Comparing standard CrossEntropy based training on the GLUE tasks to a LTN based approach using the SatAgg objective.
Why not?
MNLI, QNLI, SST-2, WNLI, RTE, CoLA
TOOD
TODO
TODO
Comparing standard CrossEntropy based training on the GLUE tasks to a LTN based approach using the SatAgg objective.
Why not?
MNLI, QNLI, SST-2, WNLI, RTE, CoLA
TOOD
TODO
TODO