Edit on GitHub

model = AutoAdapterModel.from_pretrained("roberta-base")
config = AdapterConfig.load("houlsby")
model.load_adapter("nli/qnli@ukp", config=config)

Description

QNLI adapter (with head) trained using the `run_glue.py` script with an extension that retains the best checkpoint (out of 15 epochs).

Properties

Pre-trained model
roberta-base
Adapter type
Prediction Head
  Yes
Task
Natural Language Inference
Dataset

Architecture

Name
houlsby
Non-linearity
swish
Reduction factor
16
{
  "ln_after": false,
  "ln_before": false,
  "mh_adapter": true,
  "output_adapter": true,
  "adapter_residual_before_ln": false,
  "non_linearity": null,
  "original_ln_after": true,
  "original_ln_before": false,
  "reduction_factor": null,
  "residual_before_ln": true
}

Author

  Name
Andreas Rücklé
  GitHub
  Twitter

Versions

Identifier Comment Score Download
1 DEFAULT Achieves 93.33 Accuracy on the QNLI devset

Citations

Adapter
@article{pfeiffer2020AdapterHub,
    title={AdapterHub: A Framework for Adapting Transformers},
    author={Jonas Pfeiffer,
            Andreas R\"uckl\'{e},
            Clifton Poth,
            Aishwarya Kamath,
            Ivan Vuli\'{c},
            Sebastian Ruder,
            Kyunghyun Cho,
            Iryna Gurevych},
    journal={ArXiv},
    year={2020}
}
Architecture
@misc{houlsby2019parameterefficient,
  title={Parameter-Efficient Transfer Learning for NLP},
  author={Neil Houlsby and Andrei Giurgiu and Stanislaw Jastrzebski and Bruna Morrone and Quentin de Laroussilhe and Andrea Gesmundo and Mona Attariyan and Sylvain Gelly},
  year={2019},
  eprint={1902.00751},
  archivePrefix={arXiv},
  primaryClass={cs.LG}
}
Task
@article{wang2018glue,
  title={Glue: A multi-task benchmark and analysis platform for natural language understanding},
  author={Wang, Alex and Singh, Amanpreet and Michael, Julian and Hill, Felix and Levy, Omer and Bowman, Samuel R},
  journal={arXiv preprint arXiv:1804.07461},
  year={2018}
}