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

ggml : update ggml_soft_max_ext() CUDA, SYCL

Browse source
This commit is contained in:
Georgi Gerganov 2024-05-10 11:12:19 +03:00
parent 7fdca3348c
commit d0592d495d
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
4 changed files with 38 additions and 74 deletions
Download patch file Download diff file Expand all files Collapse all files

3
ggml-kompute.cpp
View file

@ -1561,10 +1561,9 @@ static void ggml_vk_graph_compute(struct ggml_kompute_context * ctx, struct ggml
float scale;
memcpy(&scale, dst->op_params, sizeof(float));
#pragma message("TODO: add ggml_vk_soft_max() F16/F32 src1 and src2 support")
#pragma message("TODO: add ggml_vk_soft_max() F16 src1 support")
#pragma message("ref: https://github.com/ggerganov/llama.cpp/pull/5021")
GGML_ASSERT(!src1 || src1t == GGML_TYPE_F32);
GGML_ASSERT(src2 == nullptr);
ggml_vk_soft_max(seq, id_src0, id_src1, id_dst, off_src0, off_src1, off_dst, ne00, ne01, ne02, ne03, scale);
} break;