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

add q8_0 q4_0 tests

Browse source
This commit is contained in:
Johannes Gäßler 2024-05-26 22:30:46 +02:00
parent 3194a01058
commit f08776041d
2 changed files with 16 additions and 8 deletions
Download patch file Download diff file Expand all files Collapse all files

8
ggml-cuda.cu
View file

@ -2889,10 +2889,14 @@ GGML_CALL static bool ggml_backend_cuda_supports_op(ggml_backend_t backend, cons
#if defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)
return op->src[0]->ne[0] == 64 || op->src[0]->ne[0] == 128;
#else
if (op->src[0]->ne[0] == 64 || op->src[0]->ne[0] == 128) {
if (op->src[0]->ne[0] == 128) {
return true;
}
return ggml_cuda_info().devices[cuda_ctx->device].cc >= CC_VOLTA;
if (op->src[0]->ne[0] == 64 && op->src[1]->type == GGML_TYPE_F16) {
return true;
}
return ggml_cuda_info().devices[cuda_ctx->device].cc >= CC_VOLTA &&
op->src[1]->type == GGML_TYPE_F16 && op->src[2]->type == GGML_TYPE_F16;
#endif // defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)
default:
return false;