diff --git a/rag/generate.py b/rag/generate.py index b6e96fb6..25582250 100644 --- a/rag/generate.py +++ b/rag/generate.py @@ -3,7 +3,7 @@ import sys import time from pathlib import Path - +from litellm import completion import numpy as np import openai import psycopg @@ -45,7 +45,7 @@ def generate_response( api_base, api_key = get_credentials(llm=llm) while retry_count < max_retries: try: - response = openai.ChatCompletion.create( + response = completion( model=llm, temperature=temperature, stream=stream,