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

kompute : mark last few failing ops as unsupported

Browse source
This commit is contained in:
Jared Van Bortel 2024-01-25 15:47:43 -05:00
parent 445a3734b7
commit 3fbf0529ef
1 changed files with 3 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

5
ggml-kompute.cpp
View file

@ -1350,7 +1350,7 @@ static bool ggml_vk_supports_op(const struct ggml_tensor * op) {
case GGML_TYPE_Q4_0:
case GGML_TYPE_Q4_1:
case GGML_TYPE_Q6_K:
return op->ne[3] == 1;
return op->ne[2] == 1 && op->ne[3] == 1;
default:
;
}
@ -1361,11 +1361,12 @@ static bool ggml_vk_supports_op(const struct ggml_tensor * op) {
switch (op->src[0]->type) {
case GGML_TYPE_F32:
case GGML_TYPE_Q6_K:
return op->ne[3] == 1;
case GGML_TYPE_F16:
case GGML_TYPE_Q8_0:
case GGML_TYPE_Q4_0:
case GGML_TYPE_Q4_1:
case GGML_TYPE_Q6_K:
return true;
default:
;