List of packages we used and the version we tested the model on
python == 3.7.6
pytorch_transformers == 1.2.0
pytorch_pretrained_bert == 0.6.2
tensorboardX == 1.9
matplotlib == 3.1.1
nltk == 3.4.5
numpy == 1.16.4
pandas == 0.25.3
scipy == 1.3.2
seaborn == 0.9.0
scikit-image == 0.15.0
torch == 1.5.0
tqdm == 4.42.1Pretrained models are also available here:
Please refer to OpenNMT website for OpenNMT model here
Pretraining the above models with commands:
GPT2: --train_data_file YOUR_FILE --output_dir gpt2_model/ --model_type=gpt2 --model_name_or_path=gpt2 --do_train --do_eval --eval_data_file YOUR_FILE --block_size 128 --overwrite_output_dir --num_train_epochs 30
OpenNMT:-data YOUR_PROCESSED_DATA -save_model YOUR_PATH -layers 6 -rnn_size 512 -word_vec_size 512 -transformer_ff 2048 -heads 8 -encoder_type transformer -decoder_type transformer -position_encoding -train_steps 100000 -max_generator_batches 2 -dropout 0.1 -batch_size 16 -batch_type tokens -normalization tokens -accum_count 2 -optim adam -adam_beta2 0.998 -decay_method noam -warmup_steps 8000 -learning_rate 2 -max_grad_norm 0 -param_init 0 -param_init_glorot -label_smoothing 0.1 -valid_steps 2000 -save_checkpoint_steps 2000 -world_size 1 -gpu_ranks 0
Your input file should follow the format
'''
No header, answers are generated, and for diversity,
each answer is a group of sentences, separated by newlines
| id 1 | question 1 | answer 1 |
| id 2 | question 2 | answer 2 |
| id 3 | question 3 | answer 3 |
...
'''Run with arguments:
--pretrained-model-path YOUR_PATH --metric [context | fluency | diversity | logic_consistency] --file-path YOUR_FILE.csv
and this will append the score as another column of your .csv file