LONG-FORM / ENGINEERING
Articles.
Deeper explorations of the systems we build and the ideas behind them.
- ↗
LLM 推理:从 Prefill 到 Decode
seedling从推理引擎视角理解模型计算、KV Cache、请求调度与性能指标。
- ↗
Understanding the LSM tree
evergreenA write path built around sequential I/O — and the trade-offs that follow.
- ↗
Think in access paths, selectivity, and the rows you do not want to read.