69 lines
2.1 KiB
Python
69 lines
2.1 KiB
Python
import logging
|
||
import uuid
|
||
import json
|
||
|
||
from typing import Any, Dict
|
||
from fastapi import APIRouter, HTTPException, status
|
||
from fastapi.responses import StreamingResponse
|
||
|
||
from models import ChatRequest
|
||
from utils import convert_prompt_to_messages, standard_sse_generator
|
||
from config import GLM_API_KEY, GLM_UPSTREAM_URL
|
||
|
||
router = APIRouter()
|
||
logger = logging.getLogger("proxy")
|
||
|
||
|
||
@router.post("/glm/chat")
|
||
async def proxy_glm_chat(request: ChatRequest):
|
||
|
||
request_id = f"glm-{uuid.uuid4()}"
|
||
|
||
if not GLM_API_KEY:
|
||
raise HTTPException(
|
||
status_code=status.HTTP_503_SERVICE_UNAVAILABLE,
|
||
detail="GLM API key is not configured.",
|
||
)
|
||
|
||
client_request_body = request.model_dump()
|
||
logger.debug(
|
||
f"[{request_id}] === GLM New Request ===\n"
|
||
f"[{request_id}] Client Request Body:\n"
|
||
f"{json.dumps(client_request_body, indent=2, ensure_ascii=False)}"
|
||
)
|
||
|
||
messages = convert_prompt_to_messages(request.prompt)
|
||
logger.debug(
|
||
f"[{request_id}] Agent Transformed Messages:\n"
|
||
f"{json.dumps(messages, indent=2, ensure_ascii=False)}"
|
||
)
|
||
|
||
# 基础的请求体
|
||
payload: Dict[str, Any] = {
|
||
"model": "glm-4-flash",
|
||
"messages": messages,
|
||
"stream": True,
|
||
"temperature": request.temperature,
|
||
"top_p": request.top_p,
|
||
}
|
||
|
||
# 如果请求中指定了 max_tokens ,则添加到 payload
|
||
if request.max_tokens:
|
||
payload["max_tokens"] = request.max_tokens
|
||
|
||
# 如果客户端请求 web_search=True，则为 GLM 启用网页搜索工具
|
||
if request.web_search is True:
|
||
payload["tools"] = [{"type": "web_search", "web_search": {"enable": True}}]
|
||
logger.info(f"[{request_id}] Enabled web search for GLM.")
|
||
|
||
logger.debug(
|
||
f"[{request_id}] Upstream Payload:\n"
|
||
f"{json.dumps(payload, indent=2, ensure_ascii=False)}"
|
||
)
|
||
|
||
headers = {"Authorization": f"Bearer {GLM_API_KEY}"}
|
||
return StreamingResponse(
|
||
standard_sse_generator(GLM_UPSTREAM_URL, payload, headers, "GLM", request_id),
|
||
media_type="text/event-stream",
|
||
)
|