Supervised pre training! Another exploration of text generation!

Download address ：

https://arxiv.org/abs/2206.12131

The code for this article , Model , The intermediate results are open source

GitHub Home page , One stop fine tuning 、 Profiling code , In the middle

https://github.com/RUCAIBox/MVP

MVP Model

https://huggingface.co/RUCAIBox/mvp

MVP variant

https://huggingface.co/models?filter=mvp

documentation

https://huggingface.co/docs/transformers/main/en/model_doc/mvp

This paper proposes a multi task supervised pre training model for text generation MVP. Different from the previous none （ since ） Supervised pre training model , In this paper, we collect labeled generated data sets to pre train the model MVP, At the same time, continuous prompts are further trained for different tasks to stimulate the ability of the model in specific tasks . Experiments show the effectiveness and versatility of our model , Beyond the baseline model of generating tasks BART, stay 12 Data sets were obtained SOTA effect .