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

Avoid unnecessarily disabling CUDA graphs

Browse source 
As discussed in PR #6766, CUDA graphs were being disabled in the presence of long prompts.
This fixes the issue by avoiding the consective update counter from incrementing unnecessarily
for tokens in which cuda graphs are disabled due to batch size > 1.
This commit is contained in:
Alan Gray 2024-05-15 04:24:55 -07:00
parent 583fd6b000
commit c9c88ac378
1 changed files with 1 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

2
ggml-cuda.cu
View file

@ -2558,7 +2558,7 @@ GGML_CALL static enum ggml_status ggml_backend_cuda_graph_compute(ggml_backend_t
}
// Disable CUDA graphs (from the next token) if the use-case is demanding too many consecutive graph updates.
if (cuda_graph_update_required) {
if (use_cuda_graph && cuda_graph_update_required) {
cuda_ctx->cuda_graph->number_consecutive_updates++;
} else {
cuda_ctx->cuda_graph->number_consecutive_updates = 0;