The authors evaluated two diffusion language models on three mathematical reasoning benchmarks.
Hidden-state trajectories help diffusion language models escape wrong answers
LOCKR detects when iterative reasoning has stabilized incorrectly, then selectively explores and verifies targeted repair branches.
Big Tech
Guoshenghui Zhao · Tan Yu · Weijie Zhao
Rochester Institute of Technology · NVIDIA Corporation
Research Digest··2 min read
Zhao, Yu and Zhao study stable-but-wrong lock-in, where a diffusion language model settles early on an incorrect answer despite having denoising steps left.
Why this paper
From NVIDIA Corporation and Rochester Institute of Technology
In one line
Hidden-state trajectories enable detecting and repairing stable-but-wrong lock-in in diffusion language models, yielding accuracy gains of 2.21-5.37 percentage points.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§