root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-26 03:14:35 +00:00
Code Issues Actions 7 Packages Projects Releases Wiki Activity

metal : proper assert for mat-mat memory alignment (#6225)

Browse Source 
* metal : proper assert for mat-mat memory alignment

ggml-ci

* readme : add notice about the bug fix

* metal : fix the fix

ggml-ci
This commit is contained in:
Georgi Gerganov 2024-03-22 11:35:53 +02:00 committed by GitHub
parent b2075fd6a5
commit b3e94f26ba
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 41 additions and 24 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
README.md
View File

@ -17,6 +17,7 @@ Inference of Meta's [LLaMA](https://arxiv.org/abs/2302.13971) model (and others)
### Hot topics ### Hot topics
- Fix major bug in Metal batched inference https://github.com/ggerganov/llama.cpp/pull/6225
- Multi-GPU pipeline parallelizm support https://github.com/ggerganov/llama.cpp/pull/6017 - Multi-GPU pipeline parallelizm support https://github.com/ggerganov/llama.cpp/pull/6017
- Looking for contributions to add Deepseek support: https://github.com/ggerganov/llama.cpp/issues/5981 - Looking for contributions to add Deepseek support: https://github.com/ggerganov/llama.cpp/issues/5981
- Quantization blind testing: https://github.com/ggerganov/llama.cpp/discussions/5962 - Quantization blind testing: https://github.com/ggerganov/llama.cpp/discussions/5962

16
ggml-metal.m
View File

@ -1392,6 +1392,14 @@ static enum ggml_status ggml_metal_graph_compute(
(ne11 > ne11_mm_min || (ggml_is_quantized(src0t) && ne12 > 1))) { (ne11 > ne11_mm_min || (ggml_is_quantized(src0t) && ne12 > 1))) {
//printf("matrix: ne00 = %6d, ne01 = %6d, ne02 = %6d, ne11 = %6d, ne12 = %6d\n", ne00, ne01, ne02, ne11, ne12); //printf("matrix: ne00 = %6d, ne01 = %6d, ne02 = %6d, ne11 = %6d, ne12 = %6d\n", ne00, ne01, ne02, ne11, ne12);
// some Metal matrix data types require aligned pointers
// ref: https://developer.apple.com/metal/Metal-Shading-Language-Specification.pdf (Table 2.5)
switch (src0->type) {
case GGML_TYPE_F32: GGML_ASSERT(nb01 % 16 == 0); break;
case GGML_TYPE_F16: GGML_ASSERT(nb01 % 8 == 0); break;
default: break;
}
id<MTLComputePipelineState> pipeline = nil; id<MTLComputePipelineState> pipeline = nil;
switch (src0->type) { switch (src0->type) {
@ -1706,6 +1714,14 @@ static enum ggml_status ggml_metal_graph_compute(
ne20 % 32 == 0 && ne20 >= 64 && ne20 % 32 == 0 && ne20 >= 64 &&
ne11 > ne11_mm_min) { ne11 > ne11_mm_min) {
// some Metal matrix data types require aligned pointers
// ref: https://developer.apple.com/metal/Metal-Shading-Language-Specification.pdf (Table 2.5)
switch (src0->type) {
case GGML_TYPE_F32: GGML_ASSERT(nb01 % 16 == 0); break;
case GGML_TYPE_F16: GGML_ASSERT(nb01 % 8 == 0); break;
default: break;
}
id<MTLComputePipelineState> pipeline = nil; id<MTLComputePipelineState> pipeline = nil;
switch (src2->type) { switch (src2->type) {