Spaces:
Sleeping
Sleeping
| # app.py | |
| import gradio as gr | |
| import os | |
| from typing import List, Dict | |
| import json | |
| # Import our modules | |
| try: | |
| from agent.local_llm import LocalLLM | |
| from agent.tools import gaia_tools | |
| from utils.gaia_api import GaiaAPI | |
| from llama_index.core.agent import ReActAgent | |
| from llama_index.core.memory import ChatMemoryBuffer | |
| # Initialize components | |
| print("Initializing Local LLM...") | |
| local_llm = LocalLLM() | |
| llm = local_llm.get_llm() | |
| print("Creating ReAct Agent...") | |
| memory = ChatMemoryBuffer.from_defaults(token_limit=2000) | |
| agent = ReActAgent.from_tools( | |
| tools=gaia_tools, | |
| llm=llm, | |
| memory=memory, | |
| verbose=True, | |
| max_iterations=3 # Limit iterations to avoid long processing | |
| ) | |
| print("Agent initialized successfully!") | |
| AGENT_READY = True | |
| except Exception as e: | |
| print(f"Failed to initialize agent: {str(e)}") | |
| AGENT_READY = False | |
| agent = None | |
| def process_single_question(question_text: str) -> str: | |
| """Process a single GAIA question through the agent""" | |
| if not AGENT_READY: | |
| return "❌ Agent not ready. Please check the logs for initialization errors." | |
| try: | |
| # Add instruction to give direct answers only | |
| enhanced_prompt = f""" | |
| Answer the following question directly and concisely. Do not include "FINAL ANSWER" or any other prefixes in your response. Just provide the answer. | |
| Question: {question_text} | |
| """ | |
| response = agent.query(enhanced_prompt) | |
| # Clean the response to ensure it's just the answer | |
| answer = str(response).strip() | |
| # Remove common prefixes that might appear | |
| prefixes_to_remove = ["FINAL ANSWER:", "Answer:", "The answer is:", "Final answer:"] | |
| for prefix in prefixes_to_remove: | |
| if answer.startswith(prefix): | |
| answer = answer[len(prefix):].strip() | |
| return answer | |
| except Exception as e: | |
| return f"❌ Error processing question: {str(e)}" | |
| def process_all_questions() -> str: | |
| """Process all GAIA questions and prepare answers for submission""" | |
| if not AGENT_READY: | |
| return "❌ Agent not ready. Cannot process questions." | |
| try: | |
| questions = GaiaAPI.get_questions() | |
| processed_answers = [] | |
| for i, question in enumerate(questions): | |
| print(f"Processing question {i+1}/{len(questions)}: {question['task_id']}") | |
| answer = process_single_question(question['question']) | |
| processed_answers.append({ | |
| "task_id": question['task_id'], | |
| "submitted_answer": answer | |
| }) | |
| # Save answers to file for review | |
| with open("gaia_answers.json", "w") as f: | |
| json.dump(processed_answers, f, indent=2) | |
| summary = f"✅ Processed {len(processed_answers)} questions.\n" | |
| summary += f"Answers saved to gaia_answers.json\n" | |
| summary += f"First 3 answers:\n" | |
| for ans in processed_answers[:3]: | |
| summary += f"- {ans['task_id']}: {ans['submitted_answer'][:50]}...\n" | |
| return summary | |
| except Exception as e: | |
| return f"❌ Error processing all questions: {str(e)}" | |
| def submit_to_gaia(username: str, code_url: str) -> str: | |
| """Submit answers to GAIA benchmark""" | |
| if not AGENT_READY: | |
| return "❌ Agent not ready. Cannot submit." | |
| if not username or not code_url: | |
| return "❌ Please provide both username and code URL." | |
| try: | |
| # Load processed answers | |
| try: | |
| with open("gaia_answers.json", "r") as f: | |
| answers = json.load(f) | |
| except FileNotFoundError: | |
| return "❌ No processed answers found. Please process questions first." | |
| # Submit to GAIA | |
| result = GaiaAPI.submit_answers(username, code_url, answers) | |
| if "error" in result: | |
| return f"❌ Submission failed: {result['error']}" | |
| score = result.get('score', 'Unknown') | |
| return f"✅ Submission successful!\n📊 Score: {score}\n🎯 Check the leaderboard for your ranking!" | |
| except Exception as e: | |
| return f"❌ Submission error: {str(e)}" | |
| def get_sample_question() -> str: | |
| """Load a sample question for testing""" | |
| try: | |
| question = GaiaAPI.get_random_question() | |
| return question['question'] | |
| except Exception as e: | |
| return f"Error loading sample question: {str(e)}" | |
| # Create Gradio interface | |
| with gr.Blocks(title="🦙 GAIA LlamaIndex Agent") as demo: | |
| gr.Markdown(""" | |
| # 🦙 GAIA Benchmark Agent with LlamaIndex | |
| This agent uses LlamaIndex with a local LLM to tackle GAIA benchmark questions. | |
| **Status:** {"✅ Ready" if AGENT_READY else "❌ Not Ready"} | |
| """) | |
| with gr.Tab("🔬 Test Single Question"): | |
| gr.Markdown("Test the agent with individual questions") | |
| with gr.Row(): | |
| with gr.Column(): | |
| question_input = gr.Textbox( | |
| label="Question", | |
| placeholder="Enter a GAIA question or click 'Load Sample'", | |
| lines=3 | |
| ) | |
| with gr.Row(): | |
| sample_btn = gr.Button("🎲 Load Sample Question") | |
| process_btn = gr.Button("🚀 Process Question", variant="primary") | |
| with gr.Column(): | |
| answer_output = gr.Textbox( | |
| label="Agent Answer", | |
| lines=5, | |
| interactive=False | |
| ) | |
| sample_btn.click(get_sample_question, outputs=question_input) | |
| process_btn.click(process_single_question, inputs=question_input, outputs=answer_output) | |
| with gr.Tab("📊 Full Evaluation"): | |
| gr.Markdown("Process all GAIA questions and prepare for submission") | |
| with gr.Row(): | |
| process_all_btn = gr.Button("🔄 Process All Questions", variant="primary") | |
| processing_output = gr.Textbox( | |
| label="Processing Status", | |
| lines=10, | |
| interactive=False | |
| ) | |
| process_all_btn.click(process_all_questions, outputs=processing_output) | |
| with gr.Tab("🏆 Submit to GAIA"): | |
| gr.Markdown(""" | |
| Submit your processed answers to the GAIA benchmark for official scoring. | |
| **Requirements:** | |
| 1. Your Hugging Face username | |
| 2. Link to your Space code (e.g., `https://huggingface.co/spaces/YOUR_USERNAME/gaia-llamaindex-agent/tree/main`) | |
| 3. Questions must be processed first in the "Full Evaluation" tab | |
| """) | |
| with gr.Row(): | |
| with gr.Column(): | |
| username_input = gr.Textbox( | |
| label="HF Username", | |
| placeholder="your-username" | |
| ) | |
| code_url_input = gr.Textbox( | |
| label="Space Code URL", | |
| placeholder="https://huggingface.co/spaces/your-username/gaia-llamaindex-agent/tree/main" | |
| ) | |
| submit_btn = gr.Button("🎯 Submit to GAIA", variant="primary") | |
| with gr.Column(): | |
| submission_output = gr.Textbox( | |
| label="Submission Result", | |
| lines=5, | |
| interactive=False | |
| ) | |
| submit_btn.click( | |
| submit_to_gaia, | |
| inputs=[username_input, code_url_input], | |
| outputs=submission_output | |
| ) | |
| with gr.Tab("ℹ️ Info"): | |
| gr.Markdown(""" | |
| ## About This Agent | |
| This agent combines: | |
| - **LlamaIndex**: For orchestrating the agent workflow | |
| - **Local LLM**: Running entirely on Hugging Face Spaces | |
| - **ReAct Framework**: For reasoning and acting iteratively | |
| - **GAIA Tools**: Web search, calculation, file reading, etc. | |
| ## Usage Tips | |
| 1. **Start with single questions** to test the agent | |
| 2. **Process all questions** when ready for full evaluation | |
| 3. **Submit to GAIA** for official scoring | |
| ## Troubleshooting | |
| - If agent fails to initialize, check the model loading | |
| - For memory issues, try restarting the Space | |
| - For API errors, verify the GAIA endpoint URL | |
| """) | |
| if __name__ == "__main__": | |
| demo.launch(show_error=True) |