BERT-base uncased model fine-tuned on SQuAD v1

This model is block-sparse.

That means that with the right runtime it can run roughly 3x faster than an dense network, with 25% of the original weights.

This of course has some impact on the accuracy (see below).

It uses a modified version of Victor Sanh Movement Pruning method.

This model was fine-tuned from the HuggingFace BERT base uncased checkpoint on SQuAD1.1. This model is case-insensitive: it does not make a difference between english and English.

Details

Fine-tuning

• Python: 3.8.5

• Machine specs:

  CPU: Intel(R) Core(TM) i7-6700K CPU

  Memory: 64 GiB

  GPUs: 1 GeForce GTX 3090, with 24GiB memory

  GPU driver: 455.23.05, CUDA: 11.1

Results

Model size: 418M

Note that the above results didn't involve any hyperparameter search.

Example Usage

from transformers import pipeline

qa_pipeline = pipeline(
    "question-answering",
    model="madlag/bert-base-uncased-squad-v1-sparse0.25",
    tokenizer="madlag/bert-base-uncased-squad-v1-sparse0.25"
)

predictions = qa_pipeline({
    'context': "Frédéric François Chopin, born Fryderyk Franciszek Chopin (1 March 1810 – 17 October 1849), was a Polish composer and virtuoso pianist of the Romantic era who wrote primarily for solo piano.",
    'question': "Who is Frederic Chopin?",
})

print(predictions)