| from typing import Optional, Union |
|
|
| from app.utils.logger import get_logger |
| from app.utils.openai_client import build_openai_client |
|
|
| logging= get_logger(__name__) |
| class OpenAICompatibleProvider: |
| def __init__(self, api_key: str, base_url: str, model: Union[str, None]=None): |
| |
| self.client = build_openai_client(api_key, base_url, key_label="模型供应商的 API Key") |
| self.model = model |
|
|
| @property |
| def get_client(self): |
| return self.client |
|
|
| @staticmethod |
| def test_connection(api_key: str, base_url: str, model: str) -> bool: |
| """发一条最小化 chat completion 验证 key / base_url / model 三方都通。 |
| |
| 为什么不用 client.models.list(): |
| - 部分代理 / 自建供应商不实现 /v1/models(如某些 OpenAI 兼容网关) |
| - 部分供应商 key 在没有 inference 权限时 /v1/models 仍返回 200 |
| 最终用户跑的就是 chat.completions.create,所以直接测它最忠实。 |
| max_tokens=1 + temperature=0 让请求开销 < 0.0001 美元、延迟 < 2s。 |
| """ |
| try: |
| client = build_openai_client( |
| api_key, base_url, key_label="模型供应商的 API Key", timeout=15.0, |
| ) |
| client.chat.completions.create( |
| model=model, |
| messages=[{"role": "user", "content": "ping"}], |
| max_tokens=1, |
| temperature=0, |
| ) |
| logging.info(f"连通性测试成功(model={model})") |
| return True |
| except Exception as e: |
| logging.warning(f"连通性测试失败(model={model}):{e}") |
| return False |