Metraipolis › archive › dc_f8011d4e758c
Step-dLLM: Adaptive Step-aware Sparse Attention for Efficient Diffusion LLM Inference — arXiv:2611.01320
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2611.01320
- License (as archived)
- CC0-1.0
- Archived
- 2026-11-03T10:26:20.006Z
- SHA-256 of the archived text
1b684c6e1c7ecc6f46c74b964df5a969994b627a53dfda4d762ca5661d3deb12- Stable URL of this page
https://metraipolis.com/documento/dc_f8011d4e758c
How to cite
Step-dLLM: Adaptive Step-aware Sparse Attention for Efficient Diffusion LLM Inference — arXiv:2611.01320 — arXiv (en), https://arxiv.org/abs/2611.01320 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_f8011d4e758c (Metraipolis archive dc_f8011d4e758c, sha256 1b684c6e1c7ecc6f46c74b964df5a969994b627a53dfda4d762ca5661d3deb12)
BibTeX
@misc{dcf8011d4e758c,
title = {Step-dLLM: Adaptive Step-aware Sparse Attention for Efficient Diffusion LLM Inference — arXiv:2611.01320},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2611.01320},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_f8011d4e758c; sha256 1b684c6e1c7ecc6f46c74b964df5a969994b627a53dfda4d762ca5661d3deb12},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.01320. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Diffusion large language models (dLLMs) generate tokens in parallel via iterative denoising, but their bidirectional attention prevents the KV caching used by autoregressive models and recomputes the full attention map at every step, incurring quadratic per-step cost. Existing sparse attention methods for dLLMs construct a sparse mask after an initial full-attention warm-up and reuse it for the remaining steps, while allocating the sparsity budget uniformly across heads. We argue both choices ignore key structural properties of attention in dLLMs: empirically, attention patterns are only locally consistent across steps and shift abruptly at certain transitions, and different heads concentrate their attention mass to very different extents. To this end, we design Step-dLLM, which periodically refreshes a block-sparse mask at anchor steps along the denoising trajectory and reuses it within each locally consistent interval, while employing a head-adaptive global budget that more aggressively compresses heads with concentrated attention. Backed by a fused IO-aware Triton kernel for block-sparse attention, Step-dLLM applies to pre-trained LLaDA-1.5 and Dream-7B-Instruct without retraining, achieving a 2.78x end-to-end speedup over dense FlashAttention on LLaDA-1.5 at 8K context with 256 denoising steps and 90% sparsity, while improving the accuracy-latency trade-off over prior sparse-attention methods. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno). Categorie arXiv: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2611.01320 — 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 2026-11-03T10:19:50.070Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).