Tags

文献笔记 A View of Cloud ComputingReading Notes: A View of Cloud Computing

文献笔记 A Brief Introduction to Distributed SystemsReading Notes: A Brief Introduction to Distributed Systems

文献笔记 A Brief Introduction to Distributed SystemsReading Notes: A Brief Introduction to Distributed Systems

文献笔记 A View of Cloud ComputingReading Notes: A View of Cloud Computing

文献笔记 A Brief Introduction to Distributed SystemsReading Notes: A Brief Introduction to Distributed Systems

文献笔记 A Brief Introduction to Distributed SystemsReading Notes: A Brief Introduction to Distributed Systems

文献笔记 A View of Cloud ComputingReading Notes: A View of Cloud Computing

文献笔记 A View of Cloud ComputingReading Notes: A View of Cloud Computing

学习最优化课程中的一些疑惑Questions from an Optimization Course: Cartesian Products and Convex Sets

学习最优化课程中的一些疑惑Questions from an Optimization Course: Cartesian Products and Convex Sets

学习最优化课程中的一些疑惑Questions from an Optimization Course: Cartesian Products and Convex Sets

学习最优化课程中的一些疑惑Questions from an Optimization Course: Cartesian Products and Convex Sets

傻瓜操作：GraphRAG、Ollama 本地部署及踩坑记录A Noob-Friendly Guide: Deploying GraphRAG with Local Ollama, and Every Pitfall I Hit

傻瓜操作：GraphRAG、Ollama 本地部署及踩坑记录A Noob-Friendly Guide: Deploying GraphRAG with Local Ollama, and Every Pitfall I Hit

傻瓜操作：GraphRAG、Ollama 本地部署及踩坑记录A Noob-Friendly Guide: Deploying GraphRAG with Local Ollama, and Every Pitfall I Hit

傻瓜操作：GraphRAG、Ollama 本地部署及踩坑记录A Noob-Friendly Guide: Deploying GraphRAG with Local Ollama, and Every Pitfall I Hit

个人主页升级实战：Jekyll 增量改造的取舍与落地Upgrading My Personal Site: Jekyll Incremental Refactoring — Trade-offs and What Actually Shipped

从「要不要迁移框架」到「搜索、评论、SEO 全都要」的一次完整记录From 'should I migrate frameworks' to 'search, comments, SEO — I want it all'

个人主页升级实战：Jekyll 增量改造的取舍与落地Upgrading My Personal Site: Jekyll Incremental Refactoring — Trade-offs and What Actually Shipped

从「要不要迁移框架」到「搜索、评论、SEO 全都要」的一次完整记录

个人主页升级实战：Jekyll 增量改造的取舍与落地Upgrading My Personal Site: Jekyll Incremental Refactoring — Trade-offs and What Actually Shipped

从「要不要迁移框架」到「搜索、评论、SEO 全都要」的一次完整记录From 'should I migrate frameworks' to 'search, comments, SEO — I want it all'

个人主页升级实战：Jekyll 增量改造的取舍与落地Upgrading My Personal Site: Jekyll Incremental Refactoring — Trade-offs and What Actually Shipped

从「要不要迁移框架」到「搜索、评论、SEO 全都要」的一次完整记录

个人主页升级实战：Jekyll 增量改造的取舍与落地Upgrading My Personal Site: Jekyll Incremental Refactoring — Trade-offs and What Actually Shipped

从「要不要迁移框架」到「搜索、评论、SEO 全都要」的一次完整记录From 'should I migrate frameworks' to 'search, comments, SEO — I want it all'

个人主页升级实战：Jekyll 增量改造的取舍与落地Upgrading My Personal Site: Jekyll Incremental Refactoring — Trade-offs and What Actually Shipped

从「要不要迁移框架」到「搜索、评论、SEO 全都要」的一次完整记录

个人主页升级实战：Jekyll 增量改造的取舍与落地Upgrading My Personal Site: Jekyll Incremental Refactoring — Trade-offs and What Actually Shipped

从「要不要迁移框架」到「搜索、评论、SEO 全都要」的一次完整记录

个人主页升级实战：Jekyll 增量改造的取舍与落地Upgrading My Personal Site: Jekyll Incremental Refactoring — Trade-offs and What Actually Shipped

从「要不要迁移框架」到「搜索、评论、SEO 全都要」的一次完整记录From 'should I migrate frameworks' to 'search, comments, SEO — I want it all'

千卡训练的隐形账单：通信不是瓶颈The Hidden Bill of Thousand-GPU Training: Communication Is Not the Bottleneck

MLSys·im 学习笔记（五）——从单节点走到千卡集群：谁在偷走训练时间、Young-Daly 公式该往哪用、选址怎么在 61 倍里改写碳足迹MLSys·im Learning Notes (Part 5) — from one node to a thousand GPUs: who actually steals your training time, where the Young-Daly formula should point, and how site selection rewrites the carbon footprint by 61x

千卡训练的隐形账单：通信不是瓶颈The Hidden Bill of Thousand-GPU Training: Communication Is Not the Bottleneck

MLSys·im 学习笔记（五）——从单节点走到千卡集群：谁在偷走训练时间、Young-Daly 公式该往哪用、选址怎么在 61 倍里改写碳足迹MLSys·im Learning Notes (Part 5) — from one node to a thousand GPUs: who actually steals your training time, where the Young-Daly formula should point, and how site selection rewrites the carbon footprint by 61x

给开源教程修了个 demoFixing a Tutorial Demo: An Open-Source PR Retrospective

给 mlsysim 的 KV-Cache 教程重写了 §5 的 Paged Attention demo 并提了 PRI rewrote the §5 Paged Attention demo in mlsysim's KV-Cache tutorial and opened PR #2135; before merging, the maintainer revised it and shrank my numbers from 314 to 259. A retrospective of that experience.

GPU 为什么会挨饿：数据墙、投机解码与设计空间探索Why Does the GPU Starve: The Data Wall, Speculative Decoding, and Design-Space Exploration

MLSys·im 学习笔记（四）——从单点瓶颈走向全链路优化：CPU 预处理如何让 GPU 饿肚子，投机解码如何以小博大，DSE 如何在数千种配置中自动找最优解MLSys·im Learning Notes (Part 4) — from single-point analysis to whole-pipeline optimization: how CPU preprocessing starves the GPU, how speculative decoding bets small to win big, and how DSE searches thousands of configs automatically

GPU 为什么会挨饿：数据墙、投机解码与设计空间探索Why Does the GPU Starve: The Data Wall, Speculative Decoding, and Design-Space Exploration

MLSys·im 学习笔记（四）——从单点瓶颈走向全链路优化：CPU 预处理如何让 GPU 饿肚子，投机解码如何以小博大，DSE 如何在数千种配置中自动找最优解MLSys·im Learning Notes (Part 4) — from single-point analysis to whole-pipeline optimization: how CPU preprocessing starves the GPU, how speculative decoding bets small to win big, and how DSE searches thousands of configs automatically

给开源教程修了个 demoFixing a Tutorial Demo: An Open-Source PR Retrospective

给 mlsysim 的 KV-Cache 教程重写了 §5 的 Paged Attention demo 并提了 PR #2135;合并前维护者又改了一版,把数字从 314 改小到 259。复盘这次经历。I rewrote the §5 Paged Attention demo in mlsysim's KV-Cache tutorial and opened PR #2135; before merging, the maintainer revised it and shrank my numbers from 314 to 259. A retrospective of that experience.

KV-Cache 的显存账本：长上下文、并发与量化剪枝的取舍The KV-Cache Memory Ledger: Long Context, Concurrency, and the Quantization–Pruning Trade-off

MLSys·im 学习笔记（三）——用公式算清一个请求吃多少显存，序列长度如何决定并发上限，以及为什么"压缩"和"剪枝"在推理提速上完全是两回事MLSys·im Learning Notes (Part 3) — a formula for how much memory one request eats, why sequence length decides your concurrency ceiling, and why 'compression' and 'pruning' are two completely different stories for inference speedup

KV-Cache 的显存账本：长上下文、并发与量化剪枝的取舍The KV-Cache Memory Ledger: Long Context, Concurrency, and the Quantization–Pruning Trade-off

MLSys·im 学习笔记（三）——用公式算清一个请求吃多少显存，序列长度如何决定并发上限，以及为什么"压缩"和"剪枝"在推理提速上完全是两回事MLSys·im Learning Notes (Part 3) — a formula for how much memory one request eats, why sequence length decides your concurrency ceiling, and why 'compression' and 'pruning' are two completely different stories for inference speedup

内存墙与两阶段：为什么 3.2 倍算力只换来 1.6 倍加速The Memory Wall & Two Phases: Why 3.2x More FLOPS Buys Only 1.6x Speedup

MLSys·im 学习笔记（二）——从 Memory Wall 到 Prefill/Decode 双墙，一次 LLM 请求里的两套瓶颈MLSys·im Learning Notes (Part 2) — from the Memory Wall to the Prefill/Decode dual walls: two bottlenecks inside a single LLM request

内存墙与两阶段：为什么 3.2 倍算力只换来 1.6 倍加速The Memory Wall & Two Phases: Why 3.2x More FLOPS Buys Only 1.6x Speedup

MLSys·im 学习笔记（二）——从 Memory Wall 到 Prefill/Decode 双墙，一次 LLM 请求里的两套瓶颈MLSys·im Learning Notes (Part 2) — from the Memory Wall to the Prefill/Decode dual walls: two bottlenecks inside a single LLM request

用 5 行代码预测模型瓶颈：MLSys·im Roofline 教程学习笔记Predicting Model Bottlenecks in 5 Lines of Code: MLSys·im Roofline Tutorial Notes

从 Datawhale 打卡任务开始的性能分析之旅——ResNet-50、Llama-3 在 A100/H100 上的 roofline 实战与踩坑复盘MLSys·im Learning Notes (Part 1) — a performance-analysis journey from a Datawhale assignment: rooflining ResNet-50 and Llama-3 on A100/H100, with a post-mortem of every pitfall

用 5 行代码预测模型瓶颈：MLSys·im Roofline 教程学习笔记Predicting Model Bottlenecks in 5 Lines of Code: MLSys·im Roofline Tutorial Notes

从 Datawhale 打卡任务开始的性能分析之旅——ResNet-50、Llama-3 在 A100/H100 上的 roofline 实战与踩坑复盘A performance-analysis journey starting from a Datawhale assignment — rooflining ResNet-50 and Llama-3 on A100/H100, with a post-mortem of every pitfall

用 5 行代码预测模型瓶颈：MLSys·im Roofline 教程学习笔记Predicting Model Bottlenecks in 5 Lines of Code: MLSys·im Roofline Tutorial Notes

从 Datawhale 打卡任务开始的性能分析之旅——ResNet-50、Llama-3 在 A100/H100 上的 roofline 实战与踩坑复盘MLSys·im Learning Notes (Part 1) — a performance-analysis journey from a Datawhale assignment: rooflining ResNet-50 and Llama-3 on A100/H100, with a post-mortem of every pitfall

用 5 行代码预测模型瓶颈：MLSys·im Roofline 教程学习笔记Predicting Model Bottlenecks in 5 Lines of Code: MLSys·im Roofline Tutorial Notes

从 Datawhale 打卡任务开始的性能分析之旅——ResNet-50、Llama-3 在 A100/H100 上的 roofline 实战与踩坑复盘A performance-analysis journey starting from a Datawhale assignment — rooflining ResNet-50 and Llama-3 on A100/H100, with a post-mortem of every pitfall

GPU 为什么会挨饿：数据墙、投机解码与设计空间探索Why Does the GPU Starve: The Data Wall, Speculative Decoding, and Design-Space Exploration

MLSys·im 学习笔记（四）——从单点瓶颈走向全链路优化：CPU 预处理如何让 GPU 饿肚子，投机解码如何以小博大，DSE 如何在数千种配置中自动找最优解MLSys·im Learning Notes (Part 4) — from single-point analysis to whole-pipeline optimization: how CPU preprocessing starves the GPU, how speculative decoding bets small to win big, and how DSE searches thousands of configs automatically

内存墙与两阶段：为什么 3.2 倍算力只换来 1.6 倍加速The Memory Wall & Two Phases: Why 3.2x More FLOPS Buys Only 1.6x Speedup

MLSys·im 学习笔记（二）——从 Memory Wall 到 Prefill/Decode 双墙，一次 LLM 请求里的两套瓶颈MLSys·im Learning Notes (Part 2) — from the Memory Wall to the Prefill/Decode dual walls: two bottlenecks inside a single LLM request

用 5 行代码预测模型瓶颈：MLSys·im Roofline 教程学习笔记Predicting Model Bottlenecks in 5 Lines of Code: MLSys·im Roofline Tutorial Notes

从 Datawhale 打卡任务开始的性能分析之旅——ResNet-50、Llama-3 在 A100/H100 上的 roofline 实战与踩坑复盘A performance-analysis journey starting from a Datawhale assignment — rooflining ResNet-50 and Llama-3 on A100/H100, with a post-mortem of every pitfall

千卡训练的隐形账单：通信不是瓶颈The Hidden Bill of Thousand-GPU Training: Communication Is Not the Bottleneck

MLSys·im 学习笔记（五）——从单节点走到千卡集群：谁在偷走训练时间、Young-Daly 公式该往哪用、选址怎么在 61 倍里改写碳足迹MLSys·im Learning Notes (Part 5) — from one node to a thousand GPUs: who actually steals your training time, where the Young-Daly formula should point, and how site selection rewrites the carbon footprint by 61x

千卡训练的隐形账单：通信不是瓶颈The Hidden Bill of Thousand-GPU Training: Communication Is Not the Bottleneck

MLSys·im 学习笔记（五）——从单节点走到千卡集群：谁在偷走训练时间、Young-Daly 公式该往哪用、选址怎么在 61 倍里改写碳足迹MLSys·im Learning Notes (Part 5) — from one node to a thousand GPUs: who actually steals your training time, where the Young-Daly formula should point, and how site selection rewrites the carbon footprint by 61x

GPU 为什么会挨饿：数据墙、投机解码与设计空间探索Why Does the GPU Starve: The Data Wall, Speculative Decoding, and Design-Space Exploration

MLSys·im 学习笔记（四）——从单点瓶颈走向全链路优化：CPU 预处理如何让 GPU 饿肚子，投机解码如何以小博大，DSE 如何在数千种配置中自动找最优解MLSys·im Learning Notes (Part 4) — from single-point analysis to whole-pipeline optimization: how CPU preprocessing starves the GPU, how speculative decoding bets small to win big, and how DSE searches thousands of configs automatically

GPU 为什么会挨饿：数据墙、投机解码与设计空间探索Why Does the GPU Starve: The Data Wall, Speculative Decoding, and Design-Space Exploration

MLSys·im 学习笔记（四）——从单点瓶颈走向全链路优化：CPU 预处理如何让 GPU 饿肚子，投机解码如何以小博大，DSE 如何在数千种配置中自动找最优解MLSys·im Learning Notes (Part 4) — from single-point analysis to whole-pipeline optimization: how CPU preprocessing starves the GPU, how speculative decoding bets small to win big, and how DSE searches thousands of configs automatically

内存墙与两阶段：为什么 3.2 倍算力只换来 1.6 倍加速The Memory Wall & Two Phases: Why 3.2x More FLOPS Buys Only 1.6x Speedup

MLSys·im 学习笔记（二）——从 Memory Wall 到 Prefill/Decode 双墙，一次 LLM 请求里的两套瓶颈MLSys·im Learning Notes (Part 2) — from the Memory Wall to the Prefill/Decode dual walls: two bottlenecks inside a single LLM request

内存墙与两阶段：为什么 3.2 倍算力只换来 1.6 倍加速The Memory Wall & Two Phases: Why 3.2x More FLOPS Buys Only 1.6x Speedup

MLSys·im 学习笔记（二）——从 Memory Wall 到 Prefill/Decode 双墙，一次 LLM 请求里的两套瓶颈MLSys·im Learning Notes (Part 2) — from the Memory Wall to the Prefill/Decode dual walls: two bottlenecks inside a single LLM request

用 5 行代码预测模型瓶颈：MLSys·im Roofline 教程学习笔记Predicting Model Bottlenecks in 5 Lines of Code: MLSys·im Roofline Tutorial Notes

从 Datawhale 打卡任务开始的性能分析之旅——ResNet-50、Llama-3 在 A100/H100 上的 roofline 实战与踩坑复盘MLSys·im Learning Notes (Part 1) — a performance-analysis journey from a Datawhale assignment: rooflining ResNet-50 and Llama-3 on A100/H100, with a post-mortem of every pitfall

用 5 行代码预测模型瓶颈：MLSys·im Roofline 教程学习笔记Predicting Model Bottlenecks in 5 Lines of Code: MLSys·im Roofline Tutorial Notes

从 Datawhale 打卡任务开始的性能分析之旅——ResNet-50、Llama-3 在 A100/H100 上的 roofline 实战与踩坑复盘A performance-analysis journey starting from a Datawhale assignment — rooflining ResNet-50 and Llama-3 on A100/H100, with a post-mortem of every pitfall

千卡训练的隐形账单：通信不是瓶颈The Hidden Bill of Thousand-GPU Training: Communication Is Not the Bottleneck

MLSys·im 学习笔记（五）——从单节点走到千卡集群：谁在偷走训练时间、Young-Daly 公式该往哪用、选址怎么在 61 倍里改写碳足迹MLSys·im Learning Notes (Part 5) — from one node to a thousand GPUs: who actually steals your training time, where the Young-Daly formula should point, and how site selection rewrites the carbon footprint by 61x

千卡训练的隐形账单：通信不是瓶颈The Hidden Bill of Thousand-GPU Training: Communication Is Not the Bottleneck

MLSys·im 学习笔记（五）——从单节点走到千卡集群：谁在偷走训练时间、Young-Daly 公式该往哪用、选址怎么在 61 倍里改写碳足迹MLSys·im Learning Notes (Part 5) — from one node to a thousand GPUs: who actually steals your training time, where the Young-Daly formula should point, and how site selection rewrites the carbon footprint by 61x

给开源教程修了个 demoFixing a Tutorial Demo: An Open-Source PR Retrospective

给 mlsysim 的 KV-Cache 教程重写了 §5 的 Paged Attention demo 并提了 PRI rewrote the §5 Paged Attention demo in mlsysim's KV-Cache tutorial and opened PR #2135; before merging, the maintainer revised it and shrank my numbers from 314 to 259. A retrospective of that experience.

GPU 为什么会挨饿：数据墙、投机解码与设计空间探索Why Does the GPU Starve: The Data Wall, Speculative Decoding, and Design-Space Exploration

MLSys·im 学习笔记（四）——从单点瓶颈走向全链路优化：CPU 预处理如何让 GPU 饿肚子，投机解码如何以小博大，DSE 如何在数千种配置中自动找最优解MLSys·im Learning Notes (Part 4) — from single-point analysis to whole-pipeline optimization: how CPU preprocessing starves the GPU, how speculative decoding bets small to win big, and how DSE searches thousands of configs automatically

GPU 为什么会挨饿：数据墙、投机解码与设计空间探索Why Does the GPU Starve: The Data Wall, Speculative Decoding, and Design-Space Exploration

MLSys·im 学习笔记（四）——从单点瓶颈走向全链路优化：CPU 预处理如何让 GPU 饿肚子，投机解码如何以小博大，DSE 如何在数千种配置中自动找最优解MLSys·im Learning Notes (Part 4) — from single-point analysis to whole-pipeline optimization: how CPU preprocessing starves the GPU, how speculative decoding bets small to win big, and how DSE searches thousands of configs automatically

给开源教程修了个 demoFixing a Tutorial Demo: An Open-Source PR Retrospective

给 mlsysim 的 KV-Cache 教程重写了 §5 的 Paged Attention demo 并提了 PR #2135;合并前维护者又改了一版,把数字从 314 改小到 259。复盘这次经历。I rewrote the §5 Paged Attention demo in mlsysim's KV-Cache tutorial and opened PR #2135; before merging, the maintainer revised it and shrank my numbers from 314 to 259. A retrospective of that experience.

KV-Cache 的显存账本：长上下文、并发与量化剪枝的取舍The KV-Cache Memory Ledger: Long Context, Concurrency, and the Quantization–Pruning Trade-off

MLSys·im 学习笔记（三）——用公式算清一个请求吃多少显存，序列长度如何决定并发上限，以及为什么"压缩"和"剪枝"在推理提速上完全是两回事MLSys·im Learning Notes (Part 3) — a formula for how much memory one request eats, why sequence length decides your concurrency ceiling, and why 'compression' and 'pruning' are two completely different stories for inference speedup

KV-Cache 的显存账本：长上下文、并发与量化剪枝的取舍The KV-Cache Memory Ledger: Long Context, Concurrency, and the Quantization–Pruning Trade-off

MLSys·im 学习笔记（三）——用公式算清一个请求吃多少显存，序列长度如何决定并发上限，以及为什么"压缩"和"剪枝"在推理提速上完全是两回事MLSys·im Learning Notes (Part 3) — a formula for how much memory one request eats, why sequence length decides your concurrency ceiling, and why 'compression' and 'pruning' are two completely different stories for inference speedup

内存墙与两阶段：为什么 3.2 倍算力只换来 1.6 倍加速The Memory Wall & Two Phases: Why 3.2x More FLOPS Buys Only 1.6x Speedup

MLSys·im 学习笔记（二）——从 Memory Wall 到 Prefill/Decode 双墙，一次 LLM 请求里的两套瓶颈MLSys·im Learning Notes (Part 2) — from the Memory Wall to the Prefill/Decode dual walls: two bottlenecks inside a single LLM request

内存墙与两阶段：为什么 3.2 倍算力只换来 1.6 倍加速The Memory Wall & Two Phases: Why 3.2x More FLOPS Buys Only 1.6x Speedup

MLSys·im 学习笔记（二）——从 Memory Wall 到 Prefill/Decode 双墙，一次 LLM 请求里的两套瓶颈MLSys·im Learning Notes (Part 2) — from the Memory Wall to the Prefill/Decode dual walls: two bottlenecks inside a single LLM request

用 5 行代码预测模型瓶颈：MLSys·im Roofline 教程学习笔记Predicting Model Bottlenecks in 5 Lines of Code: MLSys·im Roofline Tutorial Notes

从 Datawhale 打卡任务开始的性能分析之旅——ResNet-50、Llama-3 在 A100/H100 上的 roofline 实战与踩坑复盘MLSys·im Learning Notes (Part 1) — a performance-analysis journey from a Datawhale assignment: rooflining ResNet-50 and Llama-3 on A100/H100, with a post-mortem of every pitfall

用 5 行代码预测模型瓶颈：MLSys·im Roofline 教程学习笔记Predicting Model Bottlenecks in 5 Lines of Code: MLSys·im Roofline Tutorial Notes

从 Datawhale 打卡任务开始的性能分析之旅——ResNet-50、Llama-3 在 A100/H100 上的 roofline 实战与踩坑复盘A performance-analysis journey starting from a Datawhale assignment — rooflining ResNet-50 and Llama-3 on A100/H100, with a post-mortem of every pitfall

GPU 为什么会挨饿：数据墙、投机解码与设计空间探索Why Does the GPU Starve: The Data Wall, Speculative Decoding, and Design-Space Exploration

MLSys·im 学习笔记（四）——从单点瓶颈走向全链路优化：CPU 预处理如何让 GPU 饿肚子，投机解码如何以小博大，DSE 如何在数千种配置中自动找最优解MLSys·im Learning Notes (Part 4) — from single-point analysis to whole-pipeline optimization: how CPU preprocessing starves the GPU, how speculative decoding bets small to win big, and how DSE searches thousands of configs automatically

内存墙与两阶段：为什么 3.2 倍算力只换来 1.6 倍加速The Memory Wall & Two Phases: Why 3.2x More FLOPS Buys Only 1.6x Speedup

MLSys·im 学习笔记（二）——从 Memory Wall 到 Prefill/Decode 双墙，一次 LLM 请求里的两套瓶颈MLSys·im Learning Notes (Part 2) — from the Memory Wall to the Prefill/Decode dual walls: two bottlenecks inside a single LLM request

用 5 行代码预测模型瓶颈：MLSys·im Roofline 教程学习笔记Predicting Model Bottlenecks in 5 Lines of Code: MLSys·im Roofline Tutorial Notes

从 Datawhale 打卡任务开始的性能分析之旅——ResNet-50、Llama-3 在 A100/H100 上的 roofline 实战与踩坑复盘MLSys·im Learning Notes (Part 1) — a performance-analysis journey from a Datawhale assignment: rooflining ResNet-50 and Llama-3 on A100/H100, with a post-mortem of every pitfall

内存墙与两阶段：为什么 3.2 倍算力只换来 1.6 倍加速The Memory Wall & Two Phases: Why 3.2x More FLOPS Buys Only 1.6x Speedup

MLSys·im 学习笔记（二）——从 Memory Wall 到 Prefill/Decode 双墙，一次 LLM 请求里的两套瓶颈MLSys·im Learning Notes (Part 2) — from the Memory Wall to the Prefill/Decode dual walls: two bottlenecks inside a single LLM request

内存墙与两阶段：为什么 3.2 倍算力只换来 1.6 倍加速The Memory Wall & Two Phases: Why 3.2x More FLOPS Buys Only 1.6x Speedup

MLSys·im 学习笔记（二）——从 Memory Wall 到 Prefill/Decode 双墙，一次 LLM 请求里的两套瓶颈MLSys·im Learning Notes (Part 2) — from the Memory Wall to the Prefill/Decode dual walls: two bottlenecks inside a single LLM request

交叉熵：语言模型的损失函数，为什么只能是它Cross-Entropy: The Loss Function of Language Models, and Why It Must Be This One

3Blue1Brown「压缩即智能」Part 2 学习笔记——从一个玩具编码方案讲交叉熵，为什么 p 和 q 不能对调，又为什么交叉熵公式是「平均损失仅在 q=p 时取最小值」的唯一解

GPU 为什么会挨饿：数据墙、投机解码与设计空间探索Why Does the GPU Starve: The Data Wall, Speculative Decoding, and Design-Space Exploration

MLSys·im 学习笔记（四）——从单点瓶颈走向全链路优化：CPU 预处理如何让 GPU 饿肚子，投机解码如何以小博大，DSE 如何在数千种配置中自动找最优解MLSys·im Learning Notes (Part 4) — from single-point analysis to whole-pipeline optimization: how CPU preprocessing starves the GPU, how speculative decoding bets small to win big, and how DSE searches thousands of configs automatically

KV-Cache 的显存账本：长上下文、并发与量化剪枝的取舍The KV-Cache Memory Ledger: Long Context, Concurrency, and the Quantization–Pruning Trade-off

MLSys·im 学习笔记（三）——用公式算清一个请求吃多少显存，序列长度如何决定并发上限，以及为什么"压缩"和"剪枝"在推理提速上完全是两回事MLSys·im Learning Notes (Part 3) — a formula for how much memory one request eats, why sequence length decides your concurrency ceiling, and why 'compression' and 'pruning' are two completely different stories for inference speedup

内存墙与两阶段：为什么 3.2 倍算力只换来 1.6 倍加速The Memory Wall & Two Phases: Why 3.2x More FLOPS Buys Only 1.6x Speedup

MLSys·im 学习笔记（二）——从 Memory Wall 到 Prefill/Decode 双墙，一次 LLM 请求里的两套瓶颈MLSys·im Learning Notes (Part 2) — from the Memory Wall to the Prefill/Decode dual walls: two bottlenecks inside a single LLM request

GPU 为什么会挨饿：数据墙、投机解码与设计空间探索Why Does the GPU Starve: The Data Wall, Speculative Decoding, and Design-Space Exploration

MLSys·im 学习笔记（四）——从单点瓶颈走向全链路优化：CPU 预处理如何让 GPU 饿肚子，投机解码如何以小博大，DSE 如何在数千种配置中自动找最优解MLSys·im Learning Notes (Part 4) — from single-point analysis to whole-pipeline optimization: how CPU preprocessing starves the GPU, how speculative decoding bets small to win big, and how DSE searches thousands of configs automatically

KV-Cache 的显存账本：长上下文、并发与量化剪枝的取舍The KV-Cache Memory Ledger: Long Context, Concurrency, and the Quantization–Pruning Trade-off

MLSys·im 学习笔记（三）——用公式算清一个请求吃多少显存，序列长度如何决定并发上限，以及为什么"压缩"和"剪枝"在推理提速上完全是两回事MLSys·im Learning Notes (Part 3) — a formula for how much memory one request eats, why sequence length decides your concurrency ceiling, and why 'compression' and 'pruning' are two completely different stories for inference speedup

内存墙与两阶段：为什么 3.2 倍算力只换来 1.6 倍加速The Memory Wall & Two Phases: Why 3.2x More FLOPS Buys Only 1.6x Speedup

MLSys·im 学习笔记（二）——从 Memory Wall 到 Prefill/Decode 双墙，一次 LLM 请求里的两套瓶颈MLSys·im Learning Notes (Part 2) — from the Memory Wall to the Prefill/Decode dual walls: two bottlenecks inside a single LLM request

给开源教程修了个 demoFixing a Tutorial Demo: An Open-Source PR Retrospective

给 mlsysim 的 KV-Cache 教程重写了 §5 的 Paged Attention demo 并提了 PRI rewrote the §5 Paged Attention demo in mlsysim's KV-Cache tutorial and opened PR #2135; before merging, the maintainer revised it and shrank my numbers from 314 to 259. A retrospective of that experience.

给开源教程修了个 demoFixing a Tutorial Demo: An Open-Source PR Retrospective

给 mlsysim 的 KV-Cache 教程重写了 §5 的 Paged Attention demo 并提了 PR #2135;合并前维护者又改了一版,把数字从 314 改小到 259。复盘这次经历。I rewrote the §5 Paged Attention demo in mlsysim's KV-Cache tutorial and opened PR #2135; before merging, the maintainer revised it and shrank my numbers from 314 to 259. A retrospective of that experience.

KV-Cache 的显存账本：长上下文、并发与量化剪枝的取舍The KV-Cache Memory Ledger: Long Context, Concurrency, and the Quantization–Pruning Trade-off

MLSys·im 学习笔记（三）——用公式算清一个请求吃多少显存，序列长度如何决定并发上限，以及为什么"压缩"和"剪枝"在推理提速上完全是两回事MLSys·im Learning Notes (Part 3) — a formula for how much memory one request eats, why sequence length decides your concurrency ceiling, and why 'compression' and 'pruning' are two completely different stories for inference speedup

KV-Cache 的显存账本：长上下文、并发与量化剪枝的取舍The KV-Cache Memory Ledger: Long Context, Concurrency, and the Quantization–Pruning Trade-off

MLSys·im 学习笔记（三）——用公式算清一个请求吃多少显存，序列长度如何决定并发上限，以及为什么"压缩"和"剪枝"在推理提速上完全是两回事MLSys·im Learning Notes (Part 3) — a formula for how much memory one request eats, why sequence length decides your concurrency ceiling, and why 'compression' and 'pruning' are two completely different stories for inference speedup

KV-Cache 的显存账本：长上下文、并发与量化剪枝的取舍The KV-Cache Memory Ledger: Long Context, Concurrency, and the Quantization–Pruning Trade-off

MLSys·im 学习笔记（三）——用公式算清一个请求吃多少显存，序列长度如何决定并发上限，以及为什么"压缩"和"剪枝"在推理提速上完全是两回事MLSys·im Learning Notes (Part 3) — a formula for how much memory one request eats, why sequence length decides your concurrency ceiling, and why 'compression' and 'pruning' are two completely different stories for inference speedup

KV-Cache 的显存账本：长上下文、并发与量化剪枝的取舍The KV-Cache Memory Ledger: Long Context, Concurrency, and the Quantization–Pruning Trade-off

MLSys·im 学习笔记（三）——用公式算清一个请求吃多少显存，序列长度如何决定并发上限，以及为什么"压缩"和"剪枝"在推理提速上完全是两回事MLSys·im Learning Notes (Part 3) — a formula for how much memory one request eats, why sequence length decides your concurrency ceiling, and why 'compression' and 'pruning' are two completely different stories for inference speedup

KV-Cache 的显存账本：长上下文、并发与量化剪枝的取舍The KV-Cache Memory Ledger: Long Context, Concurrency, and the Quantization–Pruning Trade-off

MLSys·im 学习笔记（三）——用公式算清一个请求吃多少显存，序列长度如何决定并发上限，以及为什么"压缩"和"剪枝"在推理提速上完全是两回事MLSys·im Learning Notes (Part 3) — a formula for how much memory one request eats, why sequence length decides your concurrency ceiling, and why 'compression' and 'pruning' are two completely different stories for inference speedup

KV-Cache 的显存账本：长上下文、并发与量化剪枝的取舍The KV-Cache Memory Ledger: Long Context, Concurrency, and the Quantization–Pruning Trade-off

MLSys·im 学习笔记（三）——用公式算清一个请求吃多少显存，序列长度如何决定并发上限，以及为什么"压缩"和"剪枝"在推理提速上完全是两回事MLSys·im Learning Notes (Part 3) — a formula for how much memory one request eats, why sequence length decides your concurrency ceiling, and why 'compression' and 'pruning' are two completely different stories for inference speedup

KV-Cache 的显存账本：长上下文、并发与量化剪枝的取舍The KV-Cache Memory Ledger: Long Context, Concurrency, and the Quantization–Pruning Trade-off

MLSys·im 学习笔记（三）——用公式算清一个请求吃多少显存，序列长度如何决定并发上限，以及为什么"压缩"和"剪枝"在推理提速上完全是两回事MLSys·im Learning Notes (Part 3) — a formula for how much memory one request eats, why sequence length decides your concurrency ceiling, and why 'compression' and 'pruning' are two completely different stories for inference speedup

KV-Cache 的显存账本：长上下文、并发与量化剪枝的取舍The KV-Cache Memory Ledger: Long Context, Concurrency, and the Quantization–Pruning Trade-off

MLSys·im 学习笔记（三）——用公式算清一个请求吃多少显存，序列长度如何决定并发上限，以及为什么"压缩"和"剪枝"在推理提速上完全是两回事MLSys·im Learning Notes (Part 3) — a formula for how much memory one request eats, why sequence length decides your concurrency ceiling, and why 'compression' and 'pruning' are two completely different stories for inference speedup

给开源教程修了个 demoFixing a Tutorial Demo: An Open-Source PR Retrospective

给 mlsysim 的 KV-Cache 教程重写了 §5 的 Paged Attention demo 并提了 PR #2135;合并前维护者又改了一版,把数字从 314 改小到 259。复盘这次经历。I rewrote the §5 Paged Attention demo in mlsysim's KV-Cache tutorial and opened PR #2135; before merging, the maintainer revised it and shrank my numbers from 314 to 259. A retrospective of that experience.

给开源教程修了个 demoFixing a Tutorial Demo: An Open-Source PR Retrospective

给 mlsysim 的 KV-Cache 教程重写了 §5 的 Paged Attention demo 并提了 PRI rewrote the §5 Paged Attention demo in mlsysim's KV-Cache tutorial and opened PR #2135; before merging, the maintainer revised it and shrank my numbers from 314 to 259. A retrospective of that experience.

给开源教程修了个 demoFixing a Tutorial Demo: An Open-Source PR Retrospective

给 mlsysim 的 KV-Cache 教程重写了 §5 的 Paged Attention demo 并提了 PR #2135;合并前维护者又改了一版,把数字从 314 改小到 259。复盘这次经历。I rewrote the §5 Paged Attention demo in mlsysim's KV-Cache tutorial and opened PR #2135; before merging, the maintainer revised it and shrank my numbers from 314 to 259. A retrospective of that experience.

GPU 为什么会挨饿：数据墙、投机解码与设计空间探索Why Does the GPU Starve: The Data Wall, Speculative Decoding, and Design-Space Exploration

MLSys·im 学习笔记（四）——从单点瓶颈走向全链路优化：CPU 预处理如何让 GPU 饿肚子，投机解码如何以小博大，DSE 如何在数千种配置中自动找最优解MLSys·im Learning Notes (Part 4) — from single-point analysis to whole-pipeline optimization: how CPU preprocessing starves the GPU, how speculative decoding bets small to win big, and how DSE searches thousands of configs automatically

GPU 为什么会挨饿：数据墙、投机解码与设计空间探索Why Does the GPU Starve: The Data Wall, Speculative Decoding, and Design-Space Exploration

MLSys·im 学习笔记（四）——从单点瓶颈走向全链路优化：CPU 预处理如何让 GPU 饿肚子，投机解码如何以小博大，DSE 如何在数千种配置中自动找最优解MLSys·im Learning Notes (Part 4) — from single-point analysis to whole-pipeline optimization: how CPU preprocessing starves the GPU, how speculative decoding bets small to win big, and how DSE searches thousands of configs automatically

GPU 为什么会挨饿：数据墙、投机解码与设计空间探索Why Does the GPU Starve: The Data Wall, Speculative Decoding, and Design-Space Exploration

MLSys·im 学习笔记（四）——从单点瓶颈走向全链路优化：CPU 预处理如何让 GPU 饿肚子，投机解码如何以小博大，DSE 如何在数千种配置中自动找最优解MLSys·im Learning Notes (Part 4) — from single-point analysis to whole-pipeline optimization: how CPU preprocessing starves the GPU, how speculative decoding bets small to win big, and how DSE searches thousands of configs automatically

GPU 为什么会挨饿：数据墙、投机解码与设计空间探索Why Does the GPU Starve: The Data Wall, Speculative Decoding, and Design-Space Exploration

MLSys·im 学习笔记（四）——从单点瓶颈走向全链路优化：CPU 预处理如何让 GPU 饿肚子，投机解码如何以小博大，DSE 如何在数千种配置中自动找最优解MLSys·im Learning Notes (Part 4) — from single-point analysis to whole-pipeline optimization: how CPU preprocessing starves the GPU, how speculative decoding bets small to win big, and how DSE searches thousands of configs automatically

GPU 为什么会挨饿：数据墙、投机解码与设计空间探索Why Does the GPU Starve: The Data Wall, Speculative Decoding, and Design-Space Exploration

MLSys·im 学习笔记（四）——从单点瓶颈走向全链路优化：CPU 预处理如何让 GPU 饿肚子，投机解码如何以小博大，DSE 如何在数千种配置中自动找最优解MLSys·im Learning Notes (Part 4) — from single-point analysis to whole-pipeline optimization: how CPU preprocessing starves the GPU, how speculative decoding bets small to win big, and how DSE searches thousands of configs automatically

GPU 为什么会挨饿：数据墙、投机解码与设计空间探索Why Does the GPU Starve: The Data Wall, Speculative Decoding, and Design-Space Exploration

MLSys·im 学习笔记（四）——从单点瓶颈走向全链路优化：CPU 预处理如何让 GPU 饿肚子，投机解码如何以小博大，DSE 如何在数千种配置中自动找最优解MLSys·im Learning Notes (Part 4) — from single-point analysis to whole-pipeline optimization: how CPU preprocessing starves the GPU, how speculative decoding bets small to win big, and how DSE searches thousands of configs automatically

给开源教程修了个 demoFixing a Tutorial Demo: An Open-Source PR Retrospective

给 mlsysim 的 KV-Cache 教程重写了 §5 的 Paged Attention demo 并提了 PRI rewrote the §5 Paged Attention demo in mlsysim's KV-Cache tutorial and opened PR #2135; before merging, the maintainer revised it and shrank my numbers from 314 to 259. A retrospective of that experience.

千卡训练的隐形账单：通信不是瓶颈The Hidden Bill of Thousand-GPU Training: Communication Is Not the Bottleneck

MLSys·im 学习笔记（五）——从单节点走到千卡集群：谁在偷走训练时间、Young-Daly 公式该往哪用、选址怎么在 61 倍里改写碳足迹MLSys·im Learning Notes (Part 5) — from one node to a thousand GPUs: who actually steals your training time, where the Young-Daly formula should point, and how site selection rewrites the carbon footprint by 61x

千卡训练的隐形账单：通信不是瓶颈The Hidden Bill of Thousand-GPU Training: Communication Is Not the Bottleneck

MLSys·im 学习笔记（五）——从单节点走到千卡集群：谁在偷走训练时间、Young-Daly 公式该往哪用、选址怎么在 61 倍里改写碳足迹MLSys·im Learning Notes (Part 5) — from one node to a thousand GPUs: who actually steals your training time, where the Young-Daly formula should point, and how site selection rewrites the carbon footprint by 61x

千卡训练的隐形账单：通信不是瓶颈The Hidden Bill of Thousand-GPU Training: Communication Is Not the Bottleneck

MLSys·im 学习笔记（五）——从单节点走到千卡集群：谁在偷走训练时间、Young-Daly 公式该往哪用、选址怎么在 61 倍里改写碳足迹MLSys·im Learning Notes (Part 5) — from one node to a thousand GPUs: who actually steals your training time, where the Young-Daly formula should point, and how site selection rewrites the carbon footprint by 61x

千卡训练的隐形账单：通信不是瓶颈The Hidden Bill of Thousand-GPU Training: Communication Is Not the Bottleneck

MLSys·im 学习笔记（五）——从单节点走到千卡集群：谁在偷走训练时间、Young-Daly 公式该往哪用、选址怎么在 61 倍里改写碳足迹MLSys·im Learning Notes (Part 5) — from one node to a thousand GPUs: who actually steals your training time, where the Young-Daly formula should point, and how site selection rewrites the carbon footprint by 61x

千卡训练的隐形账单：通信不是瓶颈The Hidden Bill of Thousand-GPU Training: Communication Is Not the Bottleneck

MLSys·im 学习笔记（五）——从单节点走到千卡集群：谁在偷走训练时间、Young-Daly 公式该往哪用、选址怎么在 61 倍里改写碳足迹MLSys·im Learning Notes (Part 5) — from one node to a thousand GPUs: who actually steals your training time, where the Young-Daly formula should point, and how site selection rewrites the carbon footprint by 61x

千卡训练的隐形账单：通信不是瓶颈The Hidden Bill of Thousand-GPU Training: Communication Is Not the Bottleneck

MLSys·im 学习笔记（五）——从单节点走到千卡集群：谁在偷走训练时间、Young-Daly 公式该往哪用、选址怎么在 61 倍里改写碳足迹MLSys·im Learning Notes (Part 5) — from one node to a thousand GPUs: who actually steals your training time, where the Young-Daly formula should point, and how site selection rewrites the carbon footprint by 61x

千卡训练的隐形账单：通信不是瓶颈The Hidden Bill of Thousand-GPU Training: Communication Is Not the Bottleneck

MLSys·im 学习笔记（五）——从单节点走到千卡集群：谁在偷走训练时间、Young-Daly 公式该往哪用、选址怎么在 61 倍里改写碳足迹MLSys·im Learning Notes (Part 5) — from one node to a thousand GPUs: who actually steals your training time, where the Young-Daly formula should point, and how site selection rewrites the carbon footprint by 61x

千卡训练的隐形账单：通信不是瓶颈The Hidden Bill of Thousand-GPU Training: Communication Is Not the Bottleneck

MLSys·im 学习笔记（五）——从单节点走到千卡集群：谁在偷走训练时间、Young-Daly 公式该往哪用、选址怎么在 61 倍里改写碳足迹MLSys·im Learning Notes (Part 5) — from one node to a thousand GPUs: who actually steals your training time, where the Young-Daly formula should point, and how site selection rewrites the carbon footprint by 61x

千卡训练的隐形账单：通信不是瓶颈The Hidden Bill of Thousand-GPU Training: Communication Is Not the Bottleneck

MLSys·im 学习笔记（五）——从单节点走到千卡集群：谁在偷走训练时间、Young-Daly 公式该往哪用、选址怎么在 61 倍里改写碳足迹MLSys·im Learning Notes (Part 5) — from one node to a thousand GPUs: who actually steals your training time, where the Young-Daly formula should point, and how site selection rewrites the carbon footprint by 61x

千卡训练的隐形账单：通信不是瓶颈The Hidden Bill of Thousand-GPU Training: Communication Is Not the Bottleneck

MLSys·im 学习笔记（五）——从单节点走到千卡集群：谁在偷走训练时间、Young-Daly 公式该往哪用、选址怎么在 61 倍里改写碳足迹MLSys·im Learning Notes (Part 5) — from one node to a thousand GPUs: who actually steals your training time, where the Young-Daly formula should point, and how site selection rewrites the carbon footprint by 61x

交叉熵：语言模型的损失函数，为什么只能是它Cross-Entropy: The Loss Function of Language Models, and Why It Must Be This One

3Blue1Brown「压缩即智能」Part 2 学习笔记——从一个玩具编码方案讲交叉熵，为什么 p 和 q 不能对调，又为什么交叉熵公式是「平均损失仅在 q=p 时取最小值」的唯一解

交叉熵：语言模型的损失函数，为什么只能是它Cross-Entropy: The Loss Function of Language Models, and Why It Must Be This One

3Blue1Brown「压缩即智能」Part 2 学习笔记——从一个玩具编码方案讲交叉熵，为什么 p 和 q 不能对调，又为什么交叉熵公式是「平均损失仅在 q=p 时取最小值」的唯一解

交叉熵：语言模型的损失函数，为什么只能是它Cross-Entropy: The Loss Function of Language Models, and Why It Must Be This One

3Blue1Brown「压缩即智能」Part 2 学习笔记——从一个玩具编码方案讲交叉熵，为什么 p 和 q 不能对调，又为什么交叉熵公式是「平均损失仅在 q=p 时取最小值」的唯一解

交叉熵：语言模型的损失函数，为什么只能是它Cross-Entropy: The Loss Function of Language Models, and Why It Must Be This One

3Blue1Brown「压缩即智能」Part 2 学习笔记——从一个玩具编码方案讲交叉熵，为什么 p 和 q 不能对调，又为什么交叉熵公式是「平均损失仅在 q=p 时取最小值」的唯一解

交叉熵：语言模型的损失函数，为什么只能是它Cross-Entropy: The Loss Function of Language Models, and Why It Must Be This One

3Blue1Brown「压缩即智能」Part 2 学习笔记——从一个玩具编码方案讲交叉熵，为什么 p 和 q 不能对调，又为什么交叉熵公式是「平均损失仅在 q=p 时取最小值」的唯一解

Readnest:我的阅读记录,为什么要搬出 NotionReadnest: Why I Moved My Reading Log Out of Notion

一个只存在于自己手机里的阅读记录本A reading log that lives on your phone and nowhere else

Readnest:我的阅读记录,为什么要搬出 NotionReadnest: Why I Moved My Reading Log Out of Notion

一个只存在于自己手机里的阅读记录本A reading log that lives on your phone and nowhere else

Readnest:我的阅读记录,为什么要搬出 NotionReadnest: Why I Moved My Reading Log Out of Notion

一个只存在于自己手机里的阅读记录本A reading log that lives on your phone and nowhere else

Readnest:我的阅读记录,为什么要搬出 NotionReadnest: Why I Moved My Reading Log Out of Notion

一个只存在于自己手机里的阅读记录本A reading log that lives on your phone and nowhere else