McGill-NLP/polytropon

Polytropon: Combining Modular Skills in Multitask Learning

Updates | Installation | Usage | Cite | Paper

Updates

23/05/03: New repository! The current repository is outdated. We recommend to use instead the full implementation of Polytropon at https://github.com/microsoft/mttl

Installation

pip install git+https://github.com/McGill-NLP/polytropon

Otherwise, if you wish to clone the repo:

git clone https://github.com/McGill-NLP/polytropon.git
cd polytropon
pip install -e .

Usage

from polytropon import SkilledMixin

# load any pretrained model from transformers
from transformers import T5ForConditionalGeneration
model = T5ForConditionalGeneration.from_pretrained("t5-small")

# merge it with polytropon
model = SkilledMixin(
  model,
  n_tasks,
  n_skills,
)

Cite

@misc{ponti2022combining,
      title={Combining Modular Skills in Multitask Learning},
      author={Edoardo M. Ponti and Alessandro Sordoni and Yoshua Bengio and Siva Reddy},
      year={2022},
      eprint={2202.13914},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}