File Formats Supported by Different Tasks

Here is the list of the file formats supported by different tasks.

For dataset files, if your datasets have been supported by datalab, you fortunately don't need to prepare the dataset. Otherwise, you can upload your custom datasets in the supported formats.

You may refer to the example files for more specific information about the formats, or click on the task link for more explanation.

To upload custom features and analysis, please refer to this instruction.

Task	File Type	File Format	Example File
conditional generation (machine translation/summarization)	dataset	TSV	cnndm_mini-dataset.tsv
		JSON	conala-dataset.json
	output	JSON	conala-baseline-output.json
		TXT	cnndm_mini-bart-output.txt
text classification	dataset	TSV	sst2-dataset.tsv
		JSON	text-classification-dataset.json
	output	JSON	text-classification-output.json
		TXT	sst2-lstm-output.txt
sequence labeling (NER/word segmentation/chunking)	dataset	CoNLL	conll2003-dataset.conll
	output	CoNLL	conll2003-elmo-output.conll
		JSON
cloze multiple choice	dataset	JSON
	output	JSON
cloze generative	dataset	JSON
	output	JSON
QA (extractive)	dataset	JSON	squad_mini-dataset.json
	output	JSON	squad_mini-example-output.json
QA (MCQ)	dataset	JSON	fig_qa-dataset.json
	output	JSON	fig_qa-bert-output.json
QA (open domain)	dataset	JSON
	output	TXT	test.dpr.nq.txt
aspect-based sentiment analysis	dataset	TSV	absa-dataset.tsv
		JSON
	output	JSON	absa-example-output-confidence.json
		TXT	absa-example-output.txt
grammatical error correction	dataset	JSON
	output	JSON
text pair classification	dataset	TSV	snli-dataset.tsv
		JSON
	output	JSON
		TXT	snli-roberta-output.txt
knowledge graph link tail prediction	dataset	JSON
	output	JSON
language modeling	dataset	JSON
		TXT
	output	JSON
		TXT
tabular classification	dataset	JSON	sst2-tabclass-dataset.json
	output	JSON
		TXT
tabular regression	dataset	JSON	sst2-tabreg-dataset.json
	output	JSON
		TXT	sst2-tabreg-lstm-output.txt