Hello,
I followed the tutorial in the README to train DiffSHEG, using the beat_english_v0.2.1 dataset, and the loss curve converged normally, but the test results were very poor. Where is wrong?
wandb:
Visualization:
https://github.com/user-attachments/assets/55e75a6c-df96-463e-bcf4-a8e09cbeb3b5