diff --git a/src/utils/utils.py b/src/utils/utils.py index c4218cd9..23090cb6 100644 --- a/src/utils/utils.py +++ b/src/utils/utils.py @@ -107,6 +107,7 @@ def get_llm_model(provider: str, **kwargs): temperature=kwargs.get("temperature", 0.0), num_ctx=kwargs.get("num_ctx", 32000), num_predict=kwargs.get("num_predict", 1024), + num_gpu=1, # Explicitly enable GPU usage base_url=kwargs.get("base_url", base_url), ) elif provider == "azure_openai":