Skip to content
This repository has been archived by the owner on Jan 15, 2024. It is now read-only.

[Script] Add BERT SST-2 finetuning result #600

Merged
merged 2 commits into from
Feb 23, 2019
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Jump to
Jump to file
Failed to load files.
Diff view
Diff view
8 changes: 6 additions & 2 deletions scripts/bert/index.rst
Original file line number Diff line number Diff line change
Expand Up @@ -49,8 +49,12 @@ It gets RTE validation accuracy of `70.8% <https://raw.githubusercontent.com/dml

$ MXNET_GPU_MEM_POOL_TYPE=Round GLUE_DIR=glue_data python3 finetune_classifier.py --task_name MNLI --max_len 80 --log_interval 100 --epsilon 1e-8 --gpu

It gets MNLI validation accuracy ,On dev_matched.tsv: 84.6%
On dev_mismatched.tsv: 84.7%. `log <https://github.com/dmlc/web-data/blob/master/gluonnlp/logs/bert/finetuned_mnli.log>`_
It gets MNLI validation accuracy of `84.55% (matched) and 84.66% (mismatched) <https://github.com/dmlc/web-data/blob/master/gluonnlp/logs/bert/finetuned_mnli.log>`_

.. code-block:: console

$GLUE_DIR=glue_data python3 finetune_classifier.py --task_name SST --epochs 4 --batch_size 16 --accumulate 1 --optimizer bertadam --gpu --lr 2e-5 --log_interval 500

It gets SST validation accuracy of `93.0% <https://raw.githubusercontent.com/dmlc/web-data/master/gluonnlp/logs/bert/finetuned_sst.log>`_.

Some other tasks can be modeled with `--task_name` parameter.