[BUG] push_to_huggingface sometimes include tags in resulting argilla.yml file

[gabrielmbmb]

Describe the bug
In some environments, when using FeedbackDataset.push_to_argilla, it generates an argilla.yaml file that includes tags that later prevents the from_huggingface function to work correctly as they yaml cannot be loaded back because of the tags.

allow_extra_metadata: true
fields:
- name: prompt
  required: true
  title: Prompt
  type: !!python/object/apply:argilla.client.feedback.schemas.enums.FieldTypes # <-- tag here
  - text
  use_markdown: true

Expected behavior
No matter what environment is used, the push_to_argilla should generate the yaml without tags.

Environment:

• Argilla Version [e.g. 1.0.0]: 1.18.0
• ElasticSearch Version [e.g. 7.10.2]: not relevant
• Docker Image (optional) [e.g. argilla:v1.0.0]: not relevant

[davidberenstein1957]

• @alvarobartt create a patch
• @alvarobartt @frascuchon @gabrielmbmb provide a better fix

[jfcalvo]

I add a link here to a possible solution suggested by @frascuchon that it's use use_enum_values in the Pydantic config for the affected model.

[jfcalvo]

I couldn't reproduce the problem locally but use_enum_values looks like working:

In [17]: class Response(BaseModel):
    ...:     status: ResponseStatus
    ...:

In [18]: Response(status=ResponseStatus.draft)
Out[18]: Response(status=<ResponseStatus.draft: 'draft'>)

In [19]: class Response(BaseModel):
    ...:     status: ResponseStatus
    ...:     class Config:
    ...:         use_enum_values = True
    ...:

In [20]: Response(status=ResponseStatus.draft)
Out[20]: Response(status='draft')