Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-09-19 17:25:07 +02:00
Code Issues Packages Projects Releases Wiki Activity

allow clip fa for anything besides cuda on gpu

Browse Source
This commit is contained in:
Concedo
2026-01-01 21:09:51 +08:00
parent 442fa7cd7c
commit 51edb6ae61
1 changed files with 6 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
gpttype_adapter.cpp
+6 -1
View File
@@ -2665,14 +2665,19 @@ ModelLoadResult gpttype_load_model(const load_model_inputs inputs, FileFormat in
printf("Clip will use CPU for this model!\n");
}
#endif
clip_flash_attn_type clip_fa = (kcpp_data->flash_attn?CLIP_FLASH_ATTN_TYPE_ENABLED:CLIP_FLASH_ATTN_TYPE_DISABLED); //kcpp: disabled in 1.102.2 as some headsizes break on turing
#if defined(GGML_USE_CUDA)
clip_fa = CLIP_FLASH_ATTN_TYPE_DISABLED; //kcpp: disabled in 1.102.2 as some headsizes break on turing
#endif
if(inputs.mmproj_cpu)
{
set_clip_uses_gpu(false);
printf("Clip forced to use CPU!\n");
clip_fa = (kcpp_data->flash_attn?CLIP_FLASH_ATTN_TYPE_ENABLED:CLIP_FLASH_ATTN_TYPE_DISABLED); //however if using CPU, fa is fine
}
clip_context_params ctx_clip_params {
/* use_gpu */ true,
/* flash_attn_type */ CLIP_FLASH_ATTN_TYPE_DISABLED, //kcpp: disabled in 1.102.2 as some headsizes break on turing
/* flash_attn_type */ clip_fa,
/* image_min_tokens */ -1,
/* image_max_tokens */ -1,
};