Ofek Avivi

Different Tasks, Different Timesteps: Temporal Task-Aware Quantization for Diffusion Language Models

Ofek Avivi*, Eliron Rahimi*, Avi Mendelson, Chaim Baskin · *equal contribution

T²Quant gives each denoising step of a diffusion LM its own activation precision, per task, from a label-free signal. It improves three recent quantizers on every model at the same 4-bit budget, by up to 6.7 points.

@misc{avivi2026t2quant,
  title         = {Different Tasks, Different Timesteps: Temporal Task-Aware Quantization for Diffusion Language Models},
  author        = {Avivi, Ofek and Rahimi, Eliron and Mendelson, Avi and Baskin, Chaim},
  year          = {2026},
  eprint        = {2610.31198},
  archivePrefix = {arXiv},
  primaryClass  = {cs.AI},
  url           = {https://arxiv.org/abs/2610.31198}
}

Unigrade

A free GPA tracker and degree planner for university students: grades, GPA, degree planning and career options in one place.