Locally Coherent Parallel Decoding in Diffusion Language Models

In three linesCoDiLA combines diffusion with local autoregressive decoding for parallel code generation. A compact auxiliary AR model (0.6B parameters) ensures syntactic coherence on diffusion latents, eliminating artifacts while preserving bidirectional generation and sub-linear latency.

Summary generated by Claude — human-verified