Edit on GitHub 

model = AutoAdapterModel.from_pretrained("facebook/bart-large")
config = AdapterConfig.load("lohfink-rossi-leaveout", non_linearity="relu", reduction_factor=16)
model.load_adapter("qa/squad2@lohfink-rossi", config=config)

Description

Adapter for bart-large using a custom architecture (Lohfink-Rossi-Leaveout) trained on the SQuAD 2.0 dataset for 15 epochs with a Cosine with Restarts learning rate scheduler ans learning rate 0.001.

Properties

Pre-trained model
facebook/bart-large
Adapter type
Task
Prediction Head
  Yes
Task
Question Answering
Dataset
SQuAD 2.0

Architecture

Name
lohfink-rossi-leaveout
Non-linearity
relu
Reduction factor
16
{
  "ln_after": false,
  "ln_before": false,
  "mh_adapter": true,
  "output_adapter": true,
  "adapter_residual_before_ln": true,
  "non_linearity": "relu",
  "original_ln_after": true,
  "original_ln_before": true,
  "reduction_factor": 16,
  "residual_before_ln": true,
  "leave_out": [
    0,
    1,
    2,
    3,
    11,
    12,
    13,
    14
  ]
}

Author

  Name
Till Lohfink & Maria Rossi (Contributed equally.)
  E-Mail
tlohfink3@gatech.edu;mrossi7@gatech.edu

Versions

Identifier Comment Score Download
1 DEFAULT 84.93

Citations

Task
@misc{rajpurkar2018know,
  title={Know What You Don't Know: Unanswerable Questions for SQuAD},
  author={Pranav Rajpurkar and Robin Jia and Percy Liang},
  year={2018},
  eprint={1806.03822},
  archivePrefix={arXiv},
  primaryClass={cs.CL}
}