Request to add GTrXL: Stabilizing Transformers for Reinforcement Learning

[ashok-arora]

Model description

Gated Transformer-XL (GTrXL) improves upon Transformer-XL by addressing its instability in reinforcement learning. It introduces gating mechanisms similar to LSTMs, which help stabilize training and improve long-term memory retention. This makes GTrXL more effective in partially observable environments where capturing long-range dependencies is crucial.

Open source status

• The model implementation is available
• The model weights are available

Provide useful links for the implementation

• https://github.com/opendilab/PPOxFamily/blob/main/chapter5_time/gtrxl.py
• https://github.com/alantess/gtrxl-torch