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

set nthread=1 when sycl, increase performance

Browse source
This commit is contained in:
jianyuzh 2024-01-15 14:33:52 +08:00 • committed by Meng, Hengyu
parent 95daece908
commit a8936f4902
2 changed files with 2 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

1
ggml.h
View file

@ -2267,6 +2267,7 @@ extern "C" {
GGML_API int ggml_cpu_has_gpublas (void); GGML_API int ggml_cpu_has_gpublas (void);
GGML_API int ggml_cpu_has_sse3 (void); GGML_API int ggml_cpu_has_sse3 (void);
GGML_API int ggml_cpu_has_ssse3 (void); GGML_API int ggml_cpu_has_ssse3 (void);
GGML_API int ggml_cpu_has_sycl (void);
GGML_API int ggml_cpu_has_vsx (void); GGML_API int ggml_cpu_has_vsx (void);
// //

2
llama.cpp
View file

@ -6703,7 +6703,7 @@ static int llama_decode_internal(
} }
const bool fully_offloaded = model.n_gpu_layers >= (int) hparams.n_layer + 1; const bool fully_offloaded = model.n_gpu_layers >= (int) hparams.n_layer + 1;
if (ggml_cpu_has_cublas() && fully_offloaded) { if ((ggml_cpu_has_cublas() || ggml_cpu_has_sycl()) && fully_offloaded) {
n_threads = 1; n_threads = 1;
} }