(self, prompt: str)
| 120 | ) |
| 121 | |
| 122 | def __call__(self, prompt: str): |
| 123 | # prompt = truncate_tokens(prompt, encoding_name=self.config["model"], max_length=self.config["max_tokens"]) |
| 124 | response = self.client.chat.completions.create( |
| 125 | temperature=0, |
| 126 | model=self.config["model"], |
| 127 | messages=[ |
| 128 | {"role": "system", "content": self.system_prompt}, |
| 129 | {"role": "user", "content": prompt}, |
| 130 | ] |
| 131 | ) |
| 132 | return response.choices[0].message.content |
| 133 | |
| 134 | class VLLM(LLM): |
| 135 | def __init__(self, config): |