Live page · Day archive

ALoDLM: Adaptively Looped Diffusion Language Models

Read the original at HF Daily Papers

Summary

Researchers introduce ALoDLM, a diffusion language model that allocates computation based on token difficulty. The system is trained at 1.7B and 8B parameter scales.

Carried by: HF Daily Papers. First seen: .