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

6
ggml-sycl.cpp
View file

@ -14738,7 +14738,12 @@ inline void ggml_sycl_op_soft_max(const ggml_tensor *src0,
GGML_ASSERT(src0->type == GGML_TYPE_F32);
GGML_ASSERT( dst->type == GGML_TYPE_F32);
const ggml_tensor * src2 = dst->src[2];
#pragma message("TODO: add ggml_sycl_op_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); // src1 contains mask and it is optional
GGML_ASSERT(!src2 || src2->type == GGML_TYPE_F32); // src2 contains positions and it is optional
const int64_t ne00 = src0->ne[0];
const int64_t nrows_x = ggml_nrows(src0);
@ -14754,7 +14759,6 @@ inline void ggml_sycl_op_soft_max(const ggml_tensor *src0,
float * src2_dd = nullptr;
sycl_pool_alloc<float> src2_f;
ggml_tensor * src2 = dst->src[2];
const bool use_src2 = src2 != nullptr;
if (use_src2) {