Papers6

#LLaDA

Balancing Understanding and Generation in Discrete Diffusion Models

This paper introduces XDLM, a single model that blends two popular diffusion styles (masked and uniform) so it both understands and generates text and images well.

#XDLM#discrete diffusion#stationary noise kernel

FourierSampler: Unlocking Non-Autoregressive Potential in Diffusion Language Models via Frequency-Guided Generation

Intermediate

Siyang He, Qiqi Wang et al.Jan 30arXiv

Diffusion language models (dLLMs) can write text in any order, but common decoding methods still prefer left-to-right, which wastes their superpower.

#diffusion language models#non-autoregressive generation#frequency-domain analysis

DIFFA-2: A Practical Diffusion Large Language Model for General Audio Understanding

Intermediate

Jiaming Zhou, Xuxin Cheng et al.Jan 30arXiv

DIFFA-2 is a new audio AI that listens to speech, sounds, and music and answers questions about them using a diffusion-style language model instead of the usual step-by-step (autoregressive) method.

#Diffusion language models#Audio understanding#Large audio language model

Residual Context Diffusion Language Models

Intermediate

Yuezhou Hu, Harman Singh et al.Jan 30arXiv

Diffusion language models (dLLMs) generate several tokens at once but usually throw away lots of helpful clues each step—RCD keeps and reuses those clues.

#diffusion language models#residual context diffusion#soft tokens

Efficient-DLM: From Autoregressive to Diffusion Language Models, and Beyond in Speed

Intermediate

Yonggan Fu, Lexington Whalen et al.Dec 16arXiv

Autoregressive (AR) models write one word at a time, which is accurate but slow, especially when your computer or GPU can’t keep many tasks in memory at once.

#diffusion language models#autoregressive models#AR-to-dLM conversion

Fast-Decoding Diffusion Language Models via Progress-Aware Confidence Schedules

Beginner

Amr Mohamed, Yang Zhang et al.Dec 2arXiv

Diffusion language models (dLLMs) can write all parts of an answer in parallel, but they usually take many tiny cleanup steps, which makes them slow.

#diffusion language models#early exit decoding#progress-aware threshold