Software

• The Re-Tuning is a new method for large language models to solve compositional tasks.

• The Social is a new challenge dataset to test the social norms of large language models.

• The STEM is a new challenge dataset to test the STEM skills of neural models.





• The Zero-Shot Robustness is a comprehensive evaluation of the robustness of zero-shot multimodal models.





• The AgentInstruct improves the zero-shot reasoning abilities of large language models on general language understanding tasks.





• The DeepStruct is the state-of-the-art pretrained language model for structure prediction.





• The DeepEx is the state-of-the-art zero-shot information extractor.





• The LASS is the state-of-the-art knowledge graph completion model.





• The PALT is the state-of-the-art parameter-lite transfer learning method for knowledge graph completion.





• The CodeSyntax is a large-scale benchmark for code syntax understanding.





• The Transformer on a Diet is a lightweight Transformer.





• The Language Models with Transformers ensembles RNNs and Transformers.





• The TextHIN performs semantic parsing over knowledge graphs.