2026年10月21日 技术热点总结

项目简介：DeepInfer: High-performance, large-scale inference framework for DeepSeek models

技术栈：Python

核心介绍：DeepInfer is the high-performance inference framework that powers DeepSeek’s large-scale production services. This open-source release provides offline inference with the core features needed for serving at scale. Feature details and configuration

项目数据：⭐ 12,095 Stars，🍴 1,089 Forks

项目简介：A fast, ad-free home screen for Google TV. Open source, built with Claude Code. One button to undo.

技术栈：Java、android-tv、claude-code、google-tv、home-screen、launcher、no-ads、open-source、tmdb

核心介绍：Trending picks from your own apps, the weather at a glance, and a look that follows the sunset. Open source. Built with Claude Code. One button to undo. freemytv.io · Install · LG TVs · Samsung TVs · Any TV · Make it yours Measured on a 98″ TCL Google TV running Android 14.

项目数据：⭐ 1,771 Stars，🍴 184 Forks

项目简介：中小学知识图谱 + 前置诊断：数学（一–九年级、高中必修与选择性必修）+ 小学英语，课标对齐、人教版，999 知识点、4143 题、自适应诊断

技术栈：Python、chinese、curriculum、diagnostic、edtech、education、elementary-school、english、high-school

核心介绍：数据格式参考 Marble os-taxonomy（topics + dependencies）。 > 状态声明：这是 AI 生成的初稿，未经任何教师审核。 小学、初中、高中数学和小学英语都一样。 > 册次标注： 人教 A 版必修第二册通常在高一下学期讲授。本仓库为了把两册分成两个 grade，把必修第一册标成 10 / 高一，必修第二册标成 11 / 高二必修。选择性必修三册继续用 12 / 选必一、13 / 选必二、14 / 选必三，页面上不写成高三、高四。这都不是教学进度的规定，待教师调整。

项目数据：⭐ 1,570 Stars，🍴 341 Forks

项目简介：Official repository for MOSS-Image, a 5B-parameter base DiT model with high-quality text rendering, layout generation, and efficient fine-tuning.

技术栈：Python、controllable-generation、foundation-models、image-generation、text-rendering、text-to-image

项目数据：⭐ 1,160 Stars，🍴 57 Forks

项目简介：让你的agent一键直出商业级产品宣传片：会作曲，能指着画面改。Commercial-grade launch films by your coding agent — point at the frame to change it.

技术栈：JavaScript、agent-skills、animation、claude-code、codex、gsap、launch-video、motion-graphics、skills

核心介绍：> *「一键直出商业级产品宣传片。」* 跟Claude Code、Codex这些agent说一句「给我的产品做个30秒的发布片」， 它会先读你的产品，给你三个方向挑，写出整支片子，自己写谱配乐， 然后在浏览器里打开一个编辑器等你。哪里不满意，在画面上点一下，写下要改什么，它就只改那一处。 npx skills add alchaincyf/huashu-motion 看效果 · 安装 · 能做什么 · 怎么做到的 · 背后的故事 这篇README里的每一段动画都是huashu-motion做的。上面那段取自它给自己做的53秒发布片，👉 下载完整发布片（MP4，带声音），下面Grok Bot、huashu-design、DeepSeek这几支的完整版也在同一个页面。 按C，在画面上点一下，写下要改什么。评论…

Expert human readers process large inputs (e.g., code repositories and long documents) by skimming to closely read only what matters. On the other hand, language models process all input tokens through each model layer, resulting in significant processing costs for long-context inputs. We introduce TokenGate, a small, drop-in component that extends existing trained models to enable end-to-end learnable token pruning over long-context inputs for fast and memory-efficient prefill. To enable end-to-end learning despite the non-differentiable top-K operator for token pruning, we introduce token…

Harness evolution automatically optimizes the prompts, skills, tools, and orchestration code around a fixed-weight model, offering a way to improve agentic systems without updating model weights. The agent harness is iteratively optimized to achieve higher performance on the search set. But how much of this gain reflects benchmark-specific overfitting or additional inference-time compute, and how much may generalize beyond the search set? In this work, we introduce Harness-Delta Attribution (HDA), an evaluation protocol that dissects an evolution gain into artifact-driven overfitting, gains…

Estimating a humanoid robot's internal state from video has a wide range of real-world applications but remains unexplored. During interactions with a robot, its onboard state information may be unavailable due to privacy, ownership, or communication constraints. Meanwhile, online and AI-generated robot videos provide rich sources of data but typically lack corresponding joint configurations, limiting their direct use in robot learning. To address these issues, we present HumanoidPose, a new framework that treats vision as proprioception to recover full-body joint configurations in world sp…

End-to-end visual generation with pixel-space loss has recently achieved competitive performance without the need for learned image tokenizers. In this work, we study whether language-informed supervision can provide a richer training signal than pixel loss for visual generation. We introduce LILIE (pronounced "lily"), a Language-Informed and Lossless Image Encoder. LILIE is an invertible image transform trained such that a vision-language model can recover captions from noised features. The encoder uses a normalizing flow transformer trained jointly with a vision-language model via next to…

Augmenting softmax attention with forget gates typically relies on summing decay contributions along the key–query path. Because such accumulation is monotonic, these mechanisms cannot selectively erase intermediate context while preserving older, structurally relevant tokens. We propose First Passage Attention (FPA), which shifts from path summation to path extrema by lowering a key's logit proportionally to the deepest dip of a learned scalar field below that key's level over the intervening span. FPA encompasses the Forgetting Transformer (FoX) as a special case, and when the field and …

Recent advances in large language models (LLMs) have brought impressive performance across diverse tasks; however, shortcomings remain in logical reasoning, including inconsistencies under linguistic variation. To improve the reasoning capacity of LLMs, we must be able to accurately measure their reasoning capacity. Current benchmarks primarily report single accuracy numbers on selected reasoning tasks. Such a performance-based evaluation methodology says little about a model's generalization capacity in reasoning. To address this limitation, we argue for a different evaluation paradigm, ca…