A technical reference covering the full inference stack of DeepSeek-V3 on an 8-node × 8 H100 cluster — from serving scheduler to GPU kernel to distributed communication.
| Language | Entry Point |
|---|---|
| 中文 | zh/README.md |
| English | en/README.md |
zh/ — Chinese edition
en/ — English edition
├── deployment-config.md
├── README.md (chapter overview)
├── ch1-serving/
├── ch2-prefill/
├── ch3-moe/
├── ch4-decode/
├── ch5-distributed/
└── ch6-performance/
Each chapter follows the inference execution path of a single request, using concrete numbers derived from the fixed deployment configuration.