haraldwolff
/
whisper.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

talk-llama : sync llama.cpp

pull/1768/merge
Georgi Gerganov 2024-02-10 10:10:59 +02:00
parent 518199c09e
commit 02b4c52c12
No known key found for this signature in database
GPG Key ID: 449E073F9DC10735
2 changed files with 699 additions and 208 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

876
examples/talk-llama/llama.cpp
View File

File diff suppressed because it is too large Load Diff

31
examples/talk-llama/llama.h
View File

@ -3,15 +3,7 @@
#include "ggml.h"
#include "ggml-backend.h"
#ifdef GGML_USE_CUBLAS
#include "ggml-cuda.h"
#define LLAMA_MAX_DEVICES GGML_CUDA_MAX_DEVICES
#elif defined(GGML_USE_SYCL)
#include "ggml-sycl.h"
#define LLAMA_MAX_DEVICES GGML_SYCL_MAX_DEVICES
#else
#define LLAMA_MAX_DEVICES 1
#endif // GGML_USE_CUBLAS
#include <stddef.h>
#include <stdint.h>
#include <stdio.h>
@ -49,11 +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(GGML_USE_VULKAN) || defined(GGML_USE_SYCL)
// 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
@ -111,6 +98,7 @@ extern "C" {
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
};
@ -199,7 +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.
@ -225,7 +213,7 @@ extern "C" {
uint32_t n_batch; // prompt processing maximum batch size
uint32_t n_threads; // number of threads to use for generation
uint32_t n_threads_batch; // number of threads to use for batch processing
int8_t rope_scaling_type; // RoPE scaling type, from `enum llama_rope_scaling_type`
int32_t rope_scaling_type; // RoPE scaling type, from `enum llama_rope_scaling_type`
// ref: https://github.com/ggerganov/llama.cpp/pull/2054
float rope_freq_base; // RoPE base frequency, 0 = from model
@ -336,9 +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);