Hi! I maintain nano-vLLM Interactive Guide, a MIT-licensed, browser-based learning project for understanding vLLM-style inference internals.
It provides Chinese-first concept labs, source anchors, teaching traces, and an optional CUDA runtime path for:
- PagedAttention and KV-block allocation
- continuous batching and scheduler decisions
- Prefix Cache reuse boundaries
- prefill/decode behavior, sampling, tensor parallelism, CUDA Graphs, and benchmark evidence boundaries
Repository: https://github.com/lora-sys/nano-vllm-interactive-guide
Live guide: https://lora-sys.github.io/nano-vllm-interactive-guide/
English discovery page: https://lora-sys.github.io/nano-vllm-interactive-guide/en/
The project explicitly distinguishes concept simulations from real CUDA traces and benchmark claims. Since this list is strongly focused on inference papers and code, would a single entry in a Learning Resources or Tutorials subsection be in scope? If not, no action is needed; I would prefer to follow the list's intended taxonomy.
Hi! I maintain nano-vLLM Interactive Guide, a MIT-licensed, browser-based learning project for understanding vLLM-style inference internals.
It provides Chinese-first concept labs, source anchors, teaching traces, and an optional CUDA runtime path for:
Repository: https://github.com/lora-sys/nano-vllm-interactive-guide
Live guide: https://lora-sys.github.io/nano-vllm-interactive-guide/
English discovery page: https://lora-sys.github.io/nano-vllm-interactive-guide/en/
The project explicitly distinguishes concept simulations from real CUDA traces and benchmark claims. Since this list is strongly focused on inference papers and code, would a single entry in a Learning Resources or Tutorials subsection be in scope? If not, no action is needed; I would prefer to follow the list's intended taxonomy.