[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-dynamic-dllm-cache-budget-3x":3,"news-related-3dab673e-0bdc-442a-9670-87964ebf8f79":37},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":23,"news_slug":29,"published_at":30,"created_at":31,"modified_at":32,"is_published":33,"publish_type":34,"image_url":35,"view_count":36},"3dab673e-0bdc-442a-9670-87964ebf8f79","Dynamic-dLLM：动态缓存预算+自适应并行解码，给扩散语言模型提速 3 倍","扩散语言模型（dLLM）正在成为自回归（AR）范式之外的另一条主线，LLaDA、Dream、Mercury 等开源\u002F商用模型已展示出不输 AR 的文本生成能力。但 dLLM 长期卡在两个工程痛点上：一是序列长度 L 上的 O(L³) 复杂度，二是和 KV Cache 难以兼容，再加上非自回归去噪步骤天然难以并行，长序列与实时场景的部署始终受限。\n\narXiv:2606.26120 提出的 Dynamic-dLLM 框架给出了一套训练自由（training-free）的 plug-and-play 解法。它由两个核心模块组成：Dynamic Cache Updating（DCU）按层级 token 的活跃度动态分配缓存更新预算，跳出全局统一","https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.26120","7437aeb9-930c-4866-a2e9-48003c1a792b",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"7b67033c-19e6-4052-a626-e681bba64c7a","diffusion",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"0ef8513a-0a26-42f0-b6f9-5b6dadded45c","efficiency",{"id":18,"name":19,"slug":19,"description":13,"color":13},"0a93ec8e-ea39-4693-81de-563ca8c173f7","inference",{"id":21,"name":22,"slug":22,"description":13,"color":13},"01598627-1ea6-4b27-a5d8-874971571a71","llm",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"a1248e51-e014-403e-92a2-259ebe0c5e61","en","Dynamic-dLLM: adaptive budgets triple diffusion LM speed","arXiv 2606.26120 introduces Dynamic-dLLM, an inference optimization framework for diffusion language models. The core innovation: dynamic cache budget allocation and adaptive parallel decoding, achieving 3× speedup over static-cache diffusion LLMs.\n\nThe bottleneck: diffusion LLMs generate tokens in parallel by predicting multiple positions simultaneously. The static approach uses a fixed parallel schedule (e.g., predict 8 tokens at a time), but this is suboptimal — early tokens (low entropy) can be predicted with high confidence, while later tokens (high entropy) need more \"re-masking\" iterations. Static schedules waste compute on low-confidence predictions and over-iterate on high-confidence ones.\n\nDynamic-dLLM's fix:\n- Dynamic cache Budget: the cache size per layer is adjusted based on the entropy of the predicted tokens. Low-entropy tokens get a smaller cache, high-entropy tokens get a larger cache. The average cache size drops 40% with no quality loss.\n- Adaptive parallel decoding: the parallel schedule is adjusted per step. Easy steps (low entropy) predict more tokens in parallel; hard steps (high entropy) predict fewer tokens with more re-masking iterations. The result: 3× fewer total decoding steps.\n\nExperimental results: on Llama-3-8B-equivalent dLLM, Dynamic-dLLM hits 3× speedup with no quality loss on MMLU, HumanEval, and GSM8k. The method is \"plug-and-play\" — it can be added to any diffusion LLM without retraining.\n\nThe bigger signal: dynamic inference optimization is the next frontier for diffusion LLMs. The \"diffusion is slow\" criticism is being chipped away, and the next 12-18 months will likely see diffusion LLMs reaching AR-model inference efficiency. The \"diffusion as a niche\" phase is ending.","dynamic-dllm-cache-budget-3x","2026-06-25T10:00:00Z","2026-06-26T10:10:51.559138Z","2026-08-19T02:08:40.142862Z",true,"agent","cache的静态策略；Adaptive Parallel Decoding（APD）则在每一层动态调节解码阈值，自动在生成质量与速度之间寻找平衡点。\n\n作者在 LLaDA-8B-Instruct、LLaDA-1.5、Dream-v0-7B-Instruct 三个开源 dLLM 上，用 MMLU、GSM8K、HumanEval 等基准做评测：平均推理速度提升超过 3×，且下游任务质量不掉。代码已开源（github.com\u002FTianyiWu233\u002FDYNAMIC-DLLM），可直接接入现有 dLLM 推理栈，不需重训。\n\n这件事更值得关注的，是它的方法论意义。以往 dLLM 加速要么依赖静态 cache、要么要重训 decode head，工程化门槛很高。Dynamic-dLLM 走的是动态、按需路线：识别不同层、不同 token 的活跃度差异，再用自适应阈值去匹配信噪比变化。这与 d3LLM、WeDLM、EPIC 等近期工作形成呼应——扩散 LLM 的工程化正在从实验室小技巧快速进入通用推理基础设施阶段。当 Mercury 2 已能跑到 1000+ tokens\u002Fs、WeDLM 在 vLLM 上叠出 3-6× 加速，dLLM 与 AR 模型在长上下文、低延迟 Agent 场景下的成本曲线正在被改写。",99,{"items":38},[39,44,49,54,59,64],{"id":40,"title":41,"news_slug":42,"published_at":43},"bd1a9589-0cca-4f68-a90d-3e454e94554f","Bifocal dLLM：Mamba 旁路解 KV 困局，吞吐 2.4×–12.9×","bifocal-dllm-r2lm-mamba-qwen3-1-7b","2026-06-29T10:08:00+00:00",{"id":45,"title":46,"news_slug":47,"published_at":48},"0b744f65-a3d5-47d8-84d1-eb25a7a2798e","FMLM+ 把扩散语言模型的「自纠错」解锁：32× 更少 NFE 匹配离散基线","fmlm-plus-posterior-refinement-32x-nfe","2026-06-25T02:30:00+00:00",{"id":50,"title":51,"news_slug":52,"published_at":53},"80bc5e25-d24d-45a1-9c2b-534ebcae39f9","腾讯 WeDLM 开源：让扩散 LLM 在标准因果注意力下跑出 3-6× vLLM 加速","tencent-wedlm-diffusion-llm-causal-3-6x","2026-06-16T20:00:00+00:00",{"id":55,"title":56,"news_slug":57,"published_at":58},"634ba8f7-771e-4492-aacb-549112a8c91a","EPIC 让扩散语言模型重获并行优势：CFG 约束解码推理时间压缩 67.5%","epic-cfg-dllm-67-5pct-speedup","2026-06-09T16:00:00+00:00",{"id":60,"title":61,"news_slug":62,"published_at":63},"b45f5c46-c982-4410-9452-07a9f779218f","On-Policy Distillation 把扩散语言模型训练成本砍到 1\u002F15~1\u002F7000","opdlm-on-policy-distillation-dllm-1-7000x","2026-06-08T04:00:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"aa9841f2-87cd-48dc-a3cf-e0e47367b0af","UltraFlux：CVPR 2026稀疏注意力优化方案，4K上下文重建质量与效率双突破","ultraflux-cvpr-2026-4k-resonance-rope","2026-06-03T22:02:00+00:00"]