HW1

Reproduce my result

#Environment

pip install -r requirements.txt

##Download Download training, validation, testing data, as well as multiple choice model and question answering model.

bash ./download.sh

##Multiple Choice

python run_multiple_choice.py \
--context_data <context.json> \
--train_data <train.json> \
--valid_data <valid.json> \
--test_data <test.json> \
--max_seq_length 512 \
--gradient_accumulation_steps 8 \
--model_name_or_path bert-base-chinese \
--learning_rate 2e-5 \
--output_dir <output directory> \
--per_device_train_batch_size 8

-model_name_or_path: Path to pretrained model.

-output_dir: Path to directory which saves the model outputs.

-context_data: Path to context.json.

-train_data: Path to train.json.

-valid_data: Path to valid.json.

-test_data: Path to test.json.

##Question Answering

python run_question_answering.py \
--context_data <context.json> \
--train_data <train.json> \
--valid_data <valid.json> \
--test_data <test.json> \
--max_seq_length 512 \
--gradient_accumulation_steps 8 \
--model_name_or_path hfl/chinese-roberta-wwm-ext-large \
--learning_rate 2e-5 \
--output_dir <output directory> \
--per_device_train_batch_size 8

-model_name_or_path: Path to pretrained model.

-output_dir: Path to directory which saves the model outputs.

-context_data: Path to context.json.

-train_data: Path to train.json.

-valid_data: Path to valid.json.

-test_data: Path to test.json.

##Reproduce my result

bash ./download.sh
bash ./run.sh /path/to/context.json /path/to/test.json /path/to/pred/prediction.csv