increase scratch for above 4096

YellowRoseCx · Jul 2, 2023 · e19483c · e19483c
1 parent b85ea58
commit e19483c
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/llama.cpp b/llama.cpp
@@ -1127,7 +1127,7 @@ static void llama_model_load_internal(
  const size_t scale = memory_type == GGML_TYPE_F32 ? 2 : 1;
 
  // this is the total memory required to run the inference
- const size_t bigctxmul = (hparams.n_ctx>2048?2:1);
+ const size_t bigctxmul = (hparams.n_ctx>4096?3:(hparams.n_ctx>2048?2:1));
  const size_t mem_required =
  ctx_size +
  mmapped_size - vram_weights + // weights in VRAM not in memory