simple gpt streaming implementation (works as 2023/6/14)

gpt-stream.py

import requests
import json
import sseclient
 
# API_KEY = 'sk-xxx'


def performRequestWithStreaming():
    reqUrl = 'https://api.openai.com/v1/chat/completions'
    reqHeaders = {
        'Accept': 'text/event-stream',
        'Authorization': 'Bearer ' + API_KEY
    }
    reqBody = {
      "model": "gpt-3.5-turbo-16k",
      "messages": [{"role": "system", "content": "you are a helpful assistant"}, {"role":"user","content":"hello"}],
      "max_tokens": 1000,
      "temperature": 0,
      "stream": True,
    }
    request = requests.post(reqUrl, stream=True, headers=reqHeaders, json=reqBody)
    client = sseclient.SSEClient(request)
    text = ""
    for event in client.events():
        # print(event.data)
        # print(json.loads(event.data))
        if event.data != '[DONE]':
            delta = json.loads(event.data)['choices'][0]['delta']
            if "content" in delta:
                text += delta["content"]
                print("\r"+text, end="", flush=True),

if __name__ == '__main__':
    performRequestWithStreaming()