Within Argilla we decided to differentiate our docs using main terminology classes and corresponding sub-classes.


Specific features that are covered by internal Argilla functionalities.




Internal Dataset classes are lightweight containers for Argilla records.


Argilla Metrics enables you to perform fine-grained analysis of your models and training datasets.


Argilla query functionalities are based on the powerful Elasticsearch query string syntax.

Semantic Search

This built-in search uses vectors for text and enables Approximate KNN for semantic search on these vectors.

MLOps Steps#

All steps that we directly or in-directly cover within the MLOps lifecycle.



๐Ÿท Labelling

manual or automatic data collection and label assignment.

๐Ÿ’ช๐Ÿฝ Training

training and evaluation of NLP models

๐Ÿ‘จ๐Ÿฝโ€๐Ÿ’ป Deploying

logging inference/prediction of your ML models during their deployment.

๐Ÿ“Š Monitoring

Dashboarding and evaluation of model performance.

NLP Tasks#

Main task categories that we cover within the NLP landscape.



๐Ÿ“•๐Ÿ“— TextClassification

Assigning predefined category labels to texts. This contains sub-tasks like detecting sentiment, semantic similarity, and multi-label classification.

๐Ÿˆด๐Ÿˆฏ๏ธ TokenClassification

Assigning predefined category labels to words and phrases within texts . This contains sub-tasks like Named Entity Recognition (NER) and Part-Of-Speech Tagging (POS).

๐Ÿ‘จ๐Ÿฝ๐Ÿ’ฌ Text2Text

Generating a text based on an input text. This contains sub-tasks like machine translation, and paraphrase generation.


Best practices and methods that can be applied during Machine Learning within our eco-system.



๐Ÿผ Basics

Simple straightforward basics for the oneโ€™s just getting started.

๐Ÿ‘จ๐Ÿฝโ€๐Ÿซ Active Learning

Actively evaluate prediction certainties to determine labels that need to be evaluated for training.

๐Ÿ‘ฎ Weak Supervision

Use rules and functions to obtain initial annotations before manually correcting them.

๐Ÿ”Ž Explainability and bias

understand and explain how a model produced a prediction and be aware of potential systematic errors.

๐Ÿ”ซ Few-shot classification

Model and techniques that perform reasonably well using only a few or zero training samples.

๐Ÿชž Semantic Search

This built-in search uses vectors for text and enables Approximate KNN for semantic search on these vectors.