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

add conditions

Browse source
This commit is contained in:
abhilash1910 2024-03-26 09:19:06 -07:00
parent d4b182ccd5
commit e9377baf7a
1 changed files with 3 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

5
ggml-sycl.cpp
View file

@ -15359,8 +15359,9 @@ static void ggml_sycl_mul_mat(const ggml_tensor * src0, const ggml_tensor * src1
bool use_dequantize_mul_mat_vec = (ggml_is_quantized(src0->type) || src0->type == GGML_TYPE_F16)
&& src1->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32
&& src0->ne[0] % GGML_SYCL_DMMV_X == 0 && src1->ne[1] == 1;
bool use_mul_mat_vec_q = min_compute_capability >= VER_4VEC && ggml_is_quantized(src0->type)
&& src1->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32;
bool use_mul_mat_vec_q = min_compute_capability >= VER_4VEC && (ggml_is_quantized(src0->type)
|| src0->type == GGML_TYPE_F16) && src1->type == GGML_TYPE_F32
&& dst->type == GGML_TYPE_F32;
bool use_mul_mat_q = min_compute_capability >= VER_4VEC && ggml_is_quantized(src0->type)
&& src1->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32
&& ggml_sycl_supports_mmq(src0->type);