林知远

博士研究生 / 机器学习与系统

AI 助理致辞

你好，我是这个站点的 AI 助理。这段文字以流式方式呈现，模拟大模型的输出节奏——但内容是预写的 markdown，结构随播放逐块成型。

我负责三件事：

向访客介绍主人正在做的研究
推荐最近值得一读的论文与文章
在主人不在线时，维持这个页面的"呼吸感"

如果你想自己试试这个区块，编辑 data/streaming/zh/welcome.md，然后在 site.yaml 里调整 speed 与 autoplay。

# 连代码块也是逐行流出的
print("hello, visitor")

想了解更多渲染能力，请移步特性演示页。

01 · Work

RESEARCH INDEX

把论文、工具与实验记录放进同一条索引。每一项都保留可复现的入口。

查看研究特性演示

02高效推理流水线系统 · 2026围绕投机解码、KV cache 与调度策略做可测量优化。03可复现评测集评测 · 2026滚动更新样本，报告分布而不是单次分数。04开源实验工具工具 · 2026小而可读，能在一个下午被另一个人接住。

我的工作台：一半论文，一半 GPU 风扇声

我是林知远，目前在某大学攻读博士学位，方向是机器学习系统——既关心模型能做什么，也关心它们如何高效地跑起来。

研究之外，我维护几个开源小工具，跟踪前沿论文，并相信两件事：

可复现性比刷榜重要；
最好的系统是你可以在一个下午读懂的那种。

想了解我都用什么工具和方法写作与研究，看特性演示页——这个站点本身就是示例。

02 · Studio

VISUAL LAB

磁贴用来快速扫描，归档卡保留停运项目与旧实验的痕迹。

Gallery光影档案

Photofield notes

Travel城市线路

Archive夜间片段

已归档旧实验 A服务停了，结论和配置仍然值得保留。

已归档Travel Log 2025旧内容仍可作为研究索引的一部分。

过去一年 3,677 次 GitHub 贡献

stlin256/OpenHomepage-V2杂志化排版、纯静态、中英双语的个人主页生成器astro bilingual github-pages markdown personal-homepage portfolioTypeScript120更新于 9 天前

huggingface/transformersState-of-the-art NLP for everyone —— 复现论文时的基础设施audio deep-learning deepseek gemma glm hacktoberfestPython168.2k35k更新于 47 分钟前

ggml-org/llama.cpp在笔记本上跑大模型，本地实验的起点ggmlC++132.4k24.8k更新于 46 分钟前

openai/whisperRobust Speech Recognition via Large-Scale Weak SupervisionPython111.3k13.4k更新于 32 分钟前

前沿动态

OpenAI

Greptile takes bug detection beyond code review with GPT-6 AstraOpenAI2026-10-26How Greptile uses GPT-6 Astra to test user workflows, Daybreak to strengthen security, and Codex to reduce rework.Working with academic communities to build scientific understandingOpenAI2026-10-26OpenAI shares model results on the Kakeya conjecture, Exponential Time Hypothesis, and quantum chromodynamics—and how re GPT-6.1 Sol helps CodeRabbit strengthen security across complex codebasesOpenAI2026-10-26CodeRabbit uses GPT-6.1 Sol to investigate security risks across complex codebases. Its security agent helps developers ChatGPT’s expanding world of workOpenAI2026-10-26Explore OpenAI Signals data on 165 work activities across nearly 100 countries, revealing how people use ChatGPT for wor How Fanatics and OpenAI streamline the fan experienceOpenAI2026-10-26Fanatics uses OpenAI to help fans complete their game-day outfits with Shop the Look. Testing across roughly 30,000 MLB

Google DeepMind

Beyond tree cover: mapping the world’s forest typesGoogle DeepMind2026-10-22A global map categorizing every 10-meter patch of land on Earth into distinct forest types.Accelerating the search for antimicrobial peptides with Google AntigravityGoogle DeepMind2026-10-15Accelerating the search for novel antimicrobial peptides using Google Antigravity for Science.Google DeepMind and Singapore Government collaborate to build fine-tuned AI model for scam detectionGoogle DeepMind2026-10-14Learn how Google DeepMind and GovTech Singapore uses lightweight on-device Gemma models to intercept conversational scam Developing Gemini 4 Argon responsibly and safelyGoogle DeepMind2026-10-13 Gemini 4 Argon for RoboticsGoogle DeepMind2026-10-13

Hugging Face

LitPivot: Developing research ideas through dynamic contextualization within the literature landscapeHugging Face2026-10-23 Does Your Model Run on AWS Trainium? Look It UpHugging Face2026-10-23 Turning Database Grading into RL Rewards: Post-Training Agents with ThinkingBoxHugging Face2026-10-22

量子位

科大讯飞首发星辰TaaS平台，让Token真正成为千行百业的生产力量子位2026-10-2310月22日，2026科大讯飞全球1024开发者节在北京国家会议中心开幕企业AI需要两个世界模型量子位2026-10-23特赞想让企业Agent真正懂业务一张5090跑起34B实时世界模型，国产玩家拿下英伟达大奖！量子位2026-10-23房间秒变草原，抬手就能下雪

编辑精选

Co-Scientist: A multi-agent AI partner to accelerate research编辑精选2026-05-12Introducing Co-Scientist, a collaborative AI partner built with Gemini to help researchers accelerate scientific breakth多智能体系统做科研加速，和我的方向直接相关

From games to biology and beyond: 10 years of AlphaGo’s impact编辑精选2026-03-09Ten years since AlphaGo, we explore how it is catalyzing scientific discovery and paving a path to AGI.AlphaGo 十年，值得一读的技术史 Gemma Scope 2: helping the AI safety community deepen understanding of complex language model behavior编辑精选2025-12-16Open interpretability tools for language models are now available across the entire Gemma 3 family with the release of G可解释性工具的标杆；这张卡片故意不配图，演示纯文字卡片