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

Free KV cache CUDA buffers upon deletion

Browse source
This commit is contained in:
JohannesGaessler 2023-06-13 11:15:30 +02:00
parent 8e3057b24b
commit ed6587491c
1 changed files with 5 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

5
llama.cpp
View file

@ -165,6 +165,11 @@ struct llama_kv_cache {
if (ctx) { if (ctx) {
ggml_free(ctx); ggml_free(ctx);
} }
#ifdef GGML_USE_CUBLAS
ggml_cuda_free_data(k);
ggml_cuda_free_data(v);
#endif // GGML_USE_CUBLAS
} }
}; };