Function llm_gpt

eval_heldout/science-world/eval.py:77–115

(prompt: List[Dict[str, str]], model: str)

    return num_tokens

def llm_gpt(prompt: List[Dict[str, str]], model: str) -> str:
    if not 'OPENAI_API_KEY' in os.environ:
        raise ValueError("OPENAI_API_KEY must be set to eval GPT models.")

    for _ in range(3):
        try:
            openai_api_key = os.environ['OPENAI_API_KEY']
            openai_api_base = os.environ.get('OPENAI_API_BASE', 'https://api.openai.com/v1')
            response = requests.post(
                openai_api_base + "/chat/completions",
                headers={
                    'Authorization': f'Bearer {openai_api_key}'
                },
                json={
                    'model': model,
                    'messages': prompt,
                    'temperature': 0.0,
                    'max_tokens': 256,
                },
                timeout=120,
            )
            text = response.json()['choices'][0]['message']['content']
            print(text)
            return text.strip()
        # if timeout or connection error, retry
        except Timeout: 
            print("Timeout, retrying...")
        except ConnectionError:
            print("Connection error, retrying...")
        except Exception:
            traceback.print_exc()
            try:
                print(response)
                print(response.text)
            except:
                pass
        time.sleep(5)
    else:
        raise Exception("Timeout after 3 retries.")

def llm_tgi(prompt: str) -> str:
    data = {