A 198.7 μJ/token Block Diffusion LLM Processor with Mask Token Similarity-Based Activation Reuse.

Yujin Moon, Jiwon Choi, Seryeong Kim, Wonhoon Park, Wooyoung Jo, Yuseon Choi, Sunjoo Whang, Hoi-Jun Yoo

Browse the full ISCAS paper archive.