(self, model_path: str)
| 8 | @serve.deployment |
| 9 | class LlamaDeployment: |
| 10 | def __init__(self, model_path: str): |
| 11 | self._llm = Llama(model_path=model_path) |
| 12 | |
| 13 | async def __call__(self, http_request: Request) -> Dict: |
| 14 | input_json = await http_request.json() |