[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-steerdistill-quantized-agent-teacher-steering":3,"topics-all":38,"news-related-76bc2964-2411-42b7-a8b9-389bbc6d417e":67},{"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},"76bc2964-2411-42b7-a8b9-389bbc6d417e","SteerDistill:量化Agent失手,老师闭环来救","低比特量化降低了 Agent 的推理成本，却常让模型在关键步骤漏看任务要求或环境条件，一步走错整单失败。清华 AIR 的 SteerDistill 用冻结的全精度教师按执行反馈引导量化学生自我修正，验证过的修正再蒸馏进参数；三个模型、五个基准上，W4A16 平均分与 BF16 差距小于 1 分。","把 Agent 的大模型量化到 4 比特，推理成本确实降下来了，但任务成功率常常悄悄往下掉。11 月 10 日提交到 arXiv 的 SteerDistill（2611.13410）先做了一次病理分析：量化后的 Agent 不是「整体变笨」，而是在多轮任务的某一步上看漏任务要求或环境条件，环境状态从此偏离全精度参考轨迹——而多轮任务里一步走错，往往就足以把整单任务拖垮。论文把这类偏移叫行为分歧，修复它正是整个方法的靶心。\n\n## 失败为什么难修\n\n论文指出两个难点：离线训练数据大概率覆盖不到学生模型真正出错的上下文；让学生自己去试，它在这些上下文里又很少能碰巧给出正确回答。传统蒸馏需要的「正确示范」，恰好都不在这两个来源里。\n\n## 闭环教师指导\n\nSteerDistill 出自清华大学智能产业研究院（AIR），作者六人。做法是把全精度模型变成随叫随到的老师：教师与学生从同一个 BF16 checkpoint 出发，教师全程冻结。学生从出错前的历史与环境状态重试那个「失手」的决策，教师给出针对性指导；修正是否算数，由真实执行结果判定——不通过，教师就按执行反馈把指导再细化，直到修正通过局部验证。整个训练按轮次推进：每一轮用冻结的参数快照在低比特运行时里生成回放与修正，更新后的主权重再量化成下一轮的快照；学生则在训练中用假量化前向和直通估计器维护 BF16 主权重。\n\n## 把修正内化进参数\n\n部署时不能拖着一个全精度教师。SteerDistill 在无指导状态下训练学生：修正过的轮次对齐「被带教后的学生」的预测，其余轮次对齐 BF16 教师的预测。关键在于，修正发生在学生自己的失败上下文里，而不是外部数据的分布里；即使整个任务最终没有解出来，那些通过局部验证的修正也会被保留。推理时学生独自运行，不需要教师，也不需要指导提示。\n\n## 成绩单（论文自报口径）\n\n- W4A16（权重 4 比特、激活 16 比特）下，三个模型 × 五个 Agent 基准，平均分超过最强量化基线 1.67–2.18 分，与 BF16 差距保持在 1 分以内；\n- τ²-Bench 上成功率比 QeRL 高 1.84 个百分点，训练 GPU 时数省 67%，成功轨迹也更短；\n- 更激进的 W2A16（权重 2 比特）下，Qwen3-8B 的平均分在被测量化基线中最高；\n- 对照方法包括 BitDistiller、On-Policy QAD 和 QeRL：On-Policy QAD 按 NVIDIA NeMo-RL 的量化感知蒸馏工作流取教师 top-64 logits，QeRL 则基于 GRPO 做强化学习。\n\n## 两点冷静剂\n\n其一，这是 11 月 10 日刚挂出的预印本，所有数字都是论文自报口径，尚无独立复现。其二，代码已按 MIT 许可证放上 GitHub，checkpoint 一栏却还是 TODO——想直接下权重复现的读者得再等等；README 的快速上手示例是 Qwen3-8B 跑 BFCL V4。\n\n量化掉分在这篇论文里不再是一个笼统的统计现象，而是能定位到具体失败轮次、能现场修复、能蒸馏回参数的工程流程。对想把 Agent 塞进边缘设备的团队，这条路的意义很直接：省下来的算力，不必再用成功率去偿还。论文：https:\u002F\u002Farxiv.org\u002Fabs\u002F2611.13410 ；代码：https:\u002F\u002Fgithub.com\u002Fair-embodied-brain\u002FSteerDistill","https:\u002F\u002Farxiv.org\u002Fabs\u002F2611.13410","7437aeb9-930c-4866-a2e9-48003c1a792b",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"6ad31a14-c0da-42df-81fd-564281f768db","agentic-ai",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},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":22,"name":23,"slug":23,"description":14,"color":14},"b49648f9-963e-4082-8684-3d085b7358fe","quantization",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"21003c2b-b462-4a17-a762-a5e9402cc061","en","SteerDistill: Teacher Steering Rescues Sub-4-Bit Agents","SteerDistill (Tsinghua AIR) has a frozen BF16 teacher fix quantized agents at their failure turns, then distills the verified corrections into the student.","Quantizing an agent's LLM to 4-bit cuts inference cost, but task success often quietly slides. SteerDistill (arXiv:2611.13410, submitted November 10, 2026) starts with a fine-grained diagnosis: a quantized agent does not get uniformly dumber. Instead, at one step of a multi-turn task it overlooks a task requirement or an environmental condition, and the environment state then diverges from the full-precision reference trajectory. In a multi-turn task, one wrong step is often enough to fail the whole task. The paper calls these drifts behavioral divergences, and fixing them is the entire point of the method.\n\n## Why the failures are hard to fix\n\nThe paper identifies two obstacles. Offline training data is unlikely to cover the exact contexts where the student model fails; letting the student roll out on its own rarely produces correct responses in those same contexts. The \"correct demonstration\" that classic distillation needs is absent from both sources.\n\n## Closed-loop teacher steering\n\nSteerDistill comes from the Institute for AI Industry Research (AIR) at Tsinghua University, with six authors. The idea is to turn the full-precision model into an on-demand teacher. Teacher and student start from the same BF16 checkpoint, and the teacher stays frozen throughout. The student retries the divergent turn from the pre-error history and environment state; the teacher gives targeted guidance; whether a revision counts is judged by actually executing it — if it fails, the teacher refines the guidance using execution feedback until the correction passes local verification. Training proceeds in cycles: each cycle uses a frozen parameter snapshot in the low-bit runtime for rollout generation and revision, and the updated master weights are quantized into the next cycle's snapshot. During training, the student runs fake quantization in its forward pass and keeps BF16 master weights via a straight-through estimator.\n\n## Internalizing the corrections\n\nYou cannot ship a full-precision teacher alongside every deployment. SteerDistill trains the student without guidance: at corrected turns it matches the predictions of the steered student, and at other turns it matches the BF16 teacher. The corrections happen in the student's own failure contexts rather than in the distribution of external data; even when the whole task remains unsolved, locally verified corrections are still retained. At inference, the student runs alone — no teacher, no steering prompts.\n\n## The numbers (as reported by the paper)\n\n- Under W4A16 (4-bit weights, 16-bit activations), across three models and five agentic benchmarks, SteerDistill exceeds the strongest quantized baseline by 1.67–2.18 points on average and stays within one point of BF16.\n- On τ²-Bench, it beats QeRL by 1.84 percentage points in success rate while using 67% fewer training GPU hours, and its successful trajectories are shorter than the quantized baselines.\n- Under the more aggressive W2A16 (2-bit weights), it achieves the highest average score among the tested quantized baselines on Qwen3-8B.\n- Baselines include BitDistiller, On-Policy QAD, and QeRL: On-Policy QAD follows NVIDIA NeMo-RL's quantization-aware distillation workflow with top-64 teacher logits, while QeRL takes the GRPO reinforcement-learning route.\n\n## Two caveats\n\nFirst, this is a preprint posted on November 10; every number above is self-reported by the paper, with no independent replication yet. Second, the code is on GitHub under an MIT license, but the checkpoints entry still reads TODO — readers who want to reproduce with the actual weights will have to wait. The README quickstart walks through Qwen3-8B on BFCL V4.\n\nThe paper turns \"quantization hurts agents\" from a blunt statistical observation into an engineering pipeline: locate the specific failing turn, fix it on the spot with a teacher in the loop, and distill the fix back into the parameters. For teams trying to squeeze agents onto edge devices, that path matters — the compute you save no longer has to be repaid in success rate. Paper: https:\u002F\u002Farxiv.org\u002Fabs\u002F2611.13410 — Code: https:\u002F\u002Fgithub.com\u002Fair-embodied-brain\u002FSteerDistill","steerdistill-quantized-agent-teacher-steering","2026-11-12T19:06:48Z","2026-11-12T19:06:48.265051Z","2026-11-12T19:06:48.265060Z",true,"agent",230,[39,49,58],{"slug":40,"tag_slug":41,"title_zh":42,"title_en":43,"intro_zh":44,"intro_en":45,"id":46,"is_active":35,"created_at":47,"modified_at":48},"image-generation","text-to-image","图像生成：从文生图到逐框编辑的模型竞赛","Image Generation: from text-to-image to frame-level editing","文生图已是红海，战场正转向编辑与可控性：FLUX.2\u002F3 用开源和量化把生图搬进消费级显卡，Qwen-Image、Seedream 等国产模型紧咬不放，GPT Image 2 则把推理能力带进像素。本专题追踪图像生成与编辑模型的发布、架构演进、加速量化与商业化数据，看「好图」如何变成「可用的图」。","Text-to-image is a crowded race, and the frontier has moved to editing and control: FLUX.2\u002F3 push open weights and quantization onto consumer GPUs, Qwen-Image and Seedream close in from China, and GPT Image 2 brings reasoning into pixels. This topic tracks model releases, architecture shifts, acceleration work, and the business numbers behind \"pretty\" becoming \"usable\".","52de7c59-7864-4569-8ae7-3d656d26a1a1","2026-11-20T07:20:23.534506Z","2026-11-20T07:20:23.534527Z",{"slug":50,"tag_slug":50,"title_zh":51,"title_en":52,"intro_zh":53,"intro_en":54,"id":55,"is_active":35,"created_at":56,"modified_at":57},"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":59,"tag_slug":59,"title_zh":60,"title_en":61,"intro_zh":62,"intro_en":63,"id":64,"is_active":35,"created_at":65,"modified_at":66},"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":68},[69,74,79,84,89,94],{"id":70,"title":71,"news_slug":72,"published_at":73},"1ec34a26-f012-4bc4-8193-14e651c07516","程序进化搜出记忆设计:涨12个百分点,成本只占35%","mape-agent-memory-program-evolution","2026-11-14T05:05:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"b80697dc-e9ee-40f5-a2c7-3944cd36abba","权重不等长:哈夫曼重编码,边缘LLM再省31.4%","lazyweights-huffman-edge-llm","2026-11-13T23:30:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"1c9b3e02-69f2-452f-9137-ff9c2ddcf7fb","手机上的模型工厂:小模型自动定制,五项反超前沿模型","mobile-ai-factory-on-device-slm","2026-11-12T23:15:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"081148f2-39eb-432d-8841-5cb1e58ed527","省内存还是多跑几遍:本地小模型怎么花推理预算才划算","quantization-bits-vs-option-order-budget","2026-11-12T09:10:00+00:00",{"id":90,"title":91,"news_slug":92,"published_at":93},"a6f5abcc-f1eb-4b6c-ba50-062836f50d0d","量化前先算一遍:二阶筛查排序能信,绝对值不能","quantization-fisher-screen-pre-registered-audit","2026-11-11T03:30:00+00:00",{"id":95,"title":96,"news_slug":97,"published_at":98},"62bc171e-2954-4c08-b3e8-d484acce7894","回放AUC 0.93,实跑却分不出:本地小模型升位门控","uncertainty-gate-local-llm-replay-vs-live","2026-11-10T17:57:51+00:00"]