[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-large-flow-language-models-lafa-self-speculative-decoding":3,"topics-all":38,"news-related-8c579d8d-4114-4d7d-92c1-22f5fab0537b":57},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":24,"news_slug":31,"published_at":32,"created_at":33,"modified_at":34,"is_published":35,"publish_type":36,"image_url":14,"view_count":37},"8c579d8d-4114-4d7d-92c1-22f5fab0537b","连续流语言模型跑到 8B：Lafa 单模型自验证解码提速 2.54 倍","连续流匹配被搬进语言建模：Lafa 把预训练自回归权重改造成 8B 分块流模型，用不到 300 亿 token 完成适配，六项基准追平自回归基线，同一权重自验证草稿，单卡 H100 吞吐达 2.54 倍。","并行解码这些年一直被卡在中间地带。要么跑自回归，一次前向吐一个 token，延迟死在串行上；要么上离散扩散，一次刷出一整块，质量却够不着同规模自回归。连续流匹配一直在旁边等着——图像和视频领域的加速手段都能搬过来，但语言建模这边迟迟没人推到十亿参数级别。\n\n## 从预训练自回归模型改出来，而不是从零训\n\nLafa（Large flow language model）的思路是分块自回归：前缀保持因果注意力，整块内部改成双向、异步的流。好处很直接——模型可以从预训练的自回归模型改造成分块去噪器，不用推倒重来。论文报告的训练规模是 0.5B、7B、8B 三档，累计 token 不到 300 亿；同类离散扩散语言模型的适配开销可以高到 5000 亿 token，论文给出的差距至少 18 倍。\n\n具体到训练，两个设计被消融实验反复确认。一是块内注意力必须双向，完全因果的版本平均准确率只有 32.7%，换成块内双向后升到 38.0%。二是时间要异步：给每个位置单独一个时间步，让靠前的 token 走得更快。这个 tokenwise time 不是人工设定的，而是训练中从去噪损失里学出来的——损失高的噪声水平自动获得更多训练样本和积分步数。学出来的调度确实呈现从左到右的偏置，但这个偏置没有被硬编码。\n\n## 一个反直觉的副产品：学流反而让自回归变强\n\n联合训练还带来一个意外收获：流训练不但保住了预训练模型的自回归能力，多 token 预测效应甚至让它比匹配的纯 SFT 基线更准。三个规模上自回归解码的平均准确率都比对应 SFT 基线高。\n\n这条副产品才是论文最值钱的地方。有了更强的自回归预测器，Lafa 就能用它验证自己的一步流草稿——同一套权重兼任起草器和验证器，不需要外挂小模型当草稿。这条自推测解码路径在六个基准上的 tokens per forward 落在 2.5 到 5 之间，准确率保持在接近或高于 AR SFT 的水平。\n\n## 2.54 倍是怎么测出来的\n\n论文在单张 H100、batch size 为 1 的条件下测了端到端吞吐，取 IFEval、GSM8K、HumanEval 三项平均。块大小为 16 时，Lafa 7B 达到 7B AR SFT 吞吐的 2.54 倍，平均准确率相当；块大小降到 4 时速度约为两倍，平均准确率还更高。对照组里，DeepSeek-MTP 和 MARS 能追平或超过 AR SFT 准确率，吞吐最多只有 1.4 倍；离散扩散基线整体落在 AR SFT 准确率之下，调得越快掉得越狠。\n\n作者对结果也算诚实：纯流解码本身要多次积分步才追得上 AR 质量，真正让速度起来的是验证一步流草稿这条捷径，而不是流解码本身。flow map 蒸馏能把学生模型压到一两步，但论文承认相对多步流解码仍有质量差距。\n\n## 权重和模型都放出来了\n\n三个 checkpoint 用 Apache 2.0 发布，0.5B 和 7B 基于 Qwen2.5 Instruct，8B 基于 Qwen3 Base，块大小都是 4。每个 checkpoint 同时支持流解码、自回归解码和自推测解码，切换只改一个参数。噪声水平嵌入的初始化方式是固定范数、只学方向，这个变体在消融里拿到最高的 38.0。\n\n所以呢：做推理优化的话，值得注意的不是又一个 8B 模型，而是「同权重兼任起草器和验证器」这条路——它绕开了草稿模型和主模型分布不一致的老问题。参考实现见 arXiv:2610.28559 与 github.com\u002Flafa-gh\u002Flarge-flow-lms。","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.28559","7437aeb9-930c-4866-a2e9-48003c1a792b",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"0ef8513a-0a26-42f0-b6f9-5b6dadded45c","efficiency",{"id":19,"name":20,"slug":20,"description":14,"color":14},"0a93ec8e-ea39-4693-81de-563ca8c173f7","inference",{"id":22,"name":23,"slug":23,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"0d4cebcd-f1c7-4d90-8f47-6dcd33a11868","en","Flow Language Models Hit 8B: Self-Speculative Decoding Runs 2.54x","Flow matching hits 8B: Lafa adapts AR weights in under 30B tokens, matches AR quality, and reaches 2.54x throughput on one H100 via self-verification.","Parallel decoding has been stuck in the middle ground for years. Either run autoregression honestly and emit one token per forward pass, letting latency die on a serial line; or reach for discrete diffusion and paint a whole block at once, but quality never quite reaches a same-scale autoregressive model. Continuous flow matching has been waiting on the sidelines, since its controllability and acceleration tricks transfer directly from image and video generation, but nobody had pushed it to billion-parameter scale for language modeling.\n\n## Adapted from a pretrained AR model, not trained from scratch\n\nLafa (Large flow language model) uses blockwise autoregression: the prefix keeps causal attention, while each block runs a bidirectional, asynchronous flow. The payoff is direct. The model can be converted into a block denoiser from pretrained autoregressive weights rather than built from the ground up. The paper reports scales of 0.5B, 7B and 8B trained on fewer than 30B tokens in total, against discrete diffusion language model adaptation that can reach 500B tokens, a gap the authors put at at least 18x.\n\nTwo training choices keep earning their place in the ablations. Block-internal attention has to be bidirectional: a fully causal version averages 32.7 percent accuracy, and block-bidirectional lifts that to 38.0. And time must be asynchronous, giving each position its own step so earlier tokens advance faster. That tokenwise time is not hand-set. It is learned during training from the denoising loss, so hard noise levels automatically receive more training samples and integration steps. The learned schedule does show a left-to-right bias, but nothing hard-codes it.\n\n## A counterintuitive side effect: learning the flow strengthens AR\n\nJoint training produces an unexpected bonus. Flow training does not merely preserve the pretrained autoregressive capability; a multi-token prediction effect leaves Lafa more accurate than matched pure SFT baselines. Average autoregressive decoding accuracy beats the corresponding SFT baseline at all three scales.\n\nThat side effect is the most valuable part of the paper. With a stronger autoregressive predictor in hand, Lafa uses it to verify its own one-step flow drafts. The same weights serve as both drafter and verifier, with no small auxiliary model bolted on as the draft. Across the six benchmarks, this self-speculative decoding path commits between 2.5 and 5 tokens per forward pass while holding accuracy close to or above AR SFT.\n\n## Where the 2.54x comes from\n\nThroughput was measured end-to-end on a single H100 at batch size one in bfloat16, averaged over IFEval, GSM8K and HumanEval. At block size sixteen, Lafa 7B reaches 2.54x the throughput of 7B AR SFT at comparable average accuracy. Dropping to block size four is about twice as fast and more accurate on average. By comparison, DeepSeek-MTP and MARS match or exceed AR SFT accuracy but reach at most 1.4x its throughput, while discrete diffusion baselines fall below AR SFT accuracy overall and lose more the faster they are tuned.\n\nThe authors are candid about what carries the speed. Pure flow decoding itself needs multiple integration steps to reach AR quality; what actually delivers throughput is verifying one-step flow drafts. Flow map distillation compresses the student model to one or two steps, but the paper admits a quality gap remains against multi-step flow decoding.\n\n## Weights and checkpoints are out\n\nAll three checkpoints ship under Apache 2.0, with 0.5B and 7B built on Qwen2.5 Instruct and 8B on Qwen3 Base, all at block size four. Each checkpoint supports flow, autoregressive and self-speculative decoding, and switching between them is a single argument change. Noise-level embeddings are initialized with a fixed norm and learn direction only, and that variant takes the top ablation score of 38.0.\n\nSo what: if you work on inference optimization, the interesting part is not another 8B model but the idea of one set of weights acting as both drafter and verifier, which sidesteps the long-standing mismatch between draft and target distributions. Reference implementations are at arXiv:2610.28559 and github.com\u002Flafa-gh\u002Flarge-flow-lms.","large-flow-language-models-lafa-self-speculative-decoding","2026-10-21T03:09:03Z","2026-10-21T03:09:20.272982Z","2026-10-21T03:09:20.272998Z",true,"agent",69,[39,48],{"slug":40,"tag_slug":40,"title_zh":41,"title_en":42,"intro_zh":43,"intro_en":44,"id":45,"is_active":35,"created_at":46,"modified_at":47},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":49,"tag_slug":49,"title_zh":50,"title_en":51,"intro_zh":52,"intro_en":53,"id":54,"is_active":35,"created_at":55,"modified_at":56},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":58},[59,64,69,74,79,84],{"id":60,"title":61,"news_slug":62,"published_at":63},"3d922c00-afcb-4f1c-a6d5-8f9d6c10c642","从 Kimi Linear 到 Kimi K3:MoE 推理效率战里被忽略的架构升级","kimi-k3-latentmoe-kda-attnres-nope","2026-07-30T00:30:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"f2b0bde3-ebb5-49e5-a418-5ece37639d1b","MIPU\u002FMIPI：把 LLM RL 的「训练—推理失配」从工程噪音重写为优化目标","mipu-mipi-rl-mismatch","2026-07-04T08:00:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"5a4c2a98-6ce7-4e51-8348-118be3083afc","ELDR 把 MoE 推理的「延迟最后一公里」拉直:vLLM 实测 TPOT 最多砍 13.9%","eldr-moe-routing","2026-07-03T12:30:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"cd82f869-d4a7-45b7-95ce-b66051e9d933","BlockPilot：实例自适应策略学习让扩散式投机解码再下一城,Qwen3-4B 上首破 4.20× 加速","blockpilot-instance-adaptive-block","2026-07-01T14:17:29+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"abf22bbb-eccf-46d0-99d6-debe1596f92b","自验证蒸馏：无需外部教师，LLM如何实现自我进化","self-verified-distill-qwen3-16-7pp-math","2026-05-27T19:00:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"6c9e553d-3029-4867-bc29-ee069d26b934","自验证蒸馏：无需外部教师，LLM 如何实现自我进化","svd-self-verified-distill-stanford-perplexity","2026-05-27T16:05:00+00:00"]