redknot

Efficient long-context LLM serving framework using head-aware KV reuse and SegPagedAttention to accelerate inference for developers deploying memory-intensive language models.