root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-26 11:24:35 +00:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

Vulkan Improvements (#5835)

Browse Source 
* Improve dequant shaders, add fast q4_0 dequant

* Optimize dmmv non-kquants for GCN

Remove unnecessary SPIR-V shader duplication

* Fix q4_0 dequant dispatch sizes

Fix backend free bug

* Optimize dequant shaders for q4_1, q5_0, q5_1 and q8_0

* Add unary and binary op shader templates

* Fix Vulkan check results

* Enable non-contiguous support for simple ops

* Add argsort

Basic q4_0 mmq shader and unit test

* Speed up q4_0 dequant code, enable mmq for q4_0

* Rework matmul pipeline selection

* Add soft_max alibi support

* Add q4_1, q5_0, q5_1 and q8_0 dequant mat mat mul shaders

* Add environment variable GGML_VK_FORCE_MAX_ALLOCATION_SIZE to limit max buffer size

Rename GGML_VULKAN_DISABLE_F16 to GGML_VK_DISABLE_F16 for consistency
This commit is contained in:
0cc4m 2024-03-05 13:33:42 +01:00 committed by GitHub
parent 21b0867433
commit 61d1c88e15
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
5 changed files with 43476 additions and 46795 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

86931
ggml-vulkan-shaders.hpp
View File

File diff suppressed because it is too large Load Diff

2128
ggml-vulkan.cpp
View File

File diff suppressed because it is too large Load Diff

1
ggml-vulkan.h
View File

@ -10,6 +10,7 @@ extern "C" {
#define GGML_VK_NAME "Vulkan" #define GGML_VK_NAME "Vulkan"
#define GGML_VK_MAX_DEVICES 16 #define GGML_VK_MAX_DEVICES 16
GGML_API void ggml_vk_instance_init(void);
GGML_API void ggml_vk_init_cpu_assist(void); GGML_API void ggml_vk_init_cpu_assist(void);
GGML_API void ggml_vk_preallocate_buffers_graph_cpu_assist(struct ggml_tensor * node); GGML_API void ggml_vk_preallocate_buffers_graph_cpu_assist(struct ggml_tensor * node);

1207
ggml_vk_generate_shaders.py
View File

File diff suppressed because it is too large Load Diff

4
llama.cpp
View File

@ -5014,8 +5014,8 @@ static struct ggml_tensor * llm_build_kqv(
ggml_mul_mat_set_prec(kq, GGML_PREC_F32); ggml_mul_mat_set_prec(kq, GGML_PREC_F32);
} }
#if defined(GGML_USE_VULKAN) || defined(GGML_USE_KOMPUTE) #if defined(GGML_USE_KOMPUTE)
#pragma message("TODO: ALiBi support in ggml_soft_max_ext is not implemented for Vulkan, and Kompute") #pragma message("TODO: ALiBi support in ggml_soft_max_ext is not implemented for Kompute")
#pragma message(" Falling back to ggml_alibi(). Will become an error in Mar 2024") #pragma message(" Falling back to ggml_alibi(). Will become an error in Mar 2024")
#pragma message("ref: https://github.com/ggerganov/llama.cpp/pull/5488") #pragma message("ref: https://github.com/ggerganov/llama.cpp/pull/5488")
if (hparams.f_max_alibi_bias > 0.0f) { if (hparams.f_max_alibi_bias > 0.0f) {