Spaces:
Sleeping
Sleeping
| import json | |
| import threading | |
| import requests | |
| import gradio as gr | |
| from agent import build_agent, answer | |
| OWNER='SystemReset' | |
| lock=threading.Lock() | |
| def require_owner(profile): | |
| if profile is None or profile.username != OWNER: | |
| raise gr.Error('Sign in as SystemReset to use these controls.') | |
| def respond(message, profile: gr.OAuthProfile | None): | |
| require_owner(profile) | |
| return answer(build_agent(use_case=True),message) | |
| def run_challenge(profile: gr.OAuthProfile | None): | |
| require_owner(profile) | |
| if not lock.acquire(blocking=False): raise gr.Error('An evaluation is already running.') | |
| try: | |
| response=requests.get('https://agents-course-unit4-scoring.hf.space/questions',timeout=60) | |
| response.raise_for_status() | |
| questions=response.json() | |
| agent=build_agent() | |
| results=[] | |
| for index,item in enumerate(questions): | |
| if item.get('file_name') or 'youtube.com/' in item['question']: | |
| row={'task_id':item['task_id'],'submitted_answer':'','status':'media unsupported'} | |
| else: | |
| try: | |
| result=answer(agent,item['question']) | |
| row={'task_id':item['task_id'],'submitted_answer':result,'status':'generated' if result else 'insufficient evidence'} | |
| except Exception as exc: | |
| row={'task_id':item['task_id'],'submitted_answer':'','status':type(exc).__name__} | |
| results.append(row) | |
| with open('challenge_results.json','w') as handle: json.dump(results,handle) | |
| yield f'Processed {index+1}/{len(questions)} questions',results | |
| finally: | |
| lock.release() | |
| def submit_challenge(profile: gr.OAuthProfile | None): | |
| require_owner(profile) | |
| if lock.locked(): raise gr.Error('Wait for evaluation to finish.') | |
| with open('challenge_results.json') as handle: results=json.load(handle) | |
| if len(results)!=20: raise gr.Error('Complete the evaluation first.') | |
| payload={'username':OWNER,'agent_code':'https://huggingface.co/spaces/SystemReset/alfred-agents-course/tree/main','answers':[{'task_id':row['task_id'],'submitted_answer':row['submitted_answer']} for row in results]} | |
| response=requests.post('https://agents-course-unit4-scoring.hf.space/submit',json=payload,timeout=120) | |
| response.raise_for_status() | |
| return response.json() | |
| with gr.Blocks() as demo: | |
| gr.Markdown('# Alfred: Course Research Agent\nSign in as SystemReset. Code is public; model calls and challenge controls are restricted to the owner.') | |
| gr.LoginButton() | |
| prompt=gr.Textbox(label='Question') | |
| reply=gr.Textbox(label='Answer') | |
| gr.Button('Ask Alfred').click(respond,inputs=[prompt],outputs=[reply]) | |
| gr.Markdown('## Final challenge\nRuns the real agent on benchmark questions. Unsupported media is left unanswered. Submit publishes generated answers and score to the course leaderboard.') | |
| status=gr.Textbox(label='Progress') | |
| results=gr.JSON(label='Generated answers') | |
| gr.Button('Run final challenge').click(run_challenge,outputs=[status,results]) | |
| gr.Button('Submit generated answers').click(submit_challenge,outputs=[results]) | |
| if __name__=='__main__': demo.queue().launch() | |