the Loss and Html are inf all the time when html training, is there anyone solved this?

[zodiac50]

[savi8sant8s]

I had the same problem today. Anyone solved?

[savi8sant8s]

I identified that in my case, I needed to map some tokens in vocab_html.json and constant.py that were not initially present in the model. After adding them, the loss started to appear normally.