rednote-machine-learning / RedKnot

Efficient Long-Context LLM Serving with Head-Aware KV Reuse and SegPagedAttention

52Hz-kiko / mini-vllm-qwen3-serving

A Mini-vLLM optimization project for Qwen3-32B TP=2 serving, Prefix Cache, and Triton PagedAttention.