GPT Dataset

Data prefix /data/arxiv/tokenized_text_document
Dataset name test
Number of samples 102400
Number of epochs 13
Sequence length 4096
Random seed 1234
Split 969, 30, 1