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

cuda: add q8_0->f32 cpy operation (#9571)

Browse source 
llama: enable K-shift for quantized KV cache
It will fail on unsupported backends or quant types.
This commit is contained in:
Ivan 2024-09-24 03:14:24 +03:00 • committed by GitHub
parent 0b3bf966f4
commit 116efee0ee
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
3 changed files with 82 additions and 9 deletions
Download patch file Download diff file Expand all files Collapse all files

3
ggml/src/ggml-cuda.cu
View file

@ -2899,6 +2899,9 @@ GGML_CALL static bool ggml_backend_cuda_supports_op(ggml_backend_t backend, cons
if (src0_type == GGML_TYPE_F32 && src1_type == GGML_TYPE_Q8_0) {
return true;
}
if (src0_type == GGML_TYPE_Q8_0 && src1_type == GGML_TYPE_F32) {
return true;
}
if (src0_type == GGML_TYPE_F32 && src1_type == GGML_TYPE_Q4_0) {
return true;
}