QA with bert-base-uncased has been completed. The dependencies needed are all included in the setup.sh, using source setup.sh to config the environment.

There are two versions of QA, question_answering.py follows the same architecture with the original classification tasks (while sharing same config files), its Loss can decrease while EM and F1 do not. Therefore, another version (question_answering_new.py & qa_args.py & qa_utils.py) are created to replace the first version which has some errors that didn't be found, you should use the second to train and test (I keep the first for the error check). The training and validation results have been included in save/ (but no model as it's a large file), it has been trained on SQuAD v1.1, you can train it on other data if the format is same as SQuAD. The more option for training and validation can be found in qa_args.py. Meanwhile, you can use tensorboard to check training results (the events file is in save/qa_bert-00/):

# First clone the repo

cd minbert-default-final-project/
tensorboard --logdir save/qa_bert-00/

Notice: I wrote this just for practicing, and I tested it on debian 10, if you config the env as setup.sh, it should work fine. Any constructive or helpful advice and bug fixes are appreciated.

The Basic classifier task model has been completed, but not train or test yet, the main task focused on is question answering part with SQuAD dataset.

This is the starting code for the default final project for the Stanford CS 224N class. You can find the handout here

In this project, you will implement some important components of the BERT model to better understanding its architecture. You will then use the embeddings produced by your BERT model on three downstream tasks: sentiment classification, paraphrase detection and semantic similarity.

After finishing the BERT implementation, you will have a simple model that simultaneously performs the three tasks. You will then implement extensions to improve on top of this baseline.

• Follow setup.sh to properly setup a conda environment and install dependencies.
• There is a detailed description of the code structure in STRUCTURE.md, including a description of which parts you will need to implement.
• You are only allowed to use libraries that are installed by setup.sh, external libraries that give you other pre-trained models or embeddings are not allowed (e.g., transformers).

Please refer to the handout for a through description of the project and its parts.

The BERT implementation part of the project was adapted from the "minbert" assignment developed at Carnegie Mellon University's CS11-711 Advanced NLP, created by Shuyan Zhou, Zhengbao Jiang, Ritam Dutt, Brendon Boldt, Aditya Veerubhotla, and Graham Neubig.

Parts of the code are from the transformers library (Apache License 2.0).