import asyncio
import os

os.environ['MODELSCOPE_ENVIRONMENT'] = ''

from app import chat_completions, ChatCompletionRequest

async def run_once():
    req = ChatCompletionRequest(model='rwkv-latest', prompt='Who is the president of France today?', stream=False, max_tokens=32, temperature=0.2, include_usage=True)
    res = await chat_completions(req)
    print(res)

if __name__ == '__main__':
    asyncio.run(run_once())