|
|
--- |
|
|
title: GAIA Agent for Hugging Face Agents Course |
|
|
emoji: 🕵🏻♂️ |
|
|
colorFrom: indigo |
|
|
colorTo: indigo |
|
|
sdk: gradio |
|
|
sdk_version: 5.25.2 |
|
|
app_file: app.py |
|
|
pinned: false |
|
|
hf_oauth: true |
|
|
|
|
|
hf_oauth_expiration_minutes: 480 |
|
|
--- |
|
|
|
|
|
Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference |
|
|
|
|
|
# GAIA Agent for Hugging Face Agents Course |
|
|
|
|
|
This project implements a powerful intelligent agent using the SmolAgents framework to tackle the GAIA benchmark questions for the Hugging Face Agents course final assessment. |
|
|
|
|
|
## Project Overview |
|
|
|
|
|
The GAIA benchmark consists of challenging questions that require an agent to use various tools, including web search, file processing, and reasoning capabilities. This agent is designed to: |
|
|
|
|
|
1. Receive questions from the GAIA API |
|
|
2. Process and understand the questions |
|
|
3. Use appropriate tools to find answers |
|
|
4. Format and return precise answers |
|
|
|
|
|
## Features |
|
|
|
|
|
- **SmolAgents Integration**: Uses CodeAgent for flexible problem-solving with Python code execution |
|
|
- **Multi-Model Support**: |
|
|
- Compatible with Hugging Face models |
|
|
- OpenAI models (GPT-4o and others) |
|
|
- X.AI's Grok models |
|
|
- Anthropic, Cohere, and Mistral models via LiteLLM |
|
|
- **Enhanced Tool Suite**: |
|
|
- Web search via DuckDuckGo |
|
|
- Python interpreter for code execution |
|
|
- File handling (reading, saving, downloading) |
|
|
- Data analysis for CSV and Excel files |
|
|
- Image processing with OCR capabilities (when available) |
|
|
- **Flexible Environment Configuration**: |
|
|
- Easy setup via environment variables or .env file |
|
|
- Fallback mechanisms for missing dependencies |
|
|
- Support for both local and secure E2B code execution |
|
|
- **Answer Processing**: |
|
|
- Special handling for reversed text questions |
|
|
- Precise answer formatting for benchmark submission |
|
|
- Automatic cleanup of model responses for exact matching |
|
|
- **Interactive UI**: Gradio interface for running the agent and submitting answers |
|
|
|
|
|
## Setup |
|
|
|
|
|
### Prerequisites |
|
|
|
|
|
- Python 3.8+ |
|
|
- Hugging Face account |
|
|
- API keys for your preferred models (HuggingFace, OpenAI, X.AI, etc.) |
|
|
|
|
|
### Installation |
|
|
|
|
|
1. Clone this repository |
|
|
2. Install the required dependencies: |
|
|
|
|
|
```bash |
|
|
pip install -r requirements.txt |
|
|
``` |
|
|
|
|
|
3. Copy the example environment file and add your API keys: |
|
|
|
|
|
```bash |
|
|
cp env.example .env |
|
|
# Edit .env with your API keys and configuration |
|
|
``` |
|
|
|
|
|
### Configuration |
|
|
|
|
|
Configure the agent by setting these environment variables or editing the `.env` file: |
|
|
|
|
|
#### API Keys |
|
|
``` |
|
|
HUGGINGFACEHUB_API_TOKEN=your_huggingface_token_here |
|
|
OPENAI_API_KEY=your_openai_key_here |
|
|
XAI_API_KEY=your_xai_api_key_here # For X.AI/Grok models |
|
|
``` |
|
|
|
|
|
#### Agent Configuration |
|
|
``` |
|
|
AGENT_MODEL_TYPE=OpenAIServerModel # HfApiModel, InferenceClientModel, LiteLLMModel, OpenAIServerModel |
|
|
AGENT_MODEL_ID=gpt-4o # Model ID depends on the model type |
|
|
AGENT_TEMPERATURE=0.2 |
|
|
AGENT_EXECUTOR_TYPE=local # local or e2b for secure execution |
|
|
AGENT_VERBOSE=true # Set to true for detailed logging |
|
|
``` |
|
|
|
|
|
#### Advanced Configuration |
|
|
``` |
|
|
AGENT_PROVIDER=hf-inference # Provider for InferenceClientModel |
|
|
AGENT_TIMEOUT=120 # Timeout in seconds for API calls |
|
|
AGENT_API_BASE=https://api.groq.com/openai/v1 # For X.AI when using OpenAIServerModel |
|
|
``` |
|
|
|
|
|
## Usage |
|
|
|
|
|
### Running the Agent |
|
|
|
|
|
Launch the Gradio interface with: |
|
|
|
|
|
```bash |
|
|
python app.py |
|
|
``` |
|
|
|
|
|
Then: |
|
|
1. Log in to your Hugging Face account using the button in the interface |
|
|
2. Click "Run Evaluation & Submit All Answers" |
|
|
|
|
|
### Testing |
|
|
|
|
|
To test the agent with sample questions before running the full evaluation: |
|
|
|
|
|
```bash |
|
|
python test_agent.py |
|
|
``` |
|
|
|
|
|
For more focused testing with specific APIs: |
|
|
|
|
|
```bash |
|
|
python test_groq_api.py # Test X.AI/Groq API integration |
|
|
python test_xai_api.py # Test X.AI API integration |
|
|
``` |
|
|
|
|
|
## Project Structure |
|
|
|
|
|
- `app.py`: Main application with Gradio interface |
|
|
- `core_agent.py`: Agent implementation with SmolAgents framework |
|
|
- `api_integration.py`: Client for interacting with GAIA API |
|
|
- `test_agent.py`: Testing script with sample questions |
|
|
- `test_groq_api.py` & `test_xai_api.py`: API-specific test scripts |
|
|
- `update_groq_key.py`: Utility for updating API keys |
|
|
- `project_planning.md`: Development roadmap and progress tracking |
|
|
- `requirements.txt`: Project dependencies |
|
|
|
|
|
## Tools Implementation |
|
|
|
|
|
The agent includes several custom tools: |
|
|
|
|
|
1. **save_and_read_file**: Save content to a temporary file and return the path |
|
|
2. **download_file_from_url**: Download a file from a URL and save it locally |
|
|
3. **extract_text_from_image**: OCR for extracting text from images (requires pytesseract) |
|
|
4. **analyze_csv_file**: Load and analyze CSV files using pandas |
|
|
5. **analyze_excel_file**: Load and analyze Excel files using pandas |
|
|
|
|
|
## Resources |
|
|
|
|
|
- [GAIA Benchmark Information](https://huggingface.co/spaces/gaia-benchmark/leaderboard) |
|
|
- [SmolAgents Documentation](https://huggingface.co/docs/smolagents/en/index) |
|
|
- [Hugging Face Agents Course](https://huggingface.co/agents-course) |
|
|
|
|
|
## License |
|
|
|
|
|
This project is licensed under the MIT License. |
