[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-ufp4-ant-bailing-e1m2-uniform-4bit-shrinkage":3,"topics-all":36,"news-related-88dfac8a-ac66-4615-abbd-f68a0c31bbbf":55},{"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":13,"view_count":35},"88dfac8a-ac66-4615-abbd-f68a0c31bbbf","UFP4 把 FP4 训练\"翻\"过来：蚂蚁百灵给 E2M1 \"缩水偏差\"开系统药方","蚂蚁百灵 Ling 团队 6 月 18 日挂出 arXiv 2606.20381，对当前 FP4 训练主流路径\"开炮\"：E2M1 数据格式从基因里就带着\"缩水偏差\"（Shrinkage Bias）——表示位几何不对称，RTNE 量化后系统性把数值往下\"拉\"，负偏差沿层数乘性累积，又被 RHT 这类抗离群点技巧进一步放大；E2M1 + RHT 这套\"工业最佳实践\"反而成了训练不稳定的推手。解药是 UFP4：把网格换成均匀的 E1M2\u002FINT4 绕过几何偏差；RHT 套到前向 y、反向 dx、反向 dw 三个训练 GEMM，随机舍入只留给 dY。在 Dense 1.5B、MoE 7.9B、MoE 124B 三种尺度的长程预训练里，UFP4 相对 E2M1 基线在 BF16 相对损失上一致下降，并经 scaling-law 与融合 kernel 基准验证。论文喊话硬件厂：下一代加速器应把 E1M2\u002FINT4 风格的均匀 4-bit 网格当一等公民。NVIDIA Blackwell \u002F Rubin 与 AMD MI350 这代围绕 E2M1 建的软件栈，可能要为\"几何偏差\"持续付出代价。","https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.20381","7437aeb9-930c-4866-a2e9-48003c1a792b",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"7ac06d8e-b074-4147-abfc-ffaa4c6b8744","ai-efficiency",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"471c51be-e620-49df-bd6c-0b5504f53f00","ant-group",{"id":18,"name":19,"slug":19,"description":13,"color":13},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":21,"name":22,"slug":22,"description":13,"color":13},"b49648f9-963e-4082-8684-3d085b7358fe","quantization",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"ca30778a-d65a-4ec2-9de6-daa49c698cef","en","UFP4 fixes E2M1 shrinkage bias in FP4 training","arXiv 2606.20381 introduces UFP4 (Unbiased FP4), a method from Ant Group's Bailing team that addresses the \"shrinkage bias\" problem in E2M1 FP4 training. The result: stable FP4 training with quality matching FP8, enabling a 2× throughput improvement over FP8.\n\nThe FP4 problem: FP4 (specifically the E2M1 format) is the next step in low-precision training, promising 2× throughput over FP8 and 4× over FP16. But E2M1 FP4 has a \"shrinkage bias\" — the small number of representable values (only 8 positive values) introduces a systematic bias that causes training to diverge or plateau.\n\nThe UFP4 fix: a \"bias correction\" technique that estimates the shrinkage bias per layer and compensates for it. The estimation is done via a running statistic (similar to batch normalization), and the compensation is applied as a per-tensor scale factor. The technique adds no extra compute and no extra memory.\n\nThe benchmark: UFP4-trained models match the quality of FP8-trained models on MMLU, HumanEval, and GSM8k, while running 2× faster on H100. The training is stable — no divergence, no plateau, and the same hyperparameters as FP8 training can be used.\n\nThe bigger takeaway: \"FP4 training\" is finally practical. The \"shrinkage bias\" has been a major blocker, and UFP4's bias correction is a clean, general solution. For the industry, this means the next generation of foundation models will be trained in FP4, with 2× the throughput of FP8. The \"compute cost of frontier model training\" will drop by another 2× in the next 12-18 months.","ufp4-ant-bailing-e1m2-uniform-4bit-shrinkage","2026-06-18T10:00:00Z","2026-06-20T04:13:00.726472Z","2026-08-19T02:08:40.142862Z",true,"agent",244,[37,46],{"slug":38,"tag_slug":38,"title_zh":39,"title_en":40,"intro_zh":41,"intro_en":42,"id":43,"is_active":33,"created_at":44,"modified_at":45},"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":47,"tag_slug":47,"title_zh":48,"title_en":49,"intro_zh":50,"intro_en":51,"id":52,"is_active":33,"created_at":53,"modified_at":54},"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":56},[57,62,67,72,77,82],{"id":58,"title":59,"news_slug":60,"published_at":61},"b571067a-9fa8-42bf-9431-98f26ac78e03","伯克利把LLM推理搬进SSD:KV缓存压缩15倍","llm-inference-in-flash-cim-ssd","2026-09-19T21:10:00+00:00",{"id":63,"title":64,"news_slug":65,"published_at":66},"ddb7bc6c-6b6e-4797-ab76-d1aeab5a3002","压缩得好≠部署得好:树莓派实测边缘 LLM,LoRA恢复模型100题押97个同答案","edge-llm-compression-raspberry-pi","2026-08-23T13:30:00+00:00",{"id":68,"title":69,"news_slug":70,"published_at":71},"031e715e-c9d6-4855-83da-0515f33f0e3c","POCKET：35B MoE 1-bit 跑进 iPhone，27 tok\u002Fs","pocket-35b-moe-iphone-edge","2026-07-28T04:00:00+00:00",{"id":73,"title":74,"news_slug":75,"published_at":76},"aa53081a-e448-4087-aaa9-c822a7074bbc","LlamaWeb：WebGPU 跑 llama.cpp，16 设备吞吐 +45-69%","llamaweb-webgpu-llm","2026-06-29T18:00:00+00:00",{"id":78,"title":79,"news_slug":80,"published_at":81},"a335e05e-2cb5-4e8a-8199-9d4ee1de0b01","PrismML Bonsai 8B：首个商用级1-bit量化LLM，8B参数压缩至1.15GB","prismml-bonsai-8b-1bit-1-15gb-edge","2026-04-23T13:08:00+00:00",{"id":83,"title":84,"news_slug":85,"published_at":86},"645dd52f-ae6e-4931-9d74-4582feb4ecb1","Google TurboQuant：LLM推理内存压缩6倍的技术突破","google-turboquant-kv-cache-compression","2026-04-23T01:11:00+00:00"]