TQL: Scaling Q-Functions with Transformers by Preventing Attention Collapse

Despite scale driving substantial recent advancements in machine learning, reinforcement learning (RL) methods still primarily use small value functions. Naively scaling value functions -- including with a transformer architecture, which is known to be highly scalable -- often results in learning instability and worse performance. In this work, we ask what prevents transformers from scaling effectively for value functions? Through empirical analysis, we identify the critical failure mode in this scaling: attention scores collapse as capacity increases. Our key insight is that we can effectively prevent this collapse and stabilize training by controlling the entropy of the attention scores, thereby enabling the use of larger models. To this end, we propose Transformer Q-Learning (TQL), a method that unlocks the scaling potential of transformers in learning value functions in RL. Our approach yields up to a 43\% improvement in performance when scaling from the smallest to the largest network sizes, while prior methods suffer from performance degradation.

Scaling value functions in reinforcement learning remains a longstanding challenge — larger models often hurt performance rather than help it unlike other areas of deep learning, even for highly scalable architectures such as transformers. We identify the root cause of this failure in transformer-based value functions, namely attention entropy collapse. To address it, we propose Transformer Q-Learning (TQL), a method that unlocks the scaling potential of transformers for value function learning in RL.