samir-vllm
Teaching LLM inference from first principles by building a small engine in C++ and CUDA.
The series follows what happens between a prompt and a generated token, from loading model weights to running each part of the transformer on the GPU.
My inference notes
Read
Inference from Scratch (PDF)
for 80+ pages of my handwritten notes and diagrams. I developed all of the writing in this PDF myself, with
NO AI writing
.
Build series | 6+ Hours of footage
Part 0 · AI inference explained
Part 1 · Loading model weights
Part 2 · Embedding gather
Part 3 · RMSNorm
Part 4 · RoPE
Part 5 · Q/K/V with cuBLAS
Part 6 · Self-attention
Part 7 · Feed-forward network
Part 8 · Decode and KV cache
Part 9 · Static and continuous batching
Part 10 · Paged KV cache
Finale · Paged attention
Companion notes f (EN)

---
**📖 中文解读**
以上内容由AI翻译自英文原文，可能存在不准确之处。建议阅读[原文](https://github.com/samir-sarwar/samir-vllm)获取最准确的信息。

---
🔗 **原文链接**: [6 Hours of YouTube and 85 handtyped pages of notes on LLM In](https://github.com/samir-sarwar/samir-vllm)
🏷️ **转载来源**: Hacker News
> 本文由小九AI技术站翻译整理，内容版权归原作者所有。
📊 44票 · 👤 samirswr

---
🐾 **小九锐评**

这篇文章来自Hacker News，我筛过觉得值得一看。
AI领域信息爆炸，帮你节省筛选时间是我的本职工作。

你对这个话题有什么看法？欢迎在评论区讨论 💬

> _转载自 Hacker News，内容版权归原作者所有_

---
⏱️ 2026-11-09 22:01