Tasksource

tasksource: 500+ dataset harmonization preprocessings for effortless extreme multi-task learning and evaluation

Huggingface Datasets is an excellent library, but it lacks standardization, and datasets often require preprocessing work to be used interchangeably. tasksource streamlines interchangeable datasets usage to scale evaluation or multi-task learning.

Each dataset is standardized to a MultipleChoice, Classification, or TokenClassification template with canonical fields. We focus on discriminative tasks (= with negative examples or classes) and do not yet support generation tasks as they are addressed by promptsource. All implemented preprocessings are in tasks.py or tasks.md. A preprocessing is a function that accepts a dataset and returns the standardized dataset. Preprocessing code is concise and human-readable.

Installation and usage:

pip install tasksource

from tasksource import list_tasks, load_task
df = list_tasks() # takes some time

for id in df[df.task_type=="MultipleChoice"].id:
    dataset = load_task(id) # all yielded datasets can be used interchangeably

Browse the 500+ curated tasks in tasks.md (200+ MultipleChoice tasks, 200+ Classification tasks), and feel free to request a new task. Datasets are downloaded to $HF_DATASETS_CACHE (like any Hugging Face dataset), so ensure you have more than 100GB of space available.

Pretrained model:

Text encoder pretrained on tasksource reached state-of-the-art results: /deberta-v3-base-tasksource-nli

Tasksource pretraining is notably helpful for RLHF reward modeling.

tasksource-instruct

The repo also contains some recasting code that was used to convert tasksource datasets to instructions format, providing one of the richest instruction-tuning dataset: /tasksource-instruct-v0

Write and use custom preprocessings

from tasksource import MultipleChoice, concatenate_dataset_dict

codah = MultipleChoice('question_propmt',choices_list='candidate_answers',
    labels='correct_answer_idx',
    dataset_name='codah', config_name='codah')
    
winogrande = MultipleChoice('sentence',['option1','option2'],'answer',
    dataset_name='winogrande',config_name='winogrande_xl',
    splits=['train','validation',None]) # these test labels are not usable
    
tasks = [winogrande.load(), codah.load()]) #  Aligned datasets (same columns) can be used interchangably  

Contact and citation

For help integrating tasksource into your experiments, please contact [email protected].

For more details, refer to this article:

@article{sileo2023tasksource,
  title={tasksource: Structured Dataset Preprocessing Annotations for Frictionless Extreme Multi-Task Learning and Evaluation},
  author={Sileo, Damien},
  url= {https://arxiv.org/abs/2301.05948},
  journal={arXiv preprint arXiv:2301.05948},
  year={2023}
}