ChunkAttention: Efficient Self-Attention with Prefix-Aware KV Cache and Two-Phase Partition

Open in new window