README.md

# BERT-mini model finetuned with M-FAC

This model is finetuned on MNLI dataset with state-of-the-art second-order optimizer M-FAC.
Check NeurIPS 2021 paper for more details on M-FAC: [https://arxiv.org/pdf/2107.03356.pdf](https://arxiv.org/pdf/2107.03356.pdf).

## Finetuning setup

For fair comparison against default Adam baseline, we finetune the model in the same framework as described here [https://github.com/huggingface/transformers/tree/master/examples/pytorch/text-classification](https://github.com/huggingface/transformers/tree/master/examples/pytorch/text-classification) and just swap Adam optimizer with M-FAC.
Hyperparameters used by M-FAC optimizer:

```bash
learning rate = 1e-4
number of gradients = 1024
dampening = 1e-6
```

## Results

We share the best model out of 5 runs with the following score on MNLI validation set:

```bash
matched_accuracy = 75.13
mismatched_accuracy = 75.93
```

Mean and standard deviation for 5 runs on MNLI validation set:

|       | Matched Accuracy | Mismatched Accuracy |
|:-----:|:----------------:|:-------------------:|
| Adam  | 73.30 ± 0.20     | 74.85 ± 0.09        |
| M-FAC | 74.59 ± 0.41     | 75.95 ± 0.14        |

Results can be reproduced by adding M-FAC optimizer code in [https://github.com/huggingface/transformers/blob/master/examples/pytorch/text-classification/run_glue.py](https://github.com/huggingface/transformers/blob/master/examples/pytorch/text-classification/run_glue.py) and running the following bash script:

```bash
CUDA_VISIBLE_DEVICES=0 python run_glue.py \
  --seed 8276 \
  --model_name_or_path prajjwal1/bert-mini \
  --task_name mnli \
  --do_train \
  --do_eval \
  --max_seq_length 128 \
  --per_device_train_batch_size 32 \
  --learning_rate 1e-4 \
  --num_train_epochs 5 \
  --output_dir out_dir/ \
  --optim MFAC \
  --optim_args '{"lr": 1e-4, "num_grads": 1024, "damp": 1e-6}'
```

We believe these results could be improved with modest tuning of hyperparameters: `per_device_train_batch_size`, `learning_rate`, `num_train_epochs`, `num_grads` and `damp`. For the sake of fair comparison  and a robust default setup we use the same hyperparameters across all models (`bert-tiny`, `bert-mini`) and all datasets (SQuAD version 2 and GLUE).

## BibTeX entry and citation info

```bibtex
@article{DBLP:journals/corr/abs-2107-03356,
  author    = {Elias Frantar and
               Eldar Kurtic and
               Dan Alistarh},
  title     = {Efficient Matrix-Free Approximations of Second-Order Information,
               with Applications to Pruning and Optimization},
  journal   = {CoRR},
  volume    = {abs/2107.03356},
  year      = {2021},
  url       = {https://arxiv.org/abs/2107.03356},
  eprinttype = {arXiv},
  eprint    = {2107.03356},
  timestamp = {Tue, 20 Jul 2021 15:08:33 +0200},
  biburl    = {https://dblp.org/rec/journals/corr/abs-2107-03356.bib},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}

```