radd-lambda-dce / README.md
jingyang Ou
update model
f0c5b40

Reparameterized Absorbing Discrete Diffusion (RADD) small model with lambda-dce loss trained for 400k iterations.

Code: https://github.com/ML-GSAI/RADD.

Paper: https://arxiv.org/abs/2406.03736.