Please provide the same data split as mentioned in the paper.
#4
by
saerom-choi
- opened
I'm curious about how the split in Table 1 maps to the data provided by Hugging Face. Could you share the id and ood splits?
Hi @saerom-choi ,
The only splits we have available are the training split and a ~1M datapoint subsplit of the the validation 4.5M validation split - we used the smaller 1M subsplit to train our models for faster training. We are not making the test splits available at this time.
Let us know if you have other questions
lbluque
changed discussion status to
closed