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

cuda : update supports_op for IQ2

Browse source 
ggml-ci
This commit is contained in:
Georgi Gerganov 2024-01-16 22:13:17 +02:00
parent bc0bb3009c
commit e9a5d54b7d
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
1 changed files with 6 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

6
ggml-cuda.cu
View file

@ -10918,6 +10918,12 @@ GGML_CALL static bool ggml_backend_cuda_supports_op(ggml_backend_t backend, cons
if (a->ne[3] != b->ne[3]) {
return false;
}
ggml_type a_type = a->type;
if (a_type == GGML_TYPE_IQ2_XXS || a_type == GGML_TYPE_IQ2_XS) {
if (b->ne[1] == 1 && ggml_nrows(b) > 1) {
return false;
}
}
return true;
} break;
case GGML_OP_GET_ROWS: