Metraipolis › archive › dc_84fdf84eddbc

FastGuide: Accelerating Reward Guidance for Diffusion Large Language Models — arXiv:2609.36202

Source: arXiv (en)
Original: https://arxiv.org/abs/2609.36202
License (as archived): CC0-1.0
Archived: 2026-10-07T16:56:25.426Z
SHA-256 of the archived text: 1a2eb4ea78d7fd29d80da85244d37a88f2ad17673e28e0be81035cab8cb6e504
Stable URL of this page: https://metraipolis.com/documento/dc_84fdf84eddbc

How to cite

FastGuide: Accelerating Reward Guidance for Diffusion Large Language Models — arXiv:2609.36202 — arXiv (en), https://arxiv.org/abs/2609.36202 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_84fdf84eddbc (Metraipolis archive dc_84fdf84eddbc, sha256 1a2eb4ea78d7fd29d80da85244d37a88f2ad17673e28e0be81035cab8cb6e504)

BibTeX

@misc{dc84fdf84eddbc,
  title = {FastGuide: Accelerating Reward Guidance for Diffusion Large Language Models — arXiv:2609.36202},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.36202},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_84fdf84eddbc; sha256 1a2eb4ea78d7fd29d80da85244d37a88f2ad17673e28e0be81035cab8cb6e504},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.36202. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Gradient-based reward guidance provides a flexible way to use downstream reward models to control masked diffusion language models at inference time. However, its computational cost remains high as each decoding iteration incurs expensive diffusion model forward passes and reward model backpropagation steps. To address this, we introduce FastGuide, an adaptive hybrid of parallel and autoregressive decoding to accelerate reward guidance for diffusion language models. In analogy to parallel decoding, FastGuide amortizes the cost of reward model backpropagation by computing guidance once per decoding step and reusing it to generate multiple tokens. Within each decoding step, FastGuide makes diffusion forward passes autoregressive by unmasking tokens one at a time while efficiently recomputing token distributions after each unmasking by utilizing KV caching techniques and sparse recomputation of attention. Lastly, to adapt hybrid decoding to the model's confidence, FastGuide defers any token that the model is unconfident about under its recomputed distribution. Experiments on three reward benchmarks demonstrate that FastGuide is up to $4.4\times$ faster than sequential reward-guided decoding while retaining similar generation quality.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-30 in cs.CL).
Categorie arXiv: Computation and Language (cs.CL) ; Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2609.36202 — 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 2026-10-07T16:27:03.184Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-30?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).