| """OpenAI 兼容 API 客户端。 |
| |
| 适用场景： |
| 1. 用户用 ``vllm serve`` / ``sglang`` / ``lmdeploy`` 等起的本地 OpenAI 兼容服务 |
| 2. OpenAI 官方 / Gemini-OpenAI-compat / Claude-OpenAI-compat 等公有云 |
| 3. 任意自托管的 OpenAI Chat Completions 协议网关 |
| |
| 只依赖标准 ``openai`` Python 客户端，不引入任何特殊鉴权 / cos url 逻辑。 |
| """ |
|
|
| from __future__ import annotations |
|
|
| import re |
| import time |
|
|
| from openai import OpenAI |
|
|
| from ..utils.image_utils import encode_image |
| from .base import APIBase |
|
|
| DEFAULT_TIMEOUT = 1200 |
|
|
|
|
| def _split_think_answer(response: str) -> tuple[str, str]: |
| """从模型输出中拆出 thinking / final answer。""" |
| if not response or not response.strip(): |
| return "", "" |
| m = re.search(r"<think>\n(.*?)\n</think>\n<answer>\n(.*?)\n</answer>", response, flags=re.DOTALL) |
| if m: |
| return m.group(1).strip(), m.group(2).strip() |
| return "", response.strip() |
|
|
|
|
| class OpenAICompatAPI(APIBase): |
| """走 OpenAI Chat Completions 协议的通用客户端。""" |
|
|
| def __init__( |
| self, |
| model_name: str, |
| base_url: str, |
| api_key: str = "EMPTY", |
| max_try: int = 3, |
| timeout: int = DEFAULT_TIMEOUT, |
| image_first: bool = True, |
| ): |
| self.model_name = model_name |
| self.base_url = base_url |
| self.api_key = api_key |
| self.max_try = max_try |
| self.timeout = timeout |
| self.image_first = image_first |
| self.client = OpenAI(base_url=base_url, api_key=api_key) |
|
|
| def __call__(self, img_path: str | None, question: str, temperature: float | None = None, **kwargs): |
| messages = self._build_messages(img_path, question) |
| return self._send(messages, temperature=temperature) |
|
|
| def _build_messages(self, img_path: str | None, question: str) -> list[dict]: |
| if not img_path: |
| assert question, "question is required when img_path is empty" |
| return [{"role": "user", "content": [{"type": "text", "text": question}]}] |
| data_uri = encode_image(img_path) |
| img_part = {"type": "image_url", "image_url": {"url": data_uri}} |
| txt_part = {"type": "text", "text": question} |
| content = [img_part, txt_part] if self.image_first else [txt_part, img_part] |
| return [{"role": "user", "content": content}] |
|
|
| def _send(self, messages: list[dict], temperature: float | None = None): |
| for attempt in range(1, self.max_try + 1): |
| try: |
| completion = self.client.chat.completions.create( |
| model=self.model_name, |
| messages=messages, |
| temperature=temperature, |
| timeout=self.timeout, |
| ) |
| response = completion.choices[0].message.content or "" |
| thinking, answer = _split_think_answer(response) |
| return True, thinking, answer |
| except Exception as e: |
| print(f"[OpenAICompatAPI] 尝试 {attempt}/{self.max_try} 失败: {e}") |
| if attempt < self.max_try: |
| time.sleep(min(2 * attempt, 10)) |
| return False, "", "" |
|
|
