Metraipolis › archive › dc_bbfb6d70303d

Fork-dLLM: Avoiding the Flexibility Trap in Diffusion Language Models — arXiv:2609.39859

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.39859
License (as archived)
CC0-1.0
Archived
2026-10-07T13:46:45.808Z
SHA-256 of the archived text
5af942262078a0f051a2f61b379c89557be9b570d9c11ccc705551c08770a800
Stable URL of this page
https://metraipolis.com/documento/dc_bbfb6d70303d

How to cite

Fork-dLLM: Avoiding the Flexibility Trap in Diffusion Language Models — arXiv:2609.39859 — arXiv (en), https://arxiv.org/abs/2609.39859 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_bbfb6d70303d (Metraipolis archive dc_bbfb6d70303d, sha256 5af942262078a0f051a2f61b379c89557be9b570d9c11ccc705551c08770a800)
BibTeX
@misc{dcbbfb6d70303d,
  title = {Fork-dLLM: Avoiding the Flexibility Trap in Diffusion Language Models — arXiv:2609.39859},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.39859},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_bbfb6d70303d; sha256 5af942262078a0f051a2f61b379c89557be9b570d9c11ccc705551c08770a800},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.39859. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Masked diffusion language models (dLLMs) have shown strong potential for faster inference through parallel token generation when combined with confidence-based samplers. However, recent work has shown that such methods can defer unmasking high-entropy fork positions at which multiple plausible continuations exist. This results in reduced generation diversity, as shown by worse pass@k scaling, and limits gains obtainable from RL post-training. To avoid this flexibility trap, prior work advocated for autoregressive (AR) sampling. Here, we show that discarding confidence-based sampling is unnecessary and, once inference cost is taken into account, wasteful. We first propose Fork-dLLM, a simple hybrid sampler that uses AR-style ordering only at uncertain fallback steps while retaining parallel generation otherwise. We then extend the same principle to post-training with ForkGRPO, which uses Fork-dLLM rollouts and applies the GRPO objective only at fallback steps, preserving exact policy-likelihood ratios while substantially reducing rollout and optimization cost. In our experiments, Fork-dLLM matches the strong pass@k scaling of AR sampling while being 2-3x more efficient, and ForkGRPO achieves downstream performance comparable to or better than AR-based GRPO baselines at a substantially lower training cost.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-10-01 in cs.LG).
Categorie arXiv: Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2609.39859 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-07T13:18:10.480Z dalla pagina https://arxiv.org/catchup/cs.LG/2026-10-01?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).