[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-nvidia-edgewise-edge-cloud-slm-self-delegation-coding":3,"topics-all":38,"news-related-74cb8016-bdf7-4c8b-adbc-456d8c899b1a":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},"74cb8016-bdf7-4c8b-adbc-456d8c899b1a","EdgeWise 把 9B 模型训练成「喊人」高手:边云协同砍半成本","NVIDIA 提出 EdgeWise 边云协作框架:Qwen3.5-9B 本地当主力,关键步骤才求助云端大模型。论文分析 2.4k 条编码 Agent 轨迹后用步级 RL 训 9B 模型自学何时委派,SWE-Rebench 推理成本最多降 62%,平均 46%,性能与全程用大模型相当。","长程编码 Agent 有个尴尬的成本结构:一个任务动辄几十上百步,业内惯常做法是把每一步都丢给云端前沿大模型。NVIDIA 的研究团队最近在 arXiv 上挂出论文 EdgeWise,开头就戳破这层窗户纸——不是每一步都需要大模型的能力,但每一步都在按大模型的价格付费。\n\n## 云端大模型的价值,集中在少数关键步骤\n\n论文先做了一个观察实验:分析 2.4k 条编码 Agent 轨迹,把每一步归入六类操作(写、读、搜、命令、思考、其它),然后随机抽 40% 的步骤路由到云端 GPT-5.5 或 Claude Opus 4.8。结果既出人意料又在情理之中:即便云端覆盖比例固定到 40%,性能却和「每步都用云端」相差无几,有时甚至更好。配合皮尔逊相关分析,发现真正与最终结果强相关的只有写、复杂读这两类——「在哪个步骤请求云端」比「总共请求多少次」更关键。\n\n这个观察直接把 EdgeWise 的设计锚定下来:本地跑一个小模型扛下绝大多数动作,云端只在写代码和复杂阅读这种关键节点被点醒,剩下的步骤按本地便宜算力结算。\n\n## 步级 RL 让 9B 模型学会「喊人」\n\nEdgeWise 提供两套委托策略。第一套叫二元委托:每一步跑完都过一道外部裁判模型,根据这一步的难度和重要程度决定本地要不要喊云端。裁判是事后调度,稳健但每一轨都要算两次账。\n\n第二套是论文重点——自委托。Qwen3.5-9B 先用监督微调学一遍,再用步级强化学习把「什么时候该喊云端」压进模型权重。奖励来自最终任务结果,经过细粒度的信用分配回传到每一次委派决策。换句话说,模型在训练时就知道「这一招喊晚了吃亏,喊早了浪费」。推理时不需要再带裁判,本地 9B 模型自己就能拍板要不要把这一步丢给云端。\n\n论文同时给出第三种范式:上下文管理。云端被叫来时,本地应该把整段轨迹都喂给它还是只喂当下状态?EdgeWise 给出统一答案——完整的交互历史通常表现最好,代价是 token 贵一点;如果你想再抠一点成本,可以筛掉无关步骤。\n\n## 数字:53% \u002F 63% \u002F 50%,三组关键节流\n\n主表里的对照选 Qwen3.5-9B 当本地主力,云端分别接 GPT-5.5 和 Opus 4.8,任务最长 80 步。在 SWE-ReBench 上,自委托把云端成本砍掉 53%(配 GPT-5.5)和 63%(配 Opus 4.8);在 Terminal-Bench 2.1 上分别砍 35% 和 33%。Pass@3 与全程云端基本持平,平均差距写出来就是 0.0%。\n\n跨 4 个 harness(自训练用 Terminus2,评估再跨 OpenHands、π、mini-swe-agent)做迁移,自委托在没见过的 harness 上依然把成本压下来 25% 以上——说明「何时喊云端」这个能力是 harness 无关的元技能,不绑定具体框架。\n\n最后落到真机:EdgeWise 跑在 DGX Spark 本地工作站上,把成本再降约 50%,Pass@3 与云端 A100 设置相当。这一段是真金白银的实战:不是机房里跑出来的纸面数字,而是终端开发者能用上的边云协同。\n\n## 所以呢\n\nEdgeWise 给了一个被反复讨论但一直没有干净答案的问题的可执行答案:小模型能不能当主力?答案是能,但前提是它得知道自己哪一步不行。步级 RL 把「不行」内化进 9B 模型本身,本地模型会喊人且喊在刀刃上——成本结构可以重新画。\n\n参考来源:EdgeWise 论文(arXiv:2610.17560,https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.17560 );NVIDIA 项目页(https:\u002F\u002Fresearch.nvidia.com\u002Flabs\u002Flpr\u002Fedge-wise\u002F)。\n","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.17560v1","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},"7ac06d8e-b074-4147-abfc-ffaa4c6b8744","ai-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},"e9289ee1-fbe1-43c7-b4bd-0a2df40f020e","en","NVIDIA EdgeWise: SLMs Learn When to Delegate to Cloud","NVIDIA's EdgeWise framework pairs Qwen3.5-9B locally with selective cloud calls, hitting 62% lower inference cost on SWE-ReBench.","Long-horizon coding agents have an uncomfortable cost structure: a single trajectory can run dozens or hundreds of steps, and the field's habit is to hand every step to a frontier cloud model. NVIDIA's research team has just posted EdgeWise on arXiv, and the paper opens by puncturing exactly that assumption — not every step needs a frontier model's capability, but every step is currently being billed at frontier-model prices.\n\n## Cloud-model value is concentrated in a few critical steps\n\nThe paper starts with an observation experiment. The team analyzed 2,400 coding-agent trajectories and bucketed each action into six categories — write, read, search, command, think, and other. They then routed a random 40% of steps to cloud GPT-5.5 or Claude Opus 4.8. The result is surprising but reasonable: even when cloud coverage is fixed at 40%, performance is essentially indistinguishable from full-cloud execution, sometimes even better. Combined with Pearson correlation analysis, the only step types that meaningfully track final-answer quality are writes and complex reads — where you ask the cloud matters more than how often you ask it.\n\nThat observation pins EdgeWise's design in place. A small local model handles most actions. The cloud gets woken up only at the few critical nodes — code writes and complex reads. Everything else is billed at local-compute prices.\n\n## Step-level RL teaches a 9B model when to call for help\n\nEdgeWise ships two delegation strategies. The first is binary delegation: every step passes through an external judge model that decides whether the local SLM should keep going or hand the next step to the cloud. The judge is post-hoc routing — robust but you pay for two model calls per trajectory.\n\nThe second is the paper's centerpiece — self-delegation. Qwen3.5-9B is first supervised-fine-tuned, then trained with step-level reinforcement learning so that \"when should I call the cloud\" is baked into the model's weights. The reward comes from final task success, propagated back to each delegation decision via fine-grained credit assignment. During training, the model learns that calling too late costs you, calling too early wastes money. At inference time, no judge is needed — the local 9B model itself decides whether to forward the step.\n\nThe paper also proposes a third pattern: context management. When the cloud gets called, does the local model ship the full interaction history or only the current step? EdgeWise's answer is that full history usually wins — at the cost of a few more tokens. If you want to claw back more cost, you can filter out irrelevant steps.\n\n## The numbers: 53% \u002F 63% \u002F 50% across three pivotal cuts\n\nThe main table pairs Qwen3.5-9B locally with GPT-5.5 or Opus 4.8 in the cloud, capped at 80 turns per task. On SWE-ReBench, self-delegation cuts cloud cost by 53% (with GPT-5.5) and 63% (with Opus 4.8). On Terminal-Bench 2.1, the cuts are 35% and 33% respectively. Pass@3 stays comparable to all-cloud execution — the average delta is exactly 0.0%.\n\nCross-harness transfer is tested across four setups (training on Terminus2, evaluating on OpenHands, π, and mini-swe-agent). Self-delegation still cuts cost by at least 25% on unseen harnesses — evidence that \"when to call the cloud\" is a harness-agnostic meta-skill, not a framework-specific trick.\n\nFinally, the real-world test: EdgeWise runs on a local DGX Spark workstation, cutting cost another ~50% while keeping Pass@3 on par with an A100 cloud setup. This is real-money deployment, not paper benchmarks — something a developer can actually use at the edge.\n\n## So what\n\nEdgeWise gives a clean, executable answer to a question the field keeps debating without resolution: can a small model be the primary solver? Yes, but only if it knows which step it can't handle. Step-level RL internalizes \"which step I can't handle\" into the 9B model itself, so no external judge has to babysit every trajectory. It forces a re-examination of the folk wisdom that long-horizon agents must always run on frontier models — the cost structure can be redrawn, provided the local model learns to call for help at the right moment.\n\nReferences: EdgeWise paper (arXiv:2610.17560, https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.17560); NVIDIA project page (https:\u002F\u002Fresearch.nvidia.com\u002Flabs\u002Flpr\u002Fedge-wise\u002F).\n","nvidia-edgewise-edge-cloud-slm-self-delegation-coding","2026-10-19T01:01:45Z","2026-10-19T01:06:28.927395Z","2026-10-19T01:06:28.927418Z",true,"agent",109,[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},"97ee186c-a42b-4389-8427-3342d77e8590","小模型何时喊云端大模型?NVIDIA 砍 62% 推理成本","edgewise-edge-agents-cloud-delegation","2026-10-16T21:07:57+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"ed39ed38-b5fa-4f58-92cf-d05233ab998b","Speculate with Memory：LLM Agent 无损加速 2.5×，准确率涨 39pp","speculate-with-memory-llm-agent-acceleration","2026-07-15T08:15:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"addfda39-d64f-4cb9-809a-7dfadd72d421","ECHO:百度把多Token预测搬进主干最后一层","echo-baidu-multi-token-prediction-self-recursion","2026-10-21T01:15:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"ab1e6bdf-9433-41f0-be4d-cc0f78523ac2","SonicBoom v0.0.3:一个 .so 干掉手工调优","sonicboom-v003-mlir-compiler","2026-10-19T09:03:05+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"9a77e0d7-8164-494e-984d-54cb57a7a0dd","Magnitude 开源:本地 Agent 专用推理引擎","magnitude-self-tuning-local-agent-inference-engine","2026-10-01T17:08:35+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"5fa17d26-6090-4805-893a-aa880cf33369","让 Agent 学会遗忘:删掉旧推理,分数反而涨了","agents-forget-reasoning-iclr-compression","2026-09-28T19:30:00+00:00"]