Spaces:
Running on Zero
Running on Zero
| import os | |
| import requests | |
| import pandas as pd | |
| import gradio as gr | |
| import spaces | |
| from io import BytesIO | |
| from smolagents import CodeAgent, DuckDuckGoSearchTool, OpenAIServerModel | |
| API_URL = "https://agents-course-unit4-scoring.hf.space" | |
| AGENT_CODE_URL = "https://huggingface.co/spaces/bacchi87/Final_Assignment_Template/tree/main" | |
| # Gemini model setup | |
| model = OpenAIServerModel( | |
| model_id="gemini-3.5-flash", | |
| api_base="https://generativelanguage.googleapis.com/v1beta/openai/", | |
| api_key=os.environ["GEMINI_API_KEY"], | |
| max_tokens=1500 | |
| ) | |
| agent = CodeAgent( | |
| tools=[DuckDuckGoSearchTool()], | |
| model=model, | |
| additional_authorized_imports=[ | |
| "math", "re", "json", "datetime", "pandas" | |
| ], | |
| max_steps=6 | |
| ) | |
| def test_gemini(): | |
| try: | |
| result = agent.run( | |
| "Reply with exactly GEMINI_OK. Do not use tools." | |
| ) | |
| return f"Gemini connection successful!\nResponse: {result}" | |
| except Exception as e: | |
| return f"Gemini connection failed:\n{e}" | |
| def get_attachment(task_id): | |
| try: | |
| response = requests.get( | |
| f"{API_URL}/files/{task_id}", timeout=30 | |
| ) | |
| response.raise_for_status() | |
| data = BytesIO(response.content) | |
| try: | |
| df = pd.read_excel(data) | |
| return df.to_string(index=False) | |
| except Exception: | |
| data.seek(0) | |
| try: | |
| df = pd.read_csv(data) | |
| return df.to_string(index=False) | |
| except Exception: | |
| return "Attachment format is not readable." | |
| except Exception as e: | |
| return f"Attachment download failed: {e}" | |
| def answer_question(question, attachment=""): | |
| prompt = f""" | |
| Answer this question accurately. | |
| Question: | |
| {question} | |
| Attached file data: | |
| {attachment} | |
| Use web search or calculations when required. | |
| Return ONLY the answer. | |
| Do not write FINAL ANSWER. | |
| """ | |
| try: | |
| result = agent.run(prompt) | |
| return str(result).strip() | |
| except Exception as e: | |
| return f"Agent error: {e}" | |
| def run_evaluation(profile: gr.OAuthProfile | None): | |
| if profile is None: | |
| return "Please login with Hugging Face.", [] | |
| username = profile.username | |
| log = f"Logged in as: {username}\n" | |
| try: | |
| response = requests.get( | |
| f"{API_URL}/questions", timeout=30 | |
| ) | |
| response.raise_for_status() | |
| questions = response.json() | |
| if not questions: | |
| return "No questions returned.", [] | |
| except Exception as e: | |
| return f"Could not fetch questions: {e}", [] | |
| answers = [] | |
| for i, item in enumerate(questions, start=1): | |
| task_id = str(item["task_id"]) | |
| question = item["question"] | |
| log += f"\nQuestion {i}/{len(questions)}: {task_id}\n" | |
| attachment = "" | |
| if item.get("file_name"): | |
| attachment = get_attachment(task_id) | |
| answer = answer_question(question, attachment) | |
| # Stop if Gemini fails | |
| if answer.startswith("Agent error:"): | |
| log += f"Model inference failed: {answer}\n" | |
| return log + "\nNo answers submitted.", answers | |
| answers.append({ | |
| "task_id": task_id, | |
| "submitted_answer": answer | |
| }) | |
| log += f"Answer: {answer}\n" | |
| # Official scoring API | |
| payload = { | |
| "username": username, | |
| "agent_code": AGENT_CODE_URL, | |
| "answers": answers | |
| } | |
| try: | |
| result = requests.post( | |
| f"{API_URL}/submit", | |
| json=payload, | |
| timeout=60 | |
| ) | |
| result.raise_for_status() | |
| score_data = result.json() | |
| log += "\n--- EVALUATION RESULT ---\n" | |
| log += f"Score: {score_data.get('score')}%\n" | |
| log += f"Correct: {score_data.get('correct_count')}\n" | |
| log += f"Attempted: {score_data.get('total_attempted')}\n" | |
| log += f"Message: {score_data.get('message')}\n" | |
| except Exception as e: | |
| log += f"\nSubmission failed: {e}\n" | |
| return log, answers | |
| with gr.Blocks() as demo: | |
| gr.Markdown("# Hugging Face Unit 4 - GAIA Evaluation") | |
| gr.LoginButton() | |
| test_btn = gr.Button("Test Gemini Connection") | |
| test_output = gr.Textbox(label="Gemini Test Result") | |
| test_btn.click( | |
| fn=test_gemini, | |
| inputs=[], | |
| outputs=[test_output] | |
| ) | |
| run_btn = gr.Button( | |
| "Run Evaluation & Submit Answers", | |
| variant="primary" | |
| ) | |
| status_box = gr.Textbox( | |
| label="Evaluation Status", | |
| lines=18 | |
| ) | |
| answers_box = gr.JSON(label="Generated Answers") | |
| run_btn.click( | |
| fn=run_evaluation, | |
| inputs=[], | |
| outputs=[status_box, answers_box] | |
| ) | |
| if __name__ == "__main__": | |
| demo.launch() |