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

workaround for deepseek not working

Browse Source
This commit is contained in:
Concedo
2024-07-05 22:39:53 +08:00
parent 6b0756506b
commit 388a2aff00
3 changed files with 17 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama.cpp
+2
View File
@@ -5607,12 +5607,14 @@ static bool llm_load_tensors(
int64_t i_gpu_start = std::max((int64_t) hparams.n_layer - n_gpu_layers, (int64_t) 0);
bool use_mmap_buffer = true;
#if defined(GGML_USE_CLBLAST)
if(clblast_offload_fallback_mode)
{
printf("\nOpenCL GPU Offload Fallback...");
clblast_offload_fallback_layers = n_gpu_layers;
i_gpu_start = std::max((int64_t) hparams.n_layer, (int64_t) 0);
}
#endif
// there is very little benefit to offloading the input layer, so always keep it on the CPU
model.buft_input = llama_default_buffer_type_cpu(true);