SparseMind

Two weeks. That's how long it took to go from GLM-5.3-Flash's first run on domestic accelerators to serving all of its production traffic, with 3.2× end-to-end throughput along the way. What I keep thinking about is who did much of the work: an Infra Agent powered by GLM-5.3. A model helping optimize the system that serves it. The conditions were hard. Limited memory and interconnect bandwidth. 1M-token context. Multimodal requests. An immature software stack where kernels were missing and documentation was often guesswork. Every optimization was a trade: compute for memory (ReplaySSM), communication for memory (intra-node tensor parallelism), precision for capacity (mixed INT8/FP8/BF16 caching), and disaggregation for scheduling freedom (Encode–Prefill–Decode). But the most important lesson wasn't about any single optimization. When the agent got stuck, it was rarely because it couldn't write the code. It was because it didn't know *why* things got worse. "Throughput down 20%" tells you something broke. It doesn't tell you which layer, which hypothesis, or what to test next. In RL terms, it's a sparse reward with a credit assignment problem. And an end-to-end benchmark that takes hours makes exploration painfully slow. Senior engineers solve this with an implicit process reward in their heads. They know when to check the timeline, when to run a microbenchmark, and which layer's output to compare. So we made that explicit. We call it dense feedback: layered verification interfaces the agent can call directly. Correctness feedback: did it compute right? System behavior feedback: where did the time go? Performance feedback: which option wins, under which conditions? Each signal has to be local, cheap, and objectively verifiable. Three things the agent found: First, precision drift in KDA's context-parallel path that grew with sequence length. The cause was TF32 rounding error compounding through chained state-matrix merges. The fix is now merged upstream in Flash Linear Attention (PR #1180). Second, KV transfer never overlapped with DeepEP dispatch. The agent followed the call chain across the Python/C++ boundary and found that the intranode path never released the GIL. After the fix, transfer overhead fell from over 30% to under 1%. Third, a decode kernel recomputing the same normalization four times because of how it was chunked. The agent restructured it and got a 1.71× speedup. The idea came from "optimization skeletons" it had distilled by reading existing kernels across SGLang, FLA, and DeepGEMM. To be clear about the boundaries: humans still defined the goals, built the feedback environment, and reviewed every high-risk change. But the engineer's role is changing, from the person who solves the problem to the person who designs the feedback. There's a deeper implication too. A layered, verifiable feedback environment built on real infrastructure tasks is exactly what training the next generation of models needs most. Every task the agent completes can become training ground for its successor. We are still far from recursive self-improvement. But the smallest loop now exists. The model optimizes the system. The system serves the model.

175

238

976

552K

SparseMind @muzhi1991study

26 days ago

@QuantumTransf 现在各家的 max 可也理解成 rl 过拟合的版本，跑分打榜专用的的，没有不 overthking 的，很多真实场景下不太多未必就是好的，类比考试的��候你得多想想总是好的

SparseMind @muzhi1991study

about 2 months ago

@sbansar @lidangzzz codex claude 的壁垒可能还没宇树高呢，一堆模型在屁股后面，也就几个月的领先吧，在 coding 领域可能几个月也没有了。宇树在运控硬件领先一个身位还有的。都是炒预期宇树也就他两估值的零头有啥可喷的。

152

muzhi1991study retweeted

鸭哥

@grapeot

about 2 months ago

DeepSeek 开源的第一个 agent harness DSH，论文里堆满了范畴论符号，很容易让人觉得又是研究员在真空里搞出来的理论产物。但在仔细读完源码并与 Codex 逐行对照后��我的工程判断很明确：对绝大多数日常写代码的开发者来说，它重得毫无必要；但对��索自进化 agent 的工程团队来说，它搭建了一套目前其他方案完全没有的底层骨架。架构上最本质的差距在于插件模型的选择。 Codex 代表了典型的声明式路线：插件只是磁盘上的文件夹，贡献的是 Markdown 写的 skill 文件、MCP server 配置或 shell 脚本。插件不进 harness 进程，也不在同一进程里跑代码，改完配置重启独立进程只需两三秒。对于搜索、加工具这类绝大多数日常需求，声明式模型简单可靠，门槛接近于零。 DSH 走的是命令式路线：插件带着自己的状态，直接跑在 harness 进程内部，互相注册与调用。一旦要在运行时替换插件，悬空引用清理、后台任务终止、依赖链协同以及崩溃回滚都会变成棘手难题。为此，DSH 引入了一套叫 Cordis 的重型运行时，单是管理 fiber 生命周期的核心模块就有 750 行代码。如果只是为了替换搜索服务或挂载常用工具，让开发者背上 Cordis 这套复杂度完全是过度设计。但 DeepSeek 包这盘饺子的真实意图，隐藏在控制流结构里。在 Codex 中，agent loop 的控制流被硬编码在 Rust 核心逻辑里，开发者只能在预设时刻挂载 hook，无法在运行时把单 agent 循环改成多 agent 协作循环。而在 DSH 里，agent loop 本身是放在 packages/core/agent-loop 中的一个普通 TypeScript 插件，向外提供 ctx.agentLoop 服务。只要实现相同的接口，运行中的控制流骨架随时可以整体卸载并替换。 Cordis 打造的那些副作用跟踪、依赖变动通知和事务性 HMR，本质上全是在支撑 agent loop 可替换这个核心目标。它保��了 agent 在运行期如果动态生成了新工具或新 loop，系统能在不中断进程的前提下平滑加载；一旦生成的代码出现错误，又能通过事务性回滚退回上一个稳定状态。 Codex 交付的是结构固定的成品家具，而 DSH 交付的是允许动态改造的生成内核。DSH 不会让你的日常 coding 变得更快，但它让 harness 本身具备了面向自进化进行物理扩展的可能。深度剖析与两套架构的完整对照： https://t.co/tqJua5Yi2y

563

488

90K