TSR Desk · science · 10 September 2026, 01:00 UTC
In-Place Instruction Following in Diffusion Language Models
- What
- In-Place Instruction Following in Diffusion Language Models
- Who
- arxiv.org
- When
- 9 September 2026, 04:00 UTC
- Category
- Science
- Primary source
- https://arxiv.org/abs/2609.07160
- What is not known
- This brief does not claim independent replication. Claims that appear only on X and not in the primary source stay unknown.
On four representative dLLMs, GRAFT raises the average IIF score from 57.75 to 73.10 (+15.35 points), with absolute gains of 15.91 and 15.57 points on literal and discourse-function constraints, while preserving general generation ability. It comes from a paper posted to arXiv on 9 September 2026. Diffusion Large Language Models (dLLMs) generate text via bidirectional iterative denoising, naturally supporting user-specified constraints anchored at arbitrary output positions, a paradigm known as In-place Prompting (IPP). We formalize this as the In-place Instruction Following (IIF) task and construct IIF-Bench, a hierarchical benchmark spanning literal, style, and discourse-function constraints, paired with a rubric-based local-global evaluation protocol. An inference-time attention-bias probe suggests that vanilla dLLMs often under-prioritize constraint spans during denoising. We then propose GRAFT, an IPP-oriented post-training framework combining constraint-aware SFT and preference optimization.
Why it counts
On four representative dLLMs, GRAFT raises the average IIF score from 57.75 to 73.10 (+15.35 points), with absolute gains of 15.91 and 15.57 points on literal and discourse-function constraints, while preserving general generation ability.
Sources
Primary source: primary source
What is not known
This brief does not claim independent replication. Claims that appear only on X and not in the primary source stay unknown.
No clip. The article still stands.