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

falcon : fix CUDA inference by making K and Q contiguous (#2830)

Browse source 
* falcon : fix CUDA inference by making K and Q contiguous

ggml-ci

* cuda : add assert to guard from non-cont ropes
This commit is contained in:
Georgi Gerganov 2023-08-27 16:40:48 +03:00 • committed by GitHub
parent da7455d046
commit eaa13a48ff
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
2 changed files with 8 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

2
ggml-cuda.cu
View file

@ -6337,9 +6337,11 @@ void ggml_cuda_soft_max(const ggml_tensor * src0, const ggml_tensor * src1, ggml
void ggml_cuda_rope(const ggml_tensor * src0, const ggml_tensor * src1, ggml_tensor * dst) {
GGML_ASSERT(src0->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32);
GGML_ASSERT(ggml_is_contiguous(src0)); // TODO: this restriction is temporary until non-cont support is implemented
const int mode = ((int32_t *) dst->op_params)[2];
const bool is_glm = mode & 4;
ggml_cuda_op(src0, src1, dst, ggml_cuda_op_rope, true, !is_glm); // flatten support not implemented for glm
}