Nano-vLLM: A Lightweight vLLM Implementation with Blazing Speed

Nano-vLLM is a lightweight implementation of vLLM, built from scratch in approximately 1200 lines of Python code. Despite its small size, it achieves inference speeds comparable to the original vLLM. It incorporates various optimizations such as prefix caching, tensor parallelism, Torch compilation, and CUDA graphs. Install via `pip install git+https://github.com/GeeeekExplorer/nano-vllm.git` and refer to example.py for usage. Benchmarks on an RTX 4070 Laptop (8GB) with the Qwen3-0.6B model show throughput slightly exceeding vLLM.