Feature Request: GAIA Evaluation Support

[dark4scope]

📊 Background

ToolOrchestra currently ranks #1 on the GAIA benchmark, but the evaluation code is not available in the repository. Adding GAIA evaluation support would greatly benefit the community.

🎯 Desired Features

1. GAIA Dataset Loading: Support for loading GAIA test/validation splits
2. Evaluation Script: Script to evaluate models on GAIA benchmark
3. Result Formatting: Output format compatible with GAIA leaderboard submission
4. Documentation: Usage instructions and examples

💡 Suggested Implementation

• Add scripts/evaluate_gaia.py for running evaluations
• Include dataset loaders in data/ directory
• Provide configuration files for different GAIA difficulty levels
• Add result parser to format outputs for leaderboard submission

📝 Additional Context

This would make it easier for researchers to:

• Reproduce the requirements.txt contains many NVIDIA-internal or unavailable packages #1 GAIA benchmark results
• Compare their approaches against ToolOrchestra
• Contribute improvements to the evaluation pipeline