"""OpenAI 兼容客户端:聊天、生图、TTS,全部按渠道转发。 性能:全局共享 AsyncClient(连接池复用,避免每请求握手)。 """ import base64 import httpx # 进程级共享连接池; comercial 场景下量大可调 max_connections _shared_client: httpx.AsyncClient | None = None def get_http() -> httpx.AsyncClient: global _shared_client if _shared_client is None or _shared_client.is_closed: _shared_client = httpx.AsyncClient( timeout=httpx.Timeout(180, connect=10), limits=httpx.Limits(max_connections=100, max_keepalive_connections=20)) return _shared_client class LLMClient: def __init__(self, base_url: str, api_key: str): self.base_url = base_url.rstrip("/") self.api_key = api_key async def chat(self, model: str, messages: list[dict], temperature: float = 0.9, max_tokens: int = 2048) -> str: r = await get_http().post( f"{self.base_url}/chat/completions", headers={"Authorization": f"Bearer {self.api_key}"}, json={"model": model, "messages": messages, "temperature": temperature, "max_tokens": max_tokens}) r.raise_for_status() return r.json()["choices"][0]["message"]["content"] or "" async def image(self, model: str, prompt: str, size: str = "1024x1024") -> str: r = await get_http().post( f"{self.base_url}/images/generations", headers={"Authorization": f"Bearer {self.api_key}"}, json={"model": model, "prompt": prompt, "size": size, "n": 1, "response_format": "url"}) r.raise_for_status() d = r.json()["data"][0] return d.get("url") or ("data:image/png;base64," + d.get("b64_json", "")) async def tts(self, model: str, text: str, voice: str) -> bytes: """优先 OpenAI 兼容 /audio/speech;失败回落阶跃 stepaudio TTS 格式。""" headers = {"Authorization": f"Bearer {self.api_key}"} try: r = await get_http().post(f"{self.base_url}/audio/speech", headers=headers, json={"model": model, "input": text, "voice": voice}) r.raise_for_status() return r.content except httpx.HTTPError: r = await get_http().post(f"{self.base_url}/audio/speech", headers=headers, json={"model": model, "text": text, "voice": voice, "response_format": "wav"}) r.raise_for_status() return r.content async def ping(self, model: str) -> tuple[bool, str]: try: reply = await self.chat(model, [{"role": "user", "content": "ping"}], temperature=0, max_tokens=8) return True, reply or "ok" except Exception as e: return False, str(e)[:300]