GPU-native multi-agent RL environments and algorithms in JAX from Foerster's FLAIR (FLAIROx) — standard MARL tooling (~835 stars), published at NeurIPS 2024 Datasets & Benchmarks and actively developed through 2026. End-to-end JIT-compiled environments make multi-agent experiments orders of magnitude faster than CPU-bound predecessors.

Paper

Library

Language Python
Framework JAX
infrastructurereinforcement-learningresearch

Related