ValueError when using prepare_for_training() on multi-label annotated Argilla dataset with single annotations

[m-newhauser]

Describe the bug
I'm getting a ValueError when trying to use the prepare_for_training() method to prepare a multi-label annotated Argilla dataset for training. I'm only getting the error when a given record has just a single annotation. Everything works fine if all records have more than one assigned annotation label.

To Reproduce

import datetime
import argilla as rg

# Create dataset with a multi-label record that only has one annotation
dataset_rg = rg.DatasetForTextClassification(
    [
        rg.TextClassificationRecord(text=None,
                                    inputs={'title': 'This is the title of an article'}, 
                                    prediction=None, 
                                    prediction_agent=None, 
                                    annotation=['LABEL_1'], 
                                    annotation_agent='team', 
                                    vectors=None, 
                                    multi_label=True, 
                                    explanation=None, 
                                    id='12345', 
                                    metadata={'split': 'train'}, 
                                    status='Validated', 
                                    event_timestamp=datetime.datetime(2023, 4, 4, 9, 57, 59, 910986), 
                                    search_keywords=None)
    ]
)

# Prepare the dataset for training
dataset_rg.prepare_for_training(framework="setfit")

Generates the following error:

ValueError: Class label 1 greater than configured num_classes 1

Expected behavior
Expect the method to return a Dataset that is ready for training.

Dataset({
    features: ['id', 'text', 'label', 'binarized_label'],
    num_rows: 1
})

Environment (please complete the following information):

• OS [e.g. iOS]: GitHub Codespaces - Ubuntu Linux
• Argilla Version [e.g. 1.0.0]: 1.6.0

Additional context
The method works properly when the given record is multi-label AND has more than one annotation label, for example:

import datetime
import argilla as rg

# Create dataset with a multi-label record that only has one annotation
dataset_rg = rg.DatasetForTextClassification(
    [
        rg.TextClassificationRecord(text=None,
                                    inputs={'title': 'This is the title of an article'}, 
                                    prediction=None, 
                                    prediction_agent=None, 
                                    annotation=['LABEL_1', 'LABEL_2'], 
                                    annotation_agent='team', 
                                    vectors=None, 
                                    multi_label=True, 
                                    explanation=None, 
                                    id='12345', 
                                    metadata={'split': 'train'}, 
                                    status='Validated', 
                                    event_timestamp=datetime.datetime(2023, 4, 4, 9, 57, 59, 910986), 
                                    search_keywords=None)
    ]
)

# Prepare the dataset for training
dataset_rg.prepare_for_training(framework="setfit")

Returns:

Dataset({
    features: ['id', 'text', 'label', 'binarized_label'],
    num_rows: 1
})

[davidberenstein1957]

@m-newhauser thanks again for reporting this. This was resolved in 1.7.0