Environments that simulate real-world tasks for training and evaluating LLM collaboration:
Writing: Multiple LLM agents collaborate to process articles.
Coding: Collaborative code generation for programming problems.
- MBPP - Mostly basic Python problems.
- HumanEval - Handwritten evaluation problems.
- CoopHumanEval - A cooperative variant of HumanEval.
Travel: Collaborative travel planning under user constraints.
- TravelPlanner - Planning multi-day trip itineraries.
Tool-Use: Collaborative function calling on BFCL v4.
- Native Parallel - Making parallel function calls.