Examples

Transformers examples are the starting point for many users of the transformers library. Hence, they are the core feature of model-hub’s support for transformers library. In fact, each transformer example corresponding to a core transformer task has an associated task in model-hub that is guaranteed to work with Determined and verified for correctness. See the table below for a summary of the model-hub transformers examples:

Task	Dataset	Filename
language-modeling	WikiText-2	language-modeling.tgz
multiple-choice	SWAG	multiple-choice.tgz
question-answering	SQuAD and SQuAD version 2	question-answering.tgz
text-classification	GLUE and XNLI	text-classification.tgz
token-classification	CoNLL-2003	token-classification.tgz
summarization	CNN/DailyMail and XSum	Coming Soon
translation	WMT-16	Coming Soon

Each of the Determined trials above subclasses the BaseTransformerTrial and makes use of helper functions in model_hub.huggingface to drastically reduce the code needed to run the example. See the api for more details.