30, 0x39}, {0xB2, 0xB3}, {0xB9, 0xB9}, {0x660, 0x669}, {0x6F0, 0x6F9}, {0x7C0, 0x7C9}, {0x966, 0x96F}, {0x9E6, 0x9EF}, {0xA66, 0xA6F}, {0xAE6, 0xAEF}, {0xB66, 0xB6F}, {0xBE6, 0xBEF}, {0xC66, 0xC6F},

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: llama_cpp
 version: !ruby/object:Gem::Version
-  version: 0.12.2
+  version: 0.12.4
 platform: ruby
 authors:
 - yoshoku
 autorequire:
 bindir: exe
 cert_chain: []
-date: 2024-01-20 00:00:00.000000000 Z
+date: 2024-02-03 00:00:00.000000000 Z
 dependencies: []
 description: llama_cpp.rb provides Ruby bindings for the llama.cpp.
 email:
@@ -45,6 +45,8 @@ files:
 - vendor/tmp/llama.cpp/ggml-cuda.cu
 - vendor/tmp/llama.cpp/ggml-cuda.h
 - vendor/tmp/llama.cpp/ggml-impl.h
+- vendor/tmp/llama.cpp/ggml-kompute.cpp
+- vendor/tmp/llama.cpp/ggml-kompute.h
 - vendor/tmp/llama.cpp/ggml-metal.h
 - vendor/tmp/llama.cpp/ggml-metal.m
 - vendor/tmp/llama.cpp/ggml-metal.metal
@@ -54,6 +56,11 @@ files:
 - vendor/tmp/llama.cpp/ggml-opencl.h
 - vendor/tmp/llama.cpp/ggml-quants.c
 - vendor/tmp/llama.cpp/ggml-quants.h
+- vendor/tmp/llama.cpp/ggml-sycl.cpp
+- vendor/tmp/llama.cpp/ggml-sycl.h
+- vendor/tmp/llama.cpp/ggml-vulkan-shaders.hpp
+- vendor/tmp/llama.cpp/ggml-vulkan.cpp
+- vendor/tmp/llama.cpp/ggml-vulkan.h
 - vendor/tmp/llama.cpp/ggml.c
 - vendor/tmp/llama.cpp/ggml.h
 - vendor/tmp/llama.cpp/llama.cpp
@@ -84,7 +91,7 @@ required_rubygems_version: !ruby/object:Gem::Requirement
     - !ruby/object:Gem::Version
       version: '0'
 requirements: []
-rubygems_version: 3.5.3
+rubygems_version: 3.4.19
 signing_key:
 specification_version: 4
 summary: Ruby bindings for the llama.cpp.