diff --git a/examples/fastapi_server.py b/examples/fastapi_server.py index e219b19..728d3f7 100644 --- a/examples/fastapi_server.py +++ b/examples/fastapi_server.py @@ -34,6 +34,7 @@ llama = llama_cpp.Llama( embedding=True, n_threads=6, n_batch=2048, + n_ctx=2048, )