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

ggml : re-enable BLAS for CPU when src0 != F32 + remove redundant full offload checks in llama.cpp (#4240)

Browse source 
* ggml : use blas even if src0 is not F32

* llama : use n_threads_batch only when n_tokens >= 32

ggml-ci

* llama : revert n_threads_batch logic

ggml-ci
This commit is contained in:
Georgi Gerganov 2023-11-28 10:32:03 +02:00 • committed by GitHub
parent b38a16dfcf
commit 8406b0924b
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
2 changed files with 2 additions and 12 deletions
Download patch file Download diff file Expand all files Collapse all files

2
ggml.c
View file

@ -9373,7 +9373,7 @@ static bool ggml_compute_forward_mul_mat_use_blas(
// TODO: find the optimal values for these
if (ggml_is_contiguous(src0) &&
ggml_is_contiguous(src1) &&
src0->type == GGML_TYPE_F32 &&
//src0->type == GGML_TYPE_F32 &&
src1->type == GGML_TYPE_F32 &&
(ne0 >= 32 && ne1 >= 32 && ne10 >= 32)) {