Search papers, labs, and topics across Lattice.
School of Computer Science, Shanghai Jiao Tong University, Shanghai, 200240, China
3
0
3
Joint commitment in dLLMs can lead to substantial accuracy gains, especially in reasoning tasks, by coordinating token selection more effectively.
Self-generated error training reduces T2T edit intensity and boosts accuracy, tackling critical failure modes in language model outputs.
Masking, not replacement, unlocks better error correction in diffusion language models by providing a cleaner context for subsequent denoising steps.
