from fastapi import FastAPI, HTTPException, Header
from fastapi.middleware.cors import CORSMiddleware
from fastapi.responses import StreamingResponse
from pydantic import BaseModel
import openai
from typing import List, Optional, Union
import logging
import httpx
import uuid
import time
import json
from datetime import datetime, timezone
import requests
import uvicorn
import random
logging.basicConfig(
level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s"
)
logger = logging.getLogger(__name__)
app = FastAPI()
app.add_middleware(
CORSMiddleware,
allow_origins=["*"],
allow_credentials=True,
allow_methods=["*"],
allow_headers=["*"],
)
MAX_RETRIES = 3
class ChatRequest(BaseModel):
messages: List[dict]
model: str
temperature: Optional[float] = 0.7
stream: Optional[bool] = False
tools: Optional[List[dict]] = []
tool_choice: Optional[str] = "auto"
class EmbeddingRequest(BaseModel):
input: Union[str, List[str]]
model: str
encoding_format: Optional[str] = "float"
async def verify_authorization(authorization: str = Header(None)):
print("Authorization header:", authorization)
if not authorization:
logger.error("Missing Authorization header")
raise HTTPException(status_code=401, detail="Missing Authorization header")
if not authorization.startswith("Bearer "):
logger.error("Invalid Authorization header format")
raise HTTPException(
status_code=401, detail="Invalid Authorization header format"
)
token = authorization.replace("Bearer ", "")
return token
def get_openai_models(api_keys):
api_key = random.choice(api_keys)
try:
client = openai.OpenAI(api_key=api_key)
models = client.models.list()
return models.model_dump()
except Exception as e:
logger.error(f"Error getting models from OpenAI with key {api_key}: {e}")
return {"error": str(e)}
def get_gemini_models(api_keys):
api_key = random.choice(api_keys)
base_url = "https://generativelanguage.googleapis.com/v1beta"
url = f"{base_url}/models?key={api_key}"
try:
response = requests.get(url)
if response.status_code == 200:
gemini_models = response.json()
return convert_to_openai_models_format(gemini_models)
else:
logger.error(f"Error getting models from Gemini with key {api_key}: {response.status_code} - {response.text}")
return {"error": f"Gemini API error: {response.status_code} - {response.text}"}
except requests.RequestException as e:
logger.error(f"Request failed: {e}")
return {"error": f"Request failed: {e}"}
def convert_to_openai_models_format(gemini_models):
openai_format = {"object": "list", "data": []}
for model in gemini_models.get("models", []):
openai_model = {
"id": model["name"].split("/")[-1],
"object": "model",
"created": int(datetime.now(timezone.utc).timestamp()),
"owned_by": "google",
"permission": [],
"root": model["name"],
"parent": None,
}
openai_format["data"].append(openai_model)
return openai_format
def convert_messages_to_gemini_format(messages):
gemini_messages = []
for msg in messages:
role = "user" if msg["role"] == "user" else "model"
parts = []
if isinstance(msg["content"], str):
parts.append({"text": msg["content"]})
elif isinstance(msg["content"], list):
for content in msg["content"]:
if isinstance(content, str):
parts.append({"text": content})
elif isinstance(content, dict) and content["type"] == "text":
parts.append({"text": content["text"]})
elif isinstance(content, dict) and content["type"] == "image_url":
image_url = content["image_url"]["url"]
if image_url.startswith("data:image"):
parts.append(
{
"inline_data": {
"mime_type": "image/jpeg",
"data": image_url.split(",")[1],
}
}
)
else:
parts.append(
{
"image_url": {
"url": image_url,
}
}
)
gemini_messages.append({"role": role, "parts": parts})
return gemini_messages
async def convert_gemini_response_to_openai(response, model, stream=False):
if stream:
chunk = response
if not chunk["candidates"]:
return None
return {
"id": "chatcmpl-" + str(uuid.uuid4()),
"object": "chat.completion.chunk",
"created": int(time.time()),
"model": model,
"choices": [
{
"index": 0,
"delta": {
"content": chunk["candidates"][0]["content"]["parts"][0]["text"]
},
"finish_reason": None,
}
],
}
else:
content = response["candidates"][0]["content"]["parts"][0]["text"]
return {
"id": "chatcmpl-" + str(uuid.uuid4()),
"object": "chat.completion",
"created": int(time.time()),
"model": model,
"choices": [
{
"index": 0,
"message": {
"role": "assistant",
"content": content,
},
"finish_reason": "stop",
}
],
"usage": {"prompt_tokens": 0, "completion_tokens": 0, "total_tokens": 0},
}
@app.get("/v1/models")
@app.get("/hf/v1/models")
async def list_models(authorization: str = Header(None)):
token = await verify_authorization(authorization)
api_keys = [key.strip() for key in token.split(',')]
all_models = []
error_messages = []
for api_key in api_keys:
if api_key.startswith("sk-"):
response = get_openai_models([api_key])
else:
response = get_gemini_models([api_key])
if "error" in response:
error_messages.append(response["error"])
else:
if isinstance(response, dict) and 'data' in response:
all_models.extend(response['data'])
else:
logger.warning(f"Unexpected response format from model list API for key {api_key}: {response}")
if error_messages and not all_models:
raise HTTPException(status_code=500, detail=f"Errors encountered: {', '.join(error_messages)}")
return {"data": all_models, "object": "list"}
@app.post("/v1/chat/completions")
@app.post("/hf/v1/chat/completions")
async def chat_completion(request: ChatRequest, authorization: str = Header(None)):
token = await verify_authorization(authorization)
api_keys = [key.strip() for key in token.split(',')]
logger.info(f"Chat completion request - Model: {request.model}")
retries = 0
while retries < MAX_RETRIES:
api_key = random.choice(api_keys)
try:
logger.info(f"Attempt {retries + 1} with API key: {api_key}")
if api_key.startswith("sk-"):
client = openai.OpenAI(api_key=api_key)
if request.stream:
logger.info("Streaming response enabled")
async def generate():
try:
stream_response = client.chat.completions.create(
model=request.model,
messages=request.messages,
temperature=request.temperature,
stream=True,
)
for chunk in stream_response:
chunk_json = chunk.model_dump_json()
yield f"data: {chunk_json}\n\n"
yield "data: [DONE]\n\n"
except Exception as e:
logger.error(f"Stream error: {str(e)}")
raise
return StreamingResponse(content=generate(), media_type="text/event-stream")
else:
response = client.chat.completions.create(
model=request.model,
messages=request.messages,
temperature=request.temperature,
)
logger.info("Chat completion successful")
return response.model_dump()
else:
gemini_messages = convert_messages_to_gemini_format(request.messages)
payload = {
"contents": gemini_messages,
"generationConfig": {
"temperature": request.temperature,
}
}
if request.stream:
logger.info("Streaming response enabled")
async def generate():
nonlocal api_key, retries, api_keys
while retries < MAX_RETRIES:
try:
async with httpx.AsyncClient() as client:
stream_url = f"https://generativelanguage.googleapis.com/v1beta/models/{request.model}:streamGenerateContent?alt=sse&key={api_key}"
async with client.stream("POST", stream_url, json=payload, timeout=60.0) as response:
if response.status_code == 429:
logger.warning(f"Rate limit reached for key: {api_key}")
retries += 1
if retries >= MAX_RETRIES:
yield f"data: {json.dumps({'error': 'Max retries reached'})}\n\n"
break
api_keys.remove(api_key)
if not api_keys:
yield f"data: {json.dumps({'error': 'All API keys exhausted'})}\n\n"
break
api_key = random.choice(api_keys)
logger.info(f"Retrying with a new API key: {api_key}")
continue
if response.status_code != 200:
logger.error(f"Error in streaming response with key {api_key}: {response.status_code} - {response.text}")
retries += 1
if retries >= MAX_RETRIES:
yield f"data: {json.dumps({'error': 'Max retries reached'})}\n\n"
break
api_keys.remove(api_key)
if not api_keys:
yield f"data: {json.dumps({'error': 'All API keys exhausted'})}\n\n"
break
api_key = random.choice(api_keys)
logger.info(f"Retrying with a new API key: {api_key}")
continue
async for line in response.aiter_lines():
if line.startswith("data: "):
try:
chunk = json.loads(line[6:])
if not chunk.get("candidates"):
continue
content = chunk["candidates"][0]["content"]["parts"][0]["text"]
new_chunk = {
"id": "chatcmpl-" + str(uuid.uuid4()),
"object": "chat.completion.chunk",
"created": int(time.time()),
"model": request.model,
"choices": [
{
"index": 0,
"delta": {
"content": content
},
"finish_reason": None,
}
],
}
yield f"data: {json.dumps(new_chunk)}\n\n"
except json.JSONDecodeError:
continue
yield "data: [DONE]\n\n"
return
except Exception as e:
logger.error(f"Stream error: {str(e)}")
retries += 1
if retries >= MAX_RETRIES:
yield f"data: {json.dumps({'error': 'Max retries reached'})}\n\n"
break
api_keys.remove(api_key)
if not api_keys:
yield f"data: {json.dumps({'error': 'All API keys exhausted'})}\n\n"
break
api_key = random.choice(api_keys)
logger.info(f"Retrying with a new API key: {api_key}")
continue
return StreamingResponse(content=generate(), media_type="text/event-stream")
else:
async with httpx.AsyncClient() as client:
non_stream_url = f"https://generativelanguage.googleapis.com/v1beta/models/{request.model}:generateContent?key={api_key}"
response = await client.post(non_stream_url, json=payload)
if response.status_code != 200:
logger.error(f"Error in non-streaming response with key {api_key}: {response.status_code} - {response.text}")
retries += 1
if retries >= MAX_RETRIES:
raise HTTPException(status_code=500, detail="Max retries reached")
api_keys.remove(api_key)
if not api_keys:
raise HTTPException(status_code=500, detail="All API keys exhausted")
api_key = random.choice(api_keys)
logger.info(f"Retrying with a new API key: {api_key}")
continue
gemini_response = response.json()
logger.info("Chat completion successful")
return await convert_gemini_response_to_openai(gemini_response, request.model)
except Exception as e:
logger.error(f"Error in chat completion: {str(e)}")
if isinstance(e, HTTPException):
raise e
retries += 1
if retries >= MAX_RETRIES:
logger.error("Max retries reached, giving up")
raise HTTPException(status_code=500, detail="Max retries reached")
api_keys.remove(api_key)
if not api_keys:
raise HTTPException(status_code=500, detail="All API keys exhausted")
api_key = random.choice(api_keys)
logger.info(f"Retrying with a new API key: {api_key}")
continue
raise HTTPException(status_code=500, detail="Unexpected error in chat completion")
@app.get("/health")
@app.get("/")
async def health_check():
logger.info("Health check endpoint called")
return {"status": "healthy"}
if __name__ == "__main__":
uvicorn.run(app, host="0.0.0.0", port=8080)