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

Free resources except for backend.

Browse source
This commit is contained in:
Markus Tavenrath 2024-06-12 13:45:41 +02:00
parent 32dd2ef133
commit 1ebe20789b
1 changed files with 6 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

8
llama.cpp
View file

@ -3936,8 +3936,12 @@ struct llama_model_loader {
}
#if defined(GGML_USE_CUDA)
for (auto const& event : events) {
ggml_backend_event_synchronize(event);
for (size_t idx = 0; idx < num_buffers;++idx) {
ggml_backend_event_synchronize(events[idx]);
ggml_backend_event_free(events[idx]);
ggml_backend_buffer_free(host_buffers[idx]);
//ggml_backend_free(backend);
}
#endif