Amazon's ALoDLM Speeds Up Diffusion LMs to 612 tok/s

Amazon AGI released ALoDLM, an adaptively looped diffusion language model that assigns looping depth per token based on prediction difficulty, achieving 612 tokens per second. Code and 1.7B/8B models are open-sourced.

2026-10-06 ~ 2026-10-06 · 2 related posts

• Amazon's ALoDLM: Token-Adaptive Looped Diffusion LMs Beat AR Baselines — amazon · 2026-10-06
• Amazon AGI's ALoDLM Beats Diffusion and AR Baselines, Hits 612 tok/s at 8B — arankomatsuzaki · 2026-10-06