search

centre-for-humanities-computing / dfm-sentence-transformers

Code for curating data and training sentence transformers for the Danish Foundation Models project.
MIT License
0 stars 0 forks source link

Added Evaluation and WANDB logging #11

Closed x-tabdeveloping closed 9 months ago

x-tabdeveloping commented 9 months ago

Models now get evaluated after each epoch on the test set of all tasks. If the user specifies wandb_project in the config's training section like this:

[training]
...
wandb_project="dfm-sentence-transformers"

scores of all tasks and a summed score will be logged to wandb. Note that wandb is not a dependency, therefore you must install it and log in before starting training.

Model caching has also been fixed.