madlag/bert-base-uncased-squad1.1-block-sparse-0.13-v1

BERT-base uncased model fine-tuned on SQuAD v1

This model is block sparse: the linear layers contains 12.5% of the original weights.

The model contains 32.1% of the original weights overall.

The training use a modified version of Victor Sanh Movement Pruning method.

That means that with the block-sparse runtime it ran 1.65x faster than an dense networks on the evaluation, at the price of some impact on the accuracy (see below).

This model was fine-tuned from the HuggingFace BERT base uncased checkpoint on SQuAD1.1, and distilled from the equivalent model csarron/bert-base-uncased-squad-v1. This model is case-insensitive: it does not make a difference between english and English.

Pruning details

A side-effect of the block pruning is that some of the attention heads are completely removed: 97 heads were removed on a total of 144 (67.4%).

Here is a detailed view on how the remaining heads are distributed in the network after pruning.

Density plot

Details

Fine-tuning

• Python: 3.8.5

• Machine specs:

Memory: 64 GiB
GPUs: 1 GeForce GTX 3090, with 24GiB memory
GPU driver: 455.23.05, CUDA: 11.1

Results

Pytorch model file size: 342M (original BERT: 438M)

Example Usage

from transformers import pipeline

qa_pipeline = pipeline(
    "question-answering",
    model="madlag/bert-base-uncased-squad1.1-block-sparse-0.13-v1",
    tokenizer="madlag/bert-base-uncased-squad1.1-block-sparse-0.13-v1"
)

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)