Last released Jun 9, 2026
Retrieval-preserving hierarchical KV cache compression for long-context LLM inference
Supported by