Inference acceleration for large language models using "stairs" assisted greedy generation

Re-assign community