bigcode/Reasoning-with-StarCoder
32
1import requests2import json3from io import StringIO4import sys5import gradio6import os7 8 9def generate_response(query, top_p):10 endpoint = os.environ.get("ENDPOINT", None)11 token = os.environ.get("TOKEN", None)12 params = {13 'accept': 'application/json',14 'Content-Type': 'application/json',15 "Authorization": f"Bearer {token}"16 }17 data = {18 'inputs': query,19 "parameters": {20 "details": True,21 "do_sample": True,22 "max_new_tokens": 256,23 "repetition_penalty": 1.2,24 "seed": 0,25 "stop": ['<|endoftext|>'],26 "temperature": 0.2,27 "top_p": top_p,28 }29 }30 try:31 res = requests.post(endpoint, data=json.dumps(data), headers=params)32 res = json.loads(res.content.decode("utf-8"))33 return res[0]["generated_text"].split("<|endoftext|>")[0].strip()34 except:35 raise gradio.Error("Connection Error. Please check your ACCESS_TOKEN")36 37 38def run_code(code):39 output_buffer = StringIO()40 sys.stdout = output_buffer41 exec(code, {})42 sys.stdout = sys.__stdout__43 pred = output_buffer.getvalue().strip()44 return pred45 