How to load an .mdl file in Python?

Question

I plan to fine tune a GPT transformer model with a custom dataset, specifically with an EmpatheticDialogues dataset, for my chatbot. The repository provides an .mdl file to their project. How will I able to load these .mdl files in my ipynb file? Is it right that I need to load any of their .mdl file to fine tune my GPT transformer model, called DialoGPT?

I already tried loading in just the .csv dataset of the said repository with the guidance of this Google Colab Notebook. However, I just get random symbols when testing the model after training it with the said dataset.

Answer 1

Evaluation with .mdl file is given clearly on their repo:

# P@1,100
python retrieval_train.py \
--batch-size 256 \
--bert-dim 300 \
--cuda \
--dataset-name empchat \
--dict-max-words 250000 \
--display-iter 100 \
--embeddings None \
--empchat-folder ${EMPATHETIC_DIALOGUES_DATA_FOLDER} \
--max-hist-len 4 \
--model bert \
--model-dir ${EVAL_SAVE_FOLDER} \
--model-name model \
--optimizer adamax \
--pretrained ${TRAIN_SAVE_FOLDER}/model.mdl \
--reactonly

# BLEU (EmpatheticDialogues context/candidates)
python retrieval_eval_bleu.py \
--bleu-dict ${PATH_TO_MODEL_WITH_TRANSFORMER_DICT} \
--empchat-cands \
--empchat-folder ${EMPATHETIC_DIALOGUES_DATA_FOLDER} \
--max-hist-len 4 \
--model ${TRAIN_SAVE_FOLDER}/model.mdl \
--name model \
--output-folder ${EVAL_SAVE_FOLDER} \
--reactonly \
--task empchat

Yes, you can use any of the 6 models.

wget https://dl.fbaipublicfiles.com/parlai/empatheticdialogues/models/normal_transformer_pretrained.mdl  # Normal Transformer, pretrained
wget https://dl.fbaipublicfiles.com/parlai/empatheticdialogues/models/normal_transformer_finetuned.mdl  # Normal Transformer, fine-tuned
wget https://dl.fbaipublicfiles.com/parlai/empatheticdialogues/models/bert_pretrained.mdl  # BERT, pretrained
wget https://dl.fbaipublicfiles.com/parlai/empatheticdialogues/models/bert_finetuned.mdl  # BERT, fine-tuned
wget https://dl.fbaipublicfiles.com/parlai/empatheticdialogues/models/bert_finetuned_emoprepend1.mdl  # BERT, fine-tuned (EmoPrepend-1)
wget https://dl.fbaipublicfiles.com/parlai/empatheticdialogues/models/fasttext_empathetic_dialogues.mdl  # fastText classifier used for EmoPrepend-1