Hugging Face

We offer wrappers for generative transformers from Hugging Face’s transformers repository for fine-tuning and evaluating in ParlAI.

GPT2

To use GPT2, run your command with the flag: -m hugging_face/gpt2.

Examples

Talk to GPT2 large in interactive mode, with beam size 10, 3-gram beam blocking, and minimum beam length 25:

parlai interactive -m hugging_face/gpt2 --add-special-tokens False --gpt2-size large --inference beam --beam-size 10 --beam-context-block-ngram 3 --beam-block-ngram 3 --beam-min-length 25

Note: In the above command, we must have the flag --add-special-tokens False if we want to use the model without finetuning it.

Here is example output from the above command:

Enter Your Message: Parrots are
[Gpt2]:  one of the most popular pets in the world. They can be trained to do a variety of tasks, such as fetching objects, opening doors, climbing ladders, and more. They are also very intelligent and can learn new skills very quickly.

Fine-tune GPT2 medium on the ConvAI2 task:

parlai train_model -m hugging_face/gpt2 --add-special-tokens True --add-start-token True --gpt2-size medium -t convai2 -bs 2 -mf <modelfile>

DialoGPT

To use DialoGPT, run your command with the flag: -m hugging_face/dialogpt.

Examples

Talk to DialoGPT large in interactive mode, with beam size 10, 3-gram beam blocking, and minimum beam length 25:

parlai interactive -m hugging_face/dialogpt --add-special-tokens False --gpt2-size large --inference beam --beam-size 10 --beam-context-block-ngram 3 --beam-block-ngram 3 --beam-min-length 25

Note: In the above command, we must have the flag --add-special-tokens False if we want to use the model without finetuning it.

Here is example output from the above command:

Enter Your Message: What do you think of parrots?
[Dialogpt]:  I love parrots. They are the best. I love them so much. I wish I had a pet parrot.

Fine-tune DialoGPT medium on the ConvAI2 task:

parlai train_model -m hugging_face/dialogpt --add-special-tokens True --delimiter '\n' --add-start-token True --gpt2-size medium -t convai2 -bs 2 -mf <modelfile>

Note: In the above command, we change the default delimiter from --delimiter '<|endoftext|>', as a personal choice.

T5

“Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer”

See https://arxiv.org/abs/1910.10683.

Implementation

The T5 model in ParlAI is based on the T5ForConditionalGeneration provided by the HuggingFace Transformers library. The model can be instantiated with any of the provided architectures there:

  • t5-small: 60 million parameters

  • t5-base: 220 million parameters

  • t5-large: 770 million parameters

  • t5-3b: 3 billion parameters

  • t5-11b: 11 billion parameters

Model Parallel: HuggingFace has implemented model parallel for T5, however it is an experimental feature, so proceed at your own risk; you can use model parallel by simply specifying --t5-model-parallel.

Basic Examples

Train t5 large on convai2.

parlai train_model -m hugging_face/t5 -mf /tmp/model_file -t convai2 -bs 24 --fp16 true -eps 1 -lr 1e-5 --optimizer adam --t5-model-arch t5-large