Downloads · 30 days
10
1% of all-time downloads
Be-Lo/xtremedistil-l6-h256-uncased-natural-questions-short
xtremedistil-l6-h256-uncased-natural-questions-short is a question answering model from Be-Lo. Use it when the input is a question plus a passage. It is set up for transformers. The card lists the license as mit.
This is a xtremedistil-l6-h256-uncased model, fine-tuned using the NaturalQuestionsShort dataset from the MRQA Shared Task 2019 repository.
Downloads · 30 days
10
1% of all-time downloads
All-time downloads
767
Public
Repo size
102 MB
Likes
0
Public
Click a slice to open those files.
.bin50.8 MB · 98%
From the Hugging Face model README
This is a xtremedistil-l6-h256-uncased model, fine-tuned using the NaturalQuestionsShort dataset from the MRQA Shared Task 2019 repository.
Language model: xtremedistil-l6-h256-uncased
Language: English
Downstream-task: Extractive QA
Training data: NaturalQuestionsShort
Eval data: NaturalQuestionsShort
Infrastructure: Google Colaboratory GPU
batch_size = 16
n_epochs = 2
base_LM_model = "xtremedistil-l6-h256-uncased"
max_seq_len = 512
learning_rate = 3e-5
optimizer = AdamW
weight_decay = 0.01
lr_schedule = Linear
warmup_steps = 0
The model was evaluated on the on the NaturalQuestionsShort dev set from the MRQA Shared Task 2019 repository.
"exact_match": 46.914926768463694,
"f1": 63.863619507647456,
This model can also be found on UKP Square. This website from the UKP lab at the TU Darmstadt is a platform to compare and evaluate cloud-hosted QA models via explainability techniques and behavioral tests.
This model was created by Janik and Ben during the DL4NLP course by Ivan Habernal