Official code for the paper titled Augmenting Transformers with Recursively Composed Multi-grained Representations

In this work, we successfully combine a composition model with bi-directional Transformers and make them jointly pre-trainable.

self-interpretable classification

Official code for the paper A Multi-Grained Self-Interpretable Symbolic-Neural Model For Single/Multi-Labeled Text Classification

Fast-R2D2

The paper version and the corresponding model pretrained on wiki-103.

r2d2

The code for paper "R2D2: Recursive Transformer based on Differentiable Tree for Interpretable Hierarchical Language Modeling"