amanotaiga/definition_modeling_project

Definition_Modeling_Project

Download Dataset

wget http://www.tkl.iis.u-tokyo.ac.jp/~ishiwatari/naacl_data.zip
unzip naacl_data.zip

Requirements

• pytorch-lightning (0.9.0)
• transformers (3.1.0)
• pytorch (1.6.0)

Run Experiments

The dataset can be specified in script (wordent, oxford, slang, wiki).

./finetune_t5_experiment.sh

definition_modeling_project's Issues

Special character in finetune_t5_experiment.sh

In finetune_t5_experiment.sh:

export CUDA_VISIBLE_DEVICES=0
python finetune.py --model_name_or_path $MODEL --data_dir $DATA_DIR --eval_batch_size $TEST_BATCH_SIZE --output_dir $T5_OUTPUT_DIR --max_source_length $MAX_LEN --max_target_length $MAX_LEN --gpus 1 --do_predict --option "forward" &

export CUDA_VISIBLE_DEVICES=2
python finetune.py --model_name_or_path $MODEL --data_dir $DATA_DIR --eval_batch_size $TEST_BATCH_SIZE --output_dir $T5_SPECIFIC_OUTPUT_DIR --max_source_length $MAX_LEN --max_target_length $MAX_LEN --do_predict --gpus 1 --option "t5_specific" &

export CUDA_VISIBLE_DEVICES=1
python finetune.py --model_name_or_path　$MODEL --data_dir $DATA_DIR --eval_batch_size $TEST_BATCH_SIZE --output_dir $T5_GENERAL_OUTPUT_DIR --max_source_length $MAX_LEN --max_target_length $MAX_LEN --do_predict --gpus 1 --option "t5_general" &
wait

There is a special character between --model_name_or_path and $MODEL instead of space in the 3rd command, causing this command to not work properly.