Based on Nano-vLLM, a simple replication of vLLM with self-contained paged attention and flash attention implementation
Open on GitHub