Skip to content
View LancyZhao's full-sized avatar
🎯
Focusing
🎯
Focusing

Block or report LancyZhao

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. 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
LancyZhao/README.md

赵兰熙 · Lancy Zhao

AI Infra 开发  ·  LLM 推理引擎与性能优化  ·  2027 届应届

Email GitHub University


专注 LLM 推理引擎与 GPU/CPU 异构性能优化——既能把 MoE 热点以 task 粒度切到 CPU 侧算,也能把异构训练节点的进程级瓶颈收成可复现的自动化诊断报告。

🛠 技术栈

Python C++ Go Java TypeScript Linux CUDA Git

方向 关键词
LLM 推理 / AI Infra SGLang · MoE CPU Offload · 快/慢 EPLB · CPU 虚拟专家 · CUDA / NV UVA · LLM 路由与成本策略
GPU / 系统性能 GPU 进程级采集 · NUMA / PCIe / GPU 拓扑 · TMA 微架构分析 · perf · eBPF · 性能剖析器开发
Linux 系统软件 系统编程 · 非侵入式监控 · 多级自举与离线部署 · AST 白名单求值
算法 / 信号处理 U-Net · 无监督学习 · 医学影像时序建模 · 信号采集与特征分析 · 故障诊断

🌐 开源贡献

在头部 LLM 开源项目中做可验证的协作,TypeScript / Python 双栈:

项目 贡献 语言 状态
run-llama/llama_index #22818 框架指南文档修复 Python ✅ 已合入主分支
BerriAI/litellm #38071 修复成本路由忽略 cache-read 单价 Python 🔵 评审中 · CI 76 项通过
vercel/ai #19247 · #19326 文档失效链接修复 TypeScript 🔵 评审中

代表作 — LiteLLM 成本路由修复: cost-based-routing 只按 input+output 单价排序、忽略 cache-read 单价。当两个 provider 输入输出价相同但 cache-read 价差 10 倍时,路由退化为"配置里谁靠前谁被选中",缓存密集场景成本最高可差约 10 倍。修复引入 cache-read 作次级排序键,并补参数化回归测试(修复前失败、修复后通过),通过类型纪律 / ruff-strict / basedpyright 三重预算门禁。

📮 联系

Popular repositories Loading

  1. ai ai Public

    Forked from vercel/ai

    Fork of vercel/ai · 用于贡献 PR #19247 与 #19326:贡献指南与架构文档失效链接修复(签名提交)

    TypeScript

  2. llama_index llama_index Public

    Forked from run-llama/llama_index

    Fork of run-llama/llama_index · 用于贡献 PR #22818:框架指南文档修复(已合入主分支)

    Python

  3. litellm litellm Public

    Forked from BerriAI/litellm

    Fork of BerriAI/litellm · 用于贡献 PR #38071:修复 cost-based-routing 忽略 cache-read 单价导致的成本路由退化(含参数化回归测试,CI 76 项通过)

    Python

  4. LancyZhao LancyZhao Public

    赵兰熙 · AI Infra / LLM 推理性能优化

  5. cardiac-mri-cs-recon cardiac-mri-cs-recon Public

    心脏 MRI 压缩感知欠采样重建与左心房分割复现实验:CS 10/20/30% 下零填充 / U-Net / 数据一致性三档对比,含真实 PSNR/SSIM/Dice 指标与方法学边界说明

    Python

  6. ai-infra-experiments ai-infra-experiments Public

    Isolated AI Infra experiments: synchronized CPU/MPS benchmarking, FFT correctness gates, batching, profiling and checkpoint fault-injection. Synthetic inputs only.

    Python