Subword-ULM transformer (DeepSPIN-3; soft-attention, 1-5 entmax)

Reported on 1 benchmark across 1 task

Note: results are matched by exact model name. Different papers may use the same name for different model variants.