Acceleration Multiple Heads Decoding for LLM via Dynamic Tree Attention

Re-assign community