Blockwise SFT for Diffusion Language Models: Reconciling Bidirectional Attention and Autoregressive Decoding
By: Bowen Sun , Yujun Cai , Ming-Hsuan Yang and more
Potential Business Impact:
Teaches AI to write better by training it block by block.
Discrete diffusion language models have shown strong potential for text generation, yet standard supervised fine-tuning (SFT) misaligns with their semi-autoregressive inference: training randomly masks tokens across the entire response, while inference generates fixed-size blocks sequentially. This mismatch introduces noisy prefixes and leaky suffixes, biasing gradients away from the desired blockwise likelihood. We propose Blockwise SFT, which partitions responses into fixed-size blocks, selects one active block per step for stochastic masking, freezes all preceding tokens, and fully hides future ones. Loss is computed only over the active block, directly mirroring the blockwise decoding process. Experiments on GSM8K, MATH, and MetaMathQA show consistent gains over classical SFT under equal compute or token budgets. Block size consistency studies and ablations confirm that improvements stem from faithful training-inference alignment rather than incidental masking effects. Our results highlight the importance of matching supervision granularity to the decoding procedure in diffusion-based language models.
Similar Papers
WeFT: Weighted Entropy-driven Fine-Tuning for dLLMs
Computation and Language
Makes AI better at solving puzzles and math.
Fast and Fluent Diffusion Language Models via Convolutional Decoding and Rejective Fine-tuning
Computation and Language
Makes AI write better and faster.
Improved Supervised Fine-Tuning for Large Language Models to Mitigate Catastrophic Forgetting
Computation and Language
Keeps AI smart while teaching it new tricks.