vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Multi GPU support, CUDA refactor, CUDA scratch buffer (#1703)

Browse source 
* CUDA multi GPU + scratch

ggml_cuda_compute_forward

Tensor parallelism

ggml_cuda_add

ggml_cuda_rms_norm

ggml_cuda_silu

CUDA scratch buffer

--main-gpu CLI option
This commit is contained in:
Johannes Gäßler 2023-06-06 21:33:23 +02:00 • committed by GitHub
parent 44f906e853
commit 17366df842
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
12 changed files with 1221 additions and 544 deletions
Download patch file Download diff file Expand all files Collapse all files

17
ggml-cuda.h
View file

@ -1,10 +1,19 @@
#pragma once
#include "ggml.h"
#ifdef __cplusplus
extern "C" {
#endif
#define GGML_CUDA_MAX_DEVICES 16
struct ggml_tensor_extra_gpu {
void * data_device[GGML_CUDA_MAX_DEVICES]; // 1 pointer for each device for split tensors
};
void ggml_init_cublas(void);
void ggml_cuda_set_tensor_split(const float * tensor_split);
void ggml_cuda_mul(const struct ggml_tensor * src0, const struct ggml_tensor * src1, struct ggml_tensor * dst);
bool ggml_cuda_can_mul_mat(const struct ggml_tensor * src0, const struct ggml_tensor * src1, struct ggml_tensor * dst);
@ -15,8 +24,12 @@ void ggml_cuda_mul_mat(const struct ggml_tensor * src0, const struct ggml_tens
void * ggml_cuda_host_malloc(size_t size);
void ggml_cuda_host_free(void * ptr);
void ggml_cuda_transform_tensor(struct ggml_tensor * tensor);
void ggml_cuda_load_data(const char * fname, struct ggml_tensor * tensors, size_t offset);
void ggml_cuda_load_data(const char * fname, struct ggml_tensor * tensors, size_t offset);
void ggml_cuda_free_data(struct ggml_tensor * tensor);
void ggml_cuda_assign_buffers(struct ggml_tensor * tensor);
void ggml_cuda_set_main_device(int main_device);
void ggml_cuda_set_scratch_size(size_t scratch_size);
bool ggml_cuda_compute_forward(struct ggml_compute_params * params, struct ggml_tensor * tensor);
#ifdef __cplusplus
}