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

ggml : add TODO's for F16/F32 mask/pos support in other backends

Browse source
This commit is contained in:
Georgi Gerganov 2024-04-23 10:01:49 +03:00
parent c129369702
commit 3864eea4cb
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
3 changed files with 17 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

5
ggml-vulkan.cpp
View file

@ -3178,6 +3178,11 @@ static vk_pipeline ggml_vk_op_get_pipeline(ggml_backend_vk_context * ctx, const
}
return nullptr;
case GGML_OP_SOFT_MAX:
#pragma message("TODO: add ggml_vk_soft_max() F16 src1 and src2 support")
#pragma message("ref: https://github.com/ggerganov/llama.cpp/pull/5021")
GGML_ASSERT(!src1 || src1->type == GGML_TYPE_F32);
GGML_ASSERT(!src2 || src2->type == GGML_TYPE_F32);
if (src0->type == GGML_TYPE_F32 && (src1 == nullptr || src1->type == GGML_TYPE_F32) && (src2 == nullptr || src2->type == GGML_TYPE_F32) && dst->type == GGML_TYPE_F32) {
return ctx->device->pipeline_soft_max_f32;
}