CoolFace
Datasetpublic

wojemann/snli_boolq_train

Dataset Card for "snli_boolq_train" This dataset contains validation and training data from both boolq and snli snli_boolq_train is a mixed dataset containing both boolq (https://huggingface.co/datasets/boolq) and the "entail" and "contradict" samples from snli (https://huggingface.co/datasets/snli). The selection of this data was to increase the finetuning dataset amount for large language models with uninitialized weights. neutral statements were… See the full description on the dataset page: https://huggingface.co/datasets/wojemann/snli_boolq_train.

sourceHugging Faceupdated 3y agoView on Hugging Face
1likes53downloads
Dataset Card

Dataset Card for "snliboolqtrain"

This dataset contains validation and training data from both boolq and snli

snliboolqtrain is a mixed dataset containing both boolq (https://huggingface.co/datasets/boolq) and the "entail" and "contradict" samples from snli (https://huggingface.co/datasets/snli). The selection of this data was to increase the finetuning dataset amount for large language models with uninitialized weights. neutral statements were removed to ensure the compatibility of the two datasets. The original boolq paper demonstrated that minimal loss to boolq accuracy could be achieved by removing neutral statements from the snli dataset (https://aclanthology.org/N19-1300.pdf)