Update example scripts relying on `model.fit` to use `SentenceTransformerTrainer` instead

[tomaarsen]

Hello!

Feature Request overview

• Many example scripts use model.fit, while I softly deprecated that method in v3.0 in favor of SentenceTransformerTrainer

Details

Many example scripts use model.fit, while I softly deprecated that method in v3.0 in favor of SentenceTransformerTrainer. I wrote a migration guide here: https://sbert.net/docs/migration_guide.html#migrating-from-v2-x-to-v3-x

But never carried out the upgrade myself on all evaluation scripts. I did a bunch myself in #2622, but there's some cases that I never tackled. For some of those it might not be possible to get a 1-1 perfect match (e.g. I believe one MS MARCO script uses model.fit because then it can fully regenerate the dataset for each batch), but it should be possible to get close.

Here's some that aren't done yet: https://github.com/search?q=repo%3Ahuggingface%2Fsentence-transformers+model.fit+language%3APython+path%3A%2F%5Eexamples%5C%2Fsentence_transformer%5C%2F%2F&type=code

If you're interested in working on this, feel free to take on just one case; I'm not expecting anyone to update all of them.

• Tom Aarsen

[itsdevandy]

Update: Looks like someone else got to this first, thanks!

Hi @tomaarsen, I’d like to pick this up and start working on it.

I'll work on migrating these two from the listed examples:

1. train_stsb_simcse.py
2. train_bi-encoder_margin-mse.py

[harshitsharma496]

Hi @tomaarsen

Picking up :

1.examples/sentence_transformer/unsupervised_learning/CT_In-Batch_Negatives/train_ct-improved_from_file.py
2. examples/sentence_transformer/unsupervised_learning/CT/train_askubuntu_ct.py

examples/sentence_transformer/training/quora_duplicate_questions/training_multi-task-learning.py --> This one is having model.fit() in docstring only not in actual code, will do this one too.

[omkar-334]

Picking up

unsupervised_learning/SimCSE/train_simcse_from_file.py
unsupervised_learning/SimCSE/train_askubuntu_simcse.py

as well

[ritoban23]

I'm working on -train_bi-encoder_mnrl.py