Overfitted Opinions

I write things down here instead of explaining them at 2 a.m. to someone trying to sleep.

Cover image for "vLLM: LLM Inference as a memory and scheduling problem"

Latest

vLLM: LLM Inference as a memory and scheduling problem

Practical deep dive into vLLM: prefill vs decode, the KV-cache bottleneck, PagedAttention, and how modern attention backends consume paged KV.

Jul 30, 202616 min read

  • large language models
  • vLLM
  • LLM inference
  • systems
  • GPU
  • model serving
  • distributed inference
  • PagedAttention

Archive

Cover image for "Hello world!"

Hello world!

Notion Test Blog

Jun 20, 20252 min read

  • hello-world