Li-Xuan Wei

This work proposes an early-bird decoding framework, motivated by the observation that tokens with similarly low entropy tend to cluster and can be jointly decoded earlier, before reaching the confidence threshold, and integrates two key enablers: a learnable network that adaptively groups tokens with similar uncertainty into variable-length blocks, rather than relying on fixed block sizes.

Discrete diffusion language models can now achieve higher accuracy without retraining the entire backbone, thanks to a lightweight recurrent memory module that bridges denoising steps.