Dynamic-Width Speculative Beam Decoding for Efficient LLM Inference

Re-assign community