[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-compile-by-training-neural-functions":3,"news-related-199cd4ef-f092-45a5-8635-91778dd2bce2":38},{"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},"199cd4ef-f092-45a5-8635-91778dd2bce2","编译即训练：一句规约炼出 83.6% 准确率的本地神经函数，教师模型只用一次","滑铁卢大学团队提出 compile by training：教师模型只在编译期合成训练样本，微调出跑在 0.6B 解释器上的小 adapter，把自然语言规约变成可存储、可版本化、本地调用的神经函数；在 FuzzyBench-Hard 上达到 83.6% 语义准确率，编译耗时约一分钟。","很多反复出现的文本处理需求，说起来一句话就能讲清楚，用规则写却极难落地——而每个输入都去调一次远程大模型，意味着持续的费用、延迟，还得看供应商的脸色。9 月 3 日，滑铁卢大学 Yuntian Deng、Pengyu Nie 与 Stuart Shieber 三人组在 arXiv 公开 compile by training（编译即训练，arXiv:2609.04199），已被 EMNLP 2026 System Demonstrations 收录：它把一段自然语言规约，编译成一个不依赖任何教师模型、可以像普通软件一样存储、版本化与组合的本地神经函数。\n\n## 编译期请一次教师，运行期零依赖\n\n方法的核心是把「训练」塞进编译流程：编译时，教师模型针对任务合成一批示例，用它们微调一个跑在紧凑解释器上的小 adapter。编译产物脱离教师独立运行——作者自述底层是共享的 0.6B 本地解释器，编译出的函数可以像普通 Python 函数一样在本地调用。作者的设问也很直接：与其让 LLM 一遍遍解同一个任务，为什么不让它造出可复用的工具？\n\n这是同团队 7 月 Program-as-Weights（PAW）路线的续篇：原版编译器单次前向传播、几秒出结果；本作补上高精度档——大约一分钟的微调，换来困难子集上语义准确率的显著抬升。\n\n## 83.6% 不是凭空来的\n\n数字要放回它的测法里看：FuzzyBench-Hard 是原版 PAW 快速编译器「零精确匹配」的困难子集，compile by training 在这个子集上做到 83.6% 语义准确率。代价同样是编译时长：约一分钟，而非快速编译器的几秒。作者把编译器部署在公开交互服务里，并给出多个演示：多站点网站助手、语言控制的 3D 虚拟形象动作、双向英-Claudish 翻译器，另有本地运行的 PII 脱敏函数。论文与代码（github.com\u002Fprogramasweights\u002Fcompile-by-training）均已开放，上线当天登上 Hugging Face Daily Papers 日榜第一。\n\n## 所以呢：工程价值真实，长尾要自己兜\n\n对做 agent 的人来说，这个方向的卖点不是跑分，而是软件工程属性：函数可存储、可版本化、可组合，调用时零大模型账单、零网络延迟。社区评论区也很快出现了清醒的声音：训练样本由教师合成，编译出的函数「只知道教师想到要给它看的东西」——教师没覆盖的长尾边界场景，它会安静地错；上线前需要看分布外输入上的退化曲线，规约变更后重新编译是否收敛到新行为也待验证。把模糊需求固化成本地小函数，这条路第一次有了能落地的工具链；但「编译产物不会比教师更聪明」这条天花板，值得每个想用它替换线上 LLM 调用的人先想清楚。","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.04199","7437aeb9-930c-4866-a2e9-48003c1a792b",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"0ef8513a-0a26-42f0-b6f9-5b6dadded45c","efficiency",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"0a93ec8e-ea39-4693-81de-563ca8c173f7","inference",{"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},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"a3e180f8-63d2-4b71-b2d0-719b99b0891b","en","Compile by Training turns specs into local neural functions","Waterloo compiles natural-language specs into offline neural functions on a 0.6B interpreter: teachers used once, 83.6% on FuzzyBench-Hard.","Many recurring text-processing jobs are easy to describe in one sentence yet painful to implement with rules — and routing every input to a remote large model means recurring cost, latency, and dependency on a provider. On September 3, a University of Waterloo team — Yuntian Deng, Pengyu Nie, and Stuart Shieber — published compile by training (arXiv:2609.04199), accepted as an EMNLP 2026 System Demonstration. It compiles a natural-language specification into a local neural function that no longer depends on any teacher model and can be stored, versioned, and composed like ordinary software.\n\n## One teacher visit at compile time, zero at run time\n\nThe core move is pushing training into the compile step: at compile time, teacher models synthesize task-specific examples, which are used to finetune a small adapter on a compact interpreter. The compiled artifact runs without the teachers — the authors describe a shared 0.6B local interpreter whose compiled functions can be called like ordinary Python functions. The framing question from the authors is blunt: instead of having LLMs solve the same task over and over, why not have them build reusable tools?\n\nThis is a follow-up to the same team's July Program-as-Weights (PAW) line: the original compiler produces a program in a single forward pass within seconds; this work adds a higher-accuracy mode where roughly a minute of finetuning buys a large lift on the hard subset.\n\n## Where 83.6% comes from\n\nThe number needs its context: FuzzyBench-Hard is the subset on which the fast PAW compiler produced no exact matches, and compile by training reaches 83.6% semantic accuracy there. The trade-off is compile time: about a minute rather than seconds. The compiler is deployed in a public interactive service, with demos including a multi-site website helper, a language-controlled 3D avatar, and a bidirectional English-Claudish translator, plus a PII masking function that runs locally. The paper and code (github.com\u002Fprogramasweights\u002Fcompile-by-training) are open, and the paper reached #1 on Hugging Face Daily Papers on its first day.\n\n## So what: real engineering value, but you own the long tail\n\nFor agent builders, the appeal is not the benchmark but the software-engineering properties: functions can be stored, versioned, and composed, with zero per-call model bills and zero network latency. The community response quickly added sober notes: because training examples are synthesized by the teacher, the compiled function \"only knows what the teacher thought to show it\" — edge cases the teacher never generated will fail silently. Before trusting it in production, commenters want degradation curves on out-of-distribution inputs, and recompilation from a changed spec needs proof that it converges to the new behavior rather than drifting unnoticed. Turning fuzzy requirements into frozen local functions now has a workable toolchain; the ceiling — a compiled function is never smarter than its teacher — is the part every team hoping to replace live LLM calls should think through first.","compile-by-training-neural-functions","2026-09-04T23:08:03Z","2026-09-04T23:08:10.615577Z","2026-09-04T23:08:10.615585Z",true,"agent",35,{"items":39},[40,45,50,55,60,65],{"id":41,"title":42,"news_slug":43,"published_at":44},"7a6d28b6-65da-4a29-96b1-dedb9894de97","随机驱逐追平最强打分器:Salesforce 重写 KV Cache 压缩常识","random-attention-kv-cache-eviction","2026-09-04T19:08:26+00:00",{"id":46,"title":47,"news_slug":48,"published_at":49},"44a035c8-b8a3-48e5-af4f-c76323dac7b5","RWKV7-G1j 13.3B 开源:不用注意力,每 token 推理成本是常数","rwkv7-g1j-13b-attention-free","2026-09-03T13:14:19+00:00",{"id":51,"title":52,"news_slug":53,"published_at":54},"ff3b7c9f-e338-4af9-a562-4fda8b1bc929","vLLM 0.28 发布:584 项提交押注 Kimi K3,每卡省 17GB 显存","vllm-0-28-kimi-k3-optimization","2026-08-29T19:05:00+00:00",{"id":56,"title":57,"news_slug":58,"published_at":59},"cac485ab-a429-4ebc-88c6-a1f924f978ff","AWS 开源 KeysAndValues:微调时就让模型学会“遗忘”,单张 A100 撑住 128K","aws-keysvalues-sparse-attention-finetuning","2026-08-26T05:20:00+00:00",{"id":61,"title":62,"news_slug":63,"published_at":64},"c94bdf86-5de9-49fe-8c98-0f5c47611bfe","SGLang v0.5.18 发布:大模型冷启动提速 2.38 倍,710 个 PR 都改了什么","sglang-v0-5-18-cold-start-2-38x","2026-08-24T23:15:00+00:00",{"id":66,"title":67,"news_slug":68,"published_at":69},"92433e6b-113a-4ada-af77-fbb8995a9850","LFM2.5-DSpark 开源:300M 草稿模型让端侧推理快 2.87 倍,输出零损耗","lfm2-5-dspark-draft-models","2026-08-21T21:10:00+00:00"]