XiangLi1999/Diffusion-LM Issues
E2E training procedure
UpdatedQuestions about the NLL loss
UpdatedAbout the tT_loss
UpdatedBaseline reproduction
UpdatedWhy not directly use Emb(W) as X_0?
Updated 2Training on A100
UpdatedLosses for E2E Training
Closed 2Where is the mbr.py file?
Closed 1Wandb log or Codalab log
Updated 1about {path-to-diffusion-lm}
Updated 6Are these normal results?
Updated 13The effect of "logp_term"
Updated 1How to control the length
Updated 1License
Closed 2Train on Multi GPU
Updated 3problem about attention_mask
Closed 2