//=time() ?>
Excited to release my latest project @huggingface with @srush: distilBART for summarization. These models are 1.25x to 4x faster than the corresponding bart-large-* summary models with very little performance sacrificed. In some cases we even gain performance! 👇