import os import gradio as gr import requests import pandas as pd import time from typing import Optional, List, Dict # Optional: import openai (pip install openai) import openai # Constants DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space" OPENAI_API_KEY = os.getenv("OPENAI_API_KEY") # Default model - you can change to "gpt-4o" or "gpt-4.1" if available OPENAI_MODEL = os.getenv("OPENAI_MODEL", "gpt-4o") # or "gpt-4.1" if not OPENAI_API_KEY: print("WARNING: OPENAI_API_KEY not set. Set it in Space secrets before running.") openai.api_key = OPENAI_API_KEY # ----------------------------- # Agent implementation (OpenAI-based) # ----------------------------- class OpenAIAgent: """ Minimal agent that uses OpenAI chat completion to answer each question. It is tuned to return *only* the final answer (no extra commentary) so that it matches the EXACT-MATCH submission requirement. """ def __init__(self, model: str = OPENAI_MODEL, temperature: float = 0.0): self.model = model self.temperature = temperature def _build_prompt_messages(self, question_text: str, file_summaries: Optional[List[str]] = None) -> List[Dict]: """ Build messages for chat completion. We instruct the model to output the answer ONLY (single-line), nothing else. No 'Final Answer' phrase. """ system = ( "You are an assistant that MUST produce a single concise answer only. " "When asked a question, respond with the exact answer text only — nothing else. " "Do NOT include explanation, reasoning steps, or any extra punctuation beyond the answer. " "If the question requires a short phrase or number, output that. " "If you do not know, output 'I don't know'." ) user_parts = [f"Question: {question_text}"] if file_summaries: # attach file summaries if provided user_parts.append("File summaries (use these to answer):") user_parts.extend(file_summaries) user = "\n".join(user_parts) return [ {"role": "system", "content": system}, {"role": "user", "content": user}, ] def _call_openai(self, messages: List[Dict], max_tokens: int = 60) -> str: """ Call OpenAI ChatCompletion API (supports gpt-4o / gpt-4.1). Return assistant text. """ if not OPENAI_API_KEY: raise RuntimeError("OPENAI_API_KEY not set in environment.") try: response = openai.ChatCompletion.create( model=self.model, messages=messages, temperature=self.temperature, max_tokens=max_tokens, top_p=1.0, n=1, ) # Extract text (handles typical response structure) text = "" # openai returns choices list with message choices = response.get("choices", []) if choices and "message" in choices[0]: text = choices[0]["message"].get("content", "") else: # fallback for older/newer SDK response shapes text = response["choices"][0]["text"] # trim return text.strip() except Exception as e: # bubble up informative exception for logging raise RuntimeError(f"OpenAI API error: {e}") def summarize_file(self, file_url: str) -> Optional[str]: """ Simple downloader + summarizer placeholder. For text files, fetch content and truncate. For images or other binary files, just return a placeholder note (could be extended). """ try: r = requests.get(file_url, timeout=10) r.raise_for_status() content_type = r.headers.get("Content-Type", "") if "text" in content_type or file_url.lower().endswith((".txt", ".md", ".csv")): text = r.text # keep first 1000 chars to avoid huge prompts summary = text[:1000].replace("\n", " ") return f"[file content preview] {summary}" else: # For non-text file, just inform the model of the file name return f"[file] downloaded from {file_url} (type: {content_type})" except Exception as e: print(f"Warning: Unable to fetch or summarize file {file_url}: {e}") return None def answer(self, question_text: str, files: Optional[List[str]] = None) -> str: """ Main entry: prepare prompt, call model, and return answer string. Ensures we strip quotes/newlines to produce a concise single-line answer. """ file_summaries = [] if files: for furl in files: s = self.summarize_file(furl) if s: file_summaries.append(s) messages = self._build_prompt_messages(question_text, file_summaries if file_summaries else None) raw = self._call_openai(messages, max_tokens=80) # Post-process: keep single-line, strip surrounding quotes, remove trailing punctuation if it's just noise ans = " ".join(raw.splitlines()).strip() # remove wrapping quotes if (ans.startswith('"') and ans.endswith('"')) or (ans.startswith("'") and ans.endswith("'")): ans = ans[1:-1].strip() # final safety: if empty, return "I don't know" if not ans: ans = "I don't know" return ans # ----------------------------- # Runner / UI glue (kept similar to original) # ----------------------------- def run_and_submit_all(profile: gr.OAuthProfile | None): """ Fetches all questions, runs the OpenAIAgent on them, submits all answers, and returns status string and results DataFrame. """ space_id = os.getenv("SPACE_ID") if profile: username = f"{profile.username}" else: return "Please Login to Hugging Face with the button.", None api_url = DEFAULT_API_URL questions_url = f"{api_url}/questions" submit_url = f"{api_url}/submit" # instantiate agent try: agent = OpenAIAgent() except Exception as e: return f"Error initializing agent: {e}", None # agent_code repo URL agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main" if space_id else "" # fetch questions try: resp = requests.get(questions_url, timeout=15) resp.raise_for_status() questions_data = resp.json() except Exception as e: return f"Error fetching questions: {e}", None if not questions_data: return "No questions returned from server.", None results_log = [] answers_payload = [] for item in questions_data: task_id = item.get("task_id") question_text = item.get("question") files = item.get("files") or [] if not task_id or question_text is None: continue try: ans = agent.answer(question_text, files) except Exception as e: ans = "I don't know" results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": f"AGENT ERROR: {e}"}) else: results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": ans}) answers_payload.append({"task_id": task_id, "submitted_answer": ans}) # small sleep to avoid rate limits time.sleep(0.5) if not answers_payload: return "Agent produced no answers.", pd.DataFrame(results_log) submission_data = {"username": username.strip(), "agent_code": agent_code, "answers": answers_payload} # submit try: r = requests.post(submit_url, json=submission_data, timeout=60) r.raise_for_status() result_data = r.json() final_status = ( f"Submission Successful!\n" f"User: {result_data.get('username')}\n" f"Overall Score: {result_data.get('score', 'N/A')}% " f"({result_data.get('correct_count', '?')}/{result_data.get('total_attempted', '?')} correct)\n" f"Message: {result_data.get('message', 'No message received.')}" ) return final_status, pd.DataFrame(results_log) except requests.exceptions.HTTPError as e: try: text = e.response.text except: text = str(e) return f"Submission failed: {text}", pd.DataFrame(results_log) except Exception as e: return f"Submission failed: {e}", pd.DataFrame(results_log) # ----------------------------- # Build Gradio Interface # ----------------------------- with gr.Blocks() as demo: gr.Markdown("# Basic Agent Evaluation Runner (OpenAI-based)") gr.Markdown( """ Instructions: 1. Add your OpenAI key as a secret named `OPENAI_API_KEY` in this Space. 2. Ensure requirements.txt contains `openai`. 3. Login, then click 'Run Evaluation & Submit All Answers'. """ ) gr.LoginButton() run_button = gr.Button("Run Evaluation & Submit All Answers") status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False) results_table = gr.DataFrame(label="Questions and Agent Answers", wrap=True) run_button.click(fn=run_and_submit_all, outputs=[status_output, results_table]) if __name__ == "__main__": demo.launch(debug=True, share=False)