Open access
Jul 2026
Adaptive Window Diffusion Decoding for Long-Context Language Generation Under Memory Constraints
Adaptive Window Diffusion Decoding (AWDD) is proposed, a memory-budgeted inference framework that adjusts window size, overlap commitment, and context-summary capacity according to observable uncertainty in the partially masked sequence to improve long-range entity reconstruction over fixed-window baselines.
N. Petrov, Rafael Costa, Clara Hoffmann
· International Journal of Adv... · 0 citations