Causal Attention with Lookahead Keys
This addresses a bottleneck in language modeling for NLP applications, offering incremental improvements over standard methods.
The paper tackled the limitation of static keys in causal attention by introducing CASTLE, which updates keys with later context while preserving autoregressive properties, resulting in reduced validation perplexity and improved downstream task performance across model scales.
In standard causal attention, each token's query, key, and value (QKV) are static and encode only preceding context. We introduce CAuSal aTtention with Lookahead kEys (CASTLE), an attention mechanism that continually updates each token's keys as the context unfolds. We term these updated keys lookahead keys because they belong to earlier positions yet integrate information from tokens that appear later relative to those positions, while strictly preserving the autoregressive property. Although the mechanism appears sequential, we derive a mathematical equivalence that avoids explicitly materializing lookahead keys at each position and enables efficient parallel training. On language modeling benchmarks, CASTLE consistently outperforms standard causal attention across model scales, reducing validation perplexity and improving performance on a range of downstream tasks.