From bf4ce086409ad71673e80be660600b130bdfceb6 Mon Sep 17 00:00:00 2001 From: ishaan-jaff Date: Mon, 9 Oct 2023 13:42:56 -0700 Subject: [PATCH] (fix) acompletion for ollama non streaing --- litellm/main.py | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-) diff --git a/litellm/main.py b/litellm/main.py index 8d8930ca490..3b3539ee1af 100644 --- a/litellm/main.py +++ b/litellm/main.py @@ -1000,9 +1000,11 @@ def completion( logging.pre_call( input=prompt, api_key=None, additional_args={"api_base": api_base, "custom_prompt_dict": litellm.custom_prompt_dict} ) - if kwargs.get('acompletion', False) == True: - async_generator = ollama.async_get_ollama_response_stream(api_base, model, prompt, optional_params) - return async_generator + if kwargs.get('acompletion', False) == True: + if optional_params.get("stream", False) == True: + # assume all ollama responses are streamed + async_generator = ollama.async_get_ollama_response_stream(api_base, model, prompt, optional_params) + return async_generator generator = ollama.get_ollama_response_stream(api_base, model, prompt, optional_params) if optional_params.get("stream", False) == True: