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

Add quant type GGML_TYPE_IQ1_S to unsupported

Browse source
This commit is contained in:
Aidan 2024-02-19 16:51:06 +00:00
parent 5d7d353707
commit 8388e73b72
1 changed files with 6 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

8
ggml-sycl.cpp
View file

@ -11946,7 +11946,8 @@ inline void ggml_sycl_op_mul_mat_vec_q(
// TODO: support these quantization types
GGML_ASSERT(!(src0->type == GGML_TYPE_IQ2_XXS ||
src0->type == GGML_TYPE_IQ2_XS ||
src0->type == GGML_TYPE_IQ3_XXS));
src0->type == GGML_TYPE_IQ3_XXS ||
src0->type == GGML_TYPE_IQ1_S));
switch (src0->type) {
case GGML_TYPE_Q4_0:
@ -14965,7 +14966,10 @@ static bool ggml_backend_sycl_supports_op(ggml_backend_t backend, const ggml_ten
if (a->ne[3] != b->ne[3]) {
return false;
}
if (a->type == GGML_TYPE_IQ1_S) {
return false;
}
if (a->type == GGML_TYPE_IQ3_XXS) {
return false;
}