English(EN) UC San Diego researchers have developed DFlash, a block diffusion model that drafts whole token blocks in a single pass for speculative decoding. The technique

DFlash 通过并行起草令牌块来加速 LLM 推理

作者 PulseAugur 编辑部 · [2 个来源] · 2026-06-24 07:21

加州大学圣迭戈分校的研究人员开发了 DFlash，这是一种新颖的推测性解码方法，可显著加速大型语言模型推理。与之前一次起草一个令牌的方法不同，DFlash 使用轻量级块扩散模型并行提出整个令牌块。据报道，与 EAGLE-3 等现有技术相比，该方法在各种模型和任务上实现了超过 6 倍的无损加速，并且在 NVIDIA Blackwell GPU 上对 GPT-OSS 120B 的吞吐量提高了 15 倍。 AI

影响 DFlash 的并行块起草可以显著降低 LLM 推理成本和延迟，从而实现更复杂和交互式的 AI 应用。

排序理由介绍 LLM 推理加速新方法的论文。[lever_c_demoted from research: ic=1 ai=1.0]

在 Mastodon — fosstodon.org 阅读 →

AI 生成摘要 · Google Gemini · 来自 2 个来源。我们如何撰写摘要 →

报道来源 [2]

MarkTechPost TIER_1 English(EN) · Asif Razzaq · 2026-06-24 07:21

DFlash Speculative Decoding Drafts Whole Token Blocks in Parallel for Up to 15x Higher Throughput on NVIDIA Blackwell

<p>UC San Diego's DFlash replaces autoregressive drafting with a lightweight block diffusion model for speculative decoding. It drafts whole token blocks in a single forward pass and conditions on target hidden features through KV injection. The paper reports up to 6.08x lossless…
Mastodon — fosstodon.org TIER_1 English(EN) · [email protected] · 2026-06-24 07:55

UC San Diego researchers have developed DFlash, a block diffusion model that drafts whole token blocks in a single pass for speculative decoding. The technique

UC San Diego researchers have developed DFlash, a block diffusion model that drafts whole token blocks in a single pass for speculative decoding. The technique delivers up to 15x higher throughput on NVIDIA Blackwell GPUs compared to traditional autoregressive methods. It works b…

报道来源 [2]

DFlash Speculative Decoding Drafts Whole Token Blocks in Parallel for Up to 15x Higher Throughput on NVIDIA Blackwell

UC San Diego researchers have developed DFlash, a block diffusion model that drafts whole token blocks in a single pass for speculative decoding. The technique

相关实体

相关话题