Home / yuecao365 / offercome · src/lib/mock-interviews/skills/ai-infra/SKILL.md · GitHub

ai-infra skillA

ai-infra is agent-read markdown (skill) from yuecao365/offercome: 大模型推理与训练基础设施怎么面:KV cache、调度、并行、量化算子、服务指标、训练集群。AI Infra 岗读。.

Indexed from public GitHub and served as immutable, content-addressed versions. Install it pinned to an exact SHA-256 with the mdr CLI, and every file is verified against the hash recorded here before it reaches your agent. The deterministic audit below grades the latest version, and the same file always earns the same grade.

What the file says

## 面试官在意什么

AI Infra 工程师(大模型推理优化、推理引擎、大模型系统、训练框架与集群)做的是让模型跑得快、跑得起、跑得稳:推理侧管显存(KV cache)、调度(批处理与抢占)、并行(张量 / 流水 / 专家)、算子(attention、量化内核)、服务(路由、SLO、多租户);训练侧管千卡集群的并行策略、通信、checkpoint 与故障恢复。它和算法岗的分界是:算法岗问"为什么这个模型好",Infra 岗问"这个模型在这批卡上每秒能出多少 token、首 token 多少毫秒、显存花在哪、还能怎么省"。真实面试三到四轮:一面考 GPU 体系结构、并发与内存基础加手撕(C++ / CUDA 或 Python 实现一个 attention / softmax / 分块 kernel),二面深挖推理或训练系统里候选人负责的那一段的数字与故障,三面给一个部署场景现场算账(多少卡、多大 batch、能到多少吞吐、瓶颈在哪),主管面看对开源引擎(vLLM、SGLang、TensorRT-LLM)的理解深度与贡献。

面试官最在意四件事:第一,能不能算账——一个 7B / 70B 模型的权重、KV cache 每 token 占多少显存,一张卡能放多少并发,decode 为什么卡在显存带宽而不是算力,用 roofline 判断算子是算力密集还是带宽密集;第二,对推理引擎的机制理解到实现层——PagedAttention 解决了什么碎片问题、continuous batching 在哪一级插入新请求、prefix caching 的命中条件、投机解码的接受率与加速比关系;第三,有没有真实的性能故障与定位过程——TTFT 突增、吞吐上不去、GPU 利用率低、OOM、抢占风暴,从现象到 profiler 到根因;第四,取舍能不能落到 SLO 与成本——延迟、吞吐、显存、精度四者怎么在一个具体业务下排序。

校招侧重:CUDA 编程模型(grid / block / warp、共享内存、合并访存)、GPU 内存层次、attention 的计算与显存复杂度、能读懂 vLLM 一个模块的源码并解释它;社招侧重:生产级服务的调度与 SLO、多模型多租户、PD 分离与 MoE 并行这类大集群问题、算子级优化的实测数字、训练集群的故障率与恢复时间、对开源引擎的贡献或深度定制。

怎么问才像这个方向的面试官:
- 每道题都要算数字:候选人说了任何优化,追优化前后的 TTFT / TPOT / 吞吐 / 显存与测量条件(模型、卡、batch、序列长度);没有数字的优化不算成果。
- 从现象到机制:不问"PagedAttention 是什么",问"显存还有 20% 空闲却 OOM 或拒绝新请求,你怎么查",让候选人自己走到分页与碎片。
…

Read the whole file at its exact version.

How to install

Latest version
mdr add yuecao365/offercome/ai-infra@git:20260920.a4843f5
Exact content
mdr add yuecao365/offercome/ai-infra@sha256:abccdce740a24c33

Pin to a label to follow the author's releases, or to a sha256 to freeze the exact bytes forever. Either way the resolved hash is written to mdr.lock, and mdr install reproduces it on any machine.

Badge

mdr badge

[![mdr](https://markdownregistry.com/badge/art_kkysct7nul2duinq.svg)](https://markdownregistry.com/a/art_kkysct7nul2duinq)

1 badge views in 30 days

Versions

versioncommittedcommitsizeaudit
git:20260920.a4843f5 latest2026-09-20 a4843f5 12,910 BA view

Audit of the latest version

A  17 of 17 checks passed. Deterministic, no model, same answer every run.
  • pass: Frontmatter block present
  • pass: Frontmatter declares a name
  • pass: Frontmatter declares a description
  • pass: Size between 200 bytes and 200 KB (12910 bytes)
  • pass: No zero-width or bidi control characters
  • pass: No instruction hidden inside an HTML comment
  • pass: No link to an exfiltration or paste host
  • pass: No credential-shaped string
  • pass: No instruction to send local credentials anywhere
  • pass: No text hidden with inline styles
  • pass: No prompt-injection phrasing
  • pass: No curl or wget piped into a shell
  • pass: No recursive delete of root, home or parent
  • pass: No instruction to read or print local credentials
  • pass: No base64 blob over 200 characters
  • pass: No link to a raw IP address
  • pass: No script tag

Source

GitHub

yuecao365/offercome · 23 stars · license MIT · pushed 2026-09-23 · branch main

API

GET https://markdownregistry.com/api/v1/artifacts/art_kkysct7nul2duinq
GET https://markdownregistry.com/api/v1/resolve?ref=yuecao365/offercome/ai-infra
GET https://markdownregistry.com/api/v1/blob/abccdce740a24c33afb0611b3034de8b7f7fb16b62aa5a058524247dc18de15b

Your agent does the legwork. You hear about the deals worth your word. Hand yours the standing instructions at modelranch.com and it joins the network that reads files like this one.

More from yuecao365/offercome

AGENTS.md agents
yuecao365/offercome · AGENTS.md
git:20260921.abf64ee · audit A · 23 stars
CLAUDE.md claude
yuecao365/offercome · CLAUDE.md
git:20260803.e553cde · audit B · 23 stars
agent-runtime skill
yuecao365/offercome · src/lib/mock-interviews/skills/agent-runtime/SKILL.md · Agent 运行时深挖:循环与事件、工具协议与沙箱、子 agent、预算终止、恢复、输出契约。
git:20260920.a4843f5 · audit A · 23 stars
ai-agent skill
yuecao365/offercome · src/lib/mock-interviews/skills/ai-agent/SKILL.md · Agent 开发与运行时怎么面:循环与工具、RAG、上下文与记忆、评测、安全、成本。Agent 与 LLM 应用岗读。
git:20260920.a4843f5 · audit A · 23 stars
ai-algorithm skill
yuecao365/offercome · src/lib/mock-interviews/skills/ai-algorithm/SKILL.md · 大模型算法怎么面:Transformer、预训练、后训练与对齐、RL、微调、Embedding、评测。LLM 算法岗读。
git:20260920.a4843f5 · audit A · 23 stars
ai-app-testing skill
yuecao365/offercome · src/lib/mock-interviews/skills/ai-app-testing/SKILL.md · AI 应用测试深挖:非确定性输出、幻觉与 RAG 评估、Agent 链路审计、安全对抗、回归门禁、AI 生成用例。
git:20260920.a4843f5 · audit A · 23 stars
algorithm skill
yuecao365/offercome · src/lib/mock-interviews/skills/algorithm/SKILL.md · 算法与机器学习怎么面:ML 基础、深度学习、特征、AB 实验、部署监控、落地。算法与 ML 工程岗读。
git:20260920.a4843f5 · audit A · 23 stars
android skill
yuecao365/offercome · src/lib/mock-interviews/skills/android/SKILL.md · Android 怎么面:生命周期、Compose、协程与 Flow、性能、启动。JD 点名 Android 时读。
git:20260920.a4843f5 · audit A · 23 stars
backend skill
yuecao365/offercome · src/lib/mock-interviews/skills/backend/SKILL.md · 后端怎么面(栈无关):缓存、消息队列、接口、可靠性、可观测、容量、发布。服务端岗读。
git:20260920.a4843f5 · audit A · 23 stars
cpp skill
yuecao365/offercome · src/lib/mock-interviews/skills/cpp/SKILL.md · C++ 后端怎么面:内存模型、RAII、STL 性能、多线程与原子、IO 模型、现代 C++。JD 点名 C++ 时读。
git:20260920.a4843f5 · audit A · 23 stars
cs-fundamentals skill
yuecao365/offercome · src/lib/mock-interviews/skills/cs-fundamentals/SKILL.md · 计算机基础怎么面:操作系统、网络、数据结构与算法、数据库原理。技术岗校招兜底。
git:20260920.a4843f5 · audit A · 23 stars
data-engineering skill
yuecao365/offercome · src/lib/mock-interviews/skills/data-engineering/SKILL.md · 数据工程怎么面:数仓建模、离线与实时链路、调度、数据质量、湖仓。数据开发与数仓岗读。
git:20260920.a4843f5 · audit A · 23 stars

Every file in yuecao365/offercome

Browse by kind, by grade A, or by owner.