diff --git a/llmsql/inference/inference_vllm.py b/llmsql/inference/inference_vllm.py index 721109e..caef064 100644 --- a/llmsql/inference/inference_vllm.py +++ b/llmsql/inference/inference_vllm.py @@ -22,6 +22,11 @@ """ import os + +os.environ["VLLM_USE_V1"] = "0" +os.environ["CUBLAS_WORKSPACE_CONFIG"] = ":4096:8" +os.environ["VLLM_ENABLE_V1_MULTIPROCESSING"] = "0" + from pathlib import Path from typing import Any