Update app/main.py

app/main.py

@@ -151,7 +151,7 @@ async def startup_event():
             log_msg = format_log_message('INFO', "Available models loaded.")
             logger.info(log_msg)
 
-@app.get("/hf/v1/models", response_model=ModelList)
+@app.get("/v1/models", response_model=ModelList)
 def list_models():
     log_msg = format_log_message('INFO', "Received request to list models", extra={'request_type': 'list_models', 'status_code': 200})
     logger.info(log_msg)
@@ -161,6 +161,8 @@ def list_models():
 async def verify_password(request: Request):
     if PASSWORD:
         auth_header = request.headers.get("Authorization")
+        if not auth_header: logger.info("no auth header")
+        if not auth_header.startswith("Bearer "): logger.info("not start with bearer")
         if not auth_header or not auth_header.startswith("Bearer "):
             raise HTTPException(
                 status_code=401, detail="Unauthorized: Missing or invalid token")
@@ -310,7 +312,7 @@ async def process_request(chat_request: ChatCompletionRequest, http_request: Req
         status_code=status.HTTP_500_INTERNAL_SERVER_ERROR, detail=msg)
 
 
-@app.post("/hf/v1/chat/completions", response_model=ChatCompletionResponse)
+@app.post("/v1/chat/completions", response_model=ChatCompletionResponse)
 async def chat_completions(request: ChatCompletionRequest, http_request: Request, _: None = Depends(verify_password)):
     return await process_request(request, http_request, "stream" if request.stream else "non-stream")