Fetching the paper…

Unlocking Compositional Generalization in Pre-trained Models Using Intermediate Representations · Around