Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-09-19 01:05:09 +02:00
Code Issues Packages Projects Releases Wiki Activity

support bf16 quantkv cache type

Browse Source
This commit is contained in:
Concedo
2026-03-28 00:01:17 +08:00
parent 326542f480
commit 0c2b679ea3
2 changed files with 8 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files
gpttype_adapter.cpp
+2 -2
View File
@@ -2673,8 +2673,8 @@ ModelLoadResult gpttype_load_model(const load_model_inputs inputs, FileFormat in
llama_ctx_params.flash_attn_type = (kcpp_data->flash_attn?LLAMA_FLASH_ATTN_TYPE_ENABLED:LLAMA_FLASH_ATTN_TYPE_DISABLED);
llama_ctx_params.swa_full = kcpp_data->swa_full;
llama_ctx_params.type_k = (inputs.quant_k>1?GGML_TYPE_Q4_0:(inputs.quant_k==1?GGML_TYPE_Q8_0:GGML_TYPE_F16));
llama_ctx_params.type_v = (inputs.quant_v>1?GGML_TYPE_Q4_0:(inputs.quant_v==1?GGML_TYPE_Q8_0:GGML_TYPE_F16));
llama_ctx_params.type_k = (inputs.quant_k==2?GGML_TYPE_Q4_0:(inputs.quant_k==1?GGML_TYPE_Q8_0:(inputs.quant_k==3?GGML_TYPE_BF16:GGML_TYPE_F16)));
llama_ctx_params.type_v = (inputs.quant_v==2?GGML_TYPE_Q4_0:(inputs.quant_v==1?GGML_TYPE_Q8_0:(inputs.quant_v==3?GGML_TYPE_BF16:GGML_TYPE_F16)));
llama_ctx_v4 = llama_init_from_model(llamamodel, llama_ctx_params);
if(load_guidance)