Skip to content
View lska367's full-sized avatar

Highlights

  • Pro

Block or report lska367

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
lska367/README.md

banner

Hi there 👋 我是杨锴(lska367)

Typing SVG

✨ Make AI usable everywhere 🌍

Email GitHub Blog Followers Profile Views


🎯 华东师范大学 · 2026 级硕士(保研)在读,研究方向:KV Cache / LLM 推理优化。

🎓 学习经历

阶段 学校 专业 时间
硕士 华东师范大学 · 计算机科学与技术学院 人工智能(保研) 2026.09 –
本科 东北林业大学 · 计算机与人工智能学院 数据科学与大数据技术 2022 – 2026

🚀 项目

vLLM Python Status

非侵入式 vLLM 运行时 tracer

观测调度 / 前向 / 采样 / KV cache 压力
定位 temp=0 输出不一致 的浮点根因
含分析工具(parquet/TTFT/flip 归因)与可视化前端

Stars

⭐ 零行为侵入 · 当前主力

PyTorch Flower Status

联邦标签噪声学习实验平台

噪声标签处理与算法对比框架
FedAvg / FedPAC / FedCorr 等

🌀 研究实验与基准

📓 kvcache_research

KV Cache Status

KV Cache 研究笔记

测量、复现与实验记录
coming soon

📝 持续沉淀

🔭 目前在做

  • KV Cache 量化 + 分层调度策略 — 做可复现的系统设计
  • 分布式 LLM 推理与训练 — Ray(Core / Train / Data)、vLLM 源码级调优
  • Kubernetes 分布式推理部署 — vLLM + Mooncake(KV Cache 跨节点传输)容器化实践

💼 工作经历

  • 益普索(中国)咨询有限公司 · AI Lab — AI 算法实习生 | 2025.12 – 2026.02

🌱 技术栈

方向 技术
Deep Learning PyTorch HF Transformers PEFT
LLM Inference vLLM KV Cache Quantization / Scheduling
Distributed Ray Training & Inference
LLM Apps LangChain / LangGraph Multi-Agent · RAG Context Engineering · LLM-as-Judge
Federated Flower (flwr) Label-Noise Learning
Systems Linux CUDA Python Git Kubernetes Docker

🧰 Toolbox


LLM 推理优化 · KV Cache · 分布式系统 · 性能分析 · 可复现设计 · ship → measure → iterate

📊 Activity Signal

贡献信号:近 52 周贡献折线 · GitHub 统计 · 常用语言

🏆 荣誉

  • 东北林业大学 2025 年度 国家励志奖学金

📫 联系

📧 yangkamboy@outlook.com | 📝 个人博客 | 欢迎交流 LLM 推理优化 / 联邦学习相关想法


Hi there 👋 I'm Kai Yang (lska367)

Typing SVG

✨ Make AI usable everywhere 🌍

Email GitHub Blog Followers Profile Views


🎯 M.S. student at ECNU (class of 2026, recommended admission). Research focus: KV Cache / LLM inference optimization.

🎓 Education

Degree School Major Period
M.S. East China Normal University · School of Computer Science and Technology Artificial Intelligence (recommended admission) 2026.09 –
B.S. Northeast Forestry University · School of Computer and Artificial Intelligence Data Science and Big Data Technology 2022 – 2026

🚀 Projects

vLLM Python Status

Non-intrusive vLLM runtime tracer

Observes scheduling / forward / sampling / KV pressure
Pinpoints temp=0 nondeterminism root causes
Analysis tools (parquet / TTFT / flip attribution) + offline webapp

Stars

⭐ Zero behavior change · main project

PyTorch Flower Status

Federated label-noise learning platform

Noise-label handling & algorithm benchmarks
FedAvg / FedPAC / FedCorr, etc.

🌀 Research experiments & benchmarks

📓 kvcache_research

KV Cache Status

KV Cache research notes

Measurements, reproductions, experiments
coming soon

📝 Ongoing notes

🔭 Currently

  • KV cache quantization + layer-wise scheduling — reproducible system design
  • Distributed LLM inference & training — Ray (Core / Train / Data), vLLM source-level tuning
  • Kubernetes for distributed inference — containerized vLLM + Mooncake (cross-node KV cache transfer)

💼 Experience

  • Ipsos (China) · AI Lab — AI Algorithm Intern | 2025.12 – 2026.02

🌱 Tech Stack

Area Stack
Deep Learning PyTorch HF Transformers PEFT
LLM Inference vLLM KV Cache Quantization / Scheduling
Distributed Ray Training & Inference
LLM Apps LangChain / LangGraph Multi-Agent · RAG Context Engineering · LLM-as-Judge
Federated Flower (flwr) Label-Noise Learning
Systems Linux CUDA Python Git Kubernetes Docker

🧰 Toolbox


LLM inference · KV cache · distributed systems · performance profiling · reproducible design · ship → measure → iterate

📊 Activity Signal

Contribution signal: 52-week line chart · GitHub stats · most used languages

🏆 Honors

  • Northeast Forestry University National Inspirational Scholarship (2025)

📫 Contact

📧 yangkamboy@outlook.com | 📝 Blog | Open to discuss LLM inference optimization / federated learning


Pinned Loading

  1. vllm-sniffer vllm-sniffer Public

    Non-invasive runtime tracer for vLLM inference: hook-free observation of scheduling/forward/sampling hot paths, per-request TTFT/TPOT timelines, and float nondeterminism (temp=0) root-cause localiz…

    Python 1