Open eu9ene opened 3 months ago
Example training log from HPLT https://github.com/hplt-project/HPLT-MT-Models/blob/main/v1.0/training/en-zh_hant/hplt/model.train.log
I changed the vocabulary size to 64000 based on the HPLT config. I don't yet understand the implications.
Chinese should be trained in two different models, one for simplified and one for traditional, as both scripts might be too large vocabulary to fit in 32k pieces.
Does it require any adjustment? Should we change any hyperparameters etc.?