feat: add ArgillaSpaCyTransformersTrainer & improve ArgillaSpaCyTrainer

[alvarobartt]

Description

This PR adds support for spacy-transformers via the new ArgillaSpaCyTransformersTrainer class allowing the user to lock the transformer model not to be updated, and the ArgillaSpaCyTrainer is improved to allow re-using the tok2vec or freeze it, if available.

Besides that, this PR also includes a new arg in ArgillaTrainer named framework_kwargs which is a Python dict to contain the framework-specific kwargs, in this case intended to be created as {"update_transformer": False} and {"freeze_tok2vec": False} for both ArgillaSpaCyTransformersTrainer and ArgillaSpaCyTrainer, respectively. Ideally, we should also move the spaCy specific-args already included as ArgillaTrainer args.

Type of change

• New feature (non-breaking change which adds functionality)
• Improvement (change adding some improvement to an existing functionality)

How Has This Been Tested

• Add unit-tests for ArgillaSpaCyTransformersTrainer

Checklist

• I added relevant documentation
• follows the style guidelines of this project
• I did a self-review of my code
• I made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• I filled out the contributor form (see text above)
• I have added relevant notes to the CHANGELOG.md file (See https://keepachangelog.com/)

[davidberenstein1957]

LGTM and thanks for the fix. Perhaps it makes more sense to me to pass the variable via the update_config to align with the usage of the rest of the frameworks. Also, I don't see the new config options being test right?

[davidberenstein1957]

Hi @alvarobartt, could you also include the implementation and tests for the FeedbackDataset?

[alvarobartt]

Hi @alvarobartt, could you also include the implementation and tests for the FeedbackDataset?

Not sure I'll have time for this release, need to focus now on the RankingQuestion, but we can create a ticket for the next one

[davidberenstein1957]

I can have a look tomorrow afternoon. I implemented most stuff so it should only require some small changes.

[alvarobartt]

Hi @davidberenstein1957 can you have a look into the unit test that is failing? Is there anything that you changed w.r.t. ArgillaSpaCyTrainer in the ArgillaTrainer for the FeedbackDatasets? I guess we're good to merge as the failing unit test is unrelated?

[davidberenstein1957]

@alvarobartt, I added the integration and some test. Can you have a final look tomorrow before merging?

[codecov]

Codecov Report

Patch coverage: 84.98% and project coverage change: -0.77 ⚠️

Comparison is base (51751ac) 90.91% compared to head (d1c9ccd) 90.14%.

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #3256      +/-   ##
===========================================
- Coverage    90.91%   90.14%   -0.77%     
===========================================
  Files          215      233      +18     
  Lines        11304    12493    +1189     
===========================================
+ Hits         10277    11262     +985     
- Misses        1027     1231     +204     

Flag	Coverage Δ
pytest	90.14% <84.98%> (-0.77%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
src/argilla/__init__.py	86.66% <ø> (+3.33%)	⬆️
...illa/client/feedback/training/frameworks/openai.py	0.00% <0.00%> (ø)
...rgilla/client/feedback/training/frameworks/peft.py	0.00% <0.00%> (ø)
...client/feedback/training/frameworks/span_marker.py	0.00% <0.00%> (ø)
src/argilla/server/contexts/datasets.py	96.01% <ø> (ø)
src/argilla/server/seeds.py	0.00% <ø> (ø)
src/argilla/tasks/training/__main__.py	30.00% <0.00%> (-1.58%)	⬇️
src/argilla/tasks/users/create.py	91.11% <ø> (-4.45%)	⬇️
src/argilla/training/autotrain_advanced.py	0.00% <0.00%> (ø)
src/argilla/training/peft.py	0.00% <0.00%> (ø)
... and 62 more

... and 5 files with indirect coverage changes

☔ View full report in Codecov by Sentry.
📢 Do you have feedback about the report comment? Let us know in this issue.