README.md

Data format

The finetuning script supports only json as input file format. The input file structure should be the same as standard QA datasets like SQuAD v2.0.

Training & Evaluation

To see list of all available options, do python question_answering.py -h. There are two ways to provide input data files to the script:

• with flag --dataset_dir <path> where <path> points to the directory containing files with prefix train, validation and test.
• with flags --train_file <path> / --train_file <path> / --validation_file <path> / --test_file <path>.

For the following commands, we are going to use the --dataset_dir <path> to provide input files.

Finetuning

For finetuning on single GPU, a minimal example is as follows:

$ python ./question_answering.py \
    --model_name_or_path "csebuetnlp/banglabert" \
    --dataset_dir "sample_inputs/" \
    --output_dir "outputs/" \
    --learning_rate=2e-5 \
    --warmup_ratio 0.1 \
    --gradient_accumulation_steps 2 \
    --weight_decay 0.1 \
    --lr_scheduler_type "linear"  \
    --per_device_train_batch_size=16 \
    --per_device_eval_batch_size=16 \
    --max_seq_length 512 \
    --logging_strategy "epoch" \
    --save_strategy "epoch" \
    --evaluation_strategy "epoch" \
    --num_train_epochs=3 \ 
    --do_train --do_eval

For a detailed example, refer to trainer.sh.

Evaluation

• To calculate metrics on test set / inference on raw data, use the following snippet:

$ python ./question_answering.py \
    --model_name_or_path <path/to/trained/model> \
    --dataset_dir "sample_inputs/" \
    --output_dir "outputs/" \
    --per_device_eval_batch_size=16 \
    --overwrite_output_dir \
    --do_predict

For a detailed example, refer to evaluate.sh.