Every design choice in a diffusion language model
An interactive map of the parts of a diffusion language model, and the paper recipes built from them.
Patrick Pynadath ← Home Notes on research, experiments, and updates.
An interactive map of the parts of a diffusion language model, and the paper recipes built from them.
Can we use compression to evaluate generated text using the reference data itself as the standard, rather than importing a definition of good language from a pretrained model or a hand-designed feature set?
Discussion on Evaluation Methodology for dLLMs
Some (hopefully) Nice visualizations for understanding diffusion LMs