BatGPT: A Bidirectional Autoregessive Talker from Generative Pre-trained Transformer
Zuchao Li, Shitou Zhang, Hai Zhao, Yifei Yang, Dongjie Yang
TL;DR
BatGPT introduces a bidirectional autoregressive LLM trained with a novel parameter expansion strategy and a reinforcement learning from human and AI feedback pipeline. By combining bidirectional pretraining, instruction tuning, RLHF, and function-preserving width expansion with progressive depth growth, BatGPT aims to alleviate memory limitations and hallucinations while enabling scalable, aligned generation. Empirical results on CMMLU and C-Eval demonstrate competitive performance on Chinese-centric tasks and cross-domain capabilities, supporting BatGPT as a viable option for diverse NLP applications. The approach outlines a scalable path for future model growth with alignment-aware training and efficiency-focused expansion techniques.
Abstract
BatGPT is a large-scale language model designed and trained jointly by Wuhan University and Shanghai Jiao Tong University. It is capable of generating highly natural and fluent text in response to various types of input, including text prompts, images, and audio. In the modeling level, we employ a bidirectional autoregressive architecture that allows the model to efficiently capture the complex dependencies of natural language, making it highly effective in tasks such as language generation, dialog systems, and question answering. Moreover, the bidirectional autoregressive modeling not only operates from left to right but also from right to left, effectively reducing fixed memory effects and alleviating model hallucinations. In the training aspect, we propose a novel parameter expansion method for leveraging the pre-training of smaller models and employ reinforcement learning from both AI and human feedback, aimed at improving the model's alignment performance. Overall, these approaches significantly improve the effectiveness of BatGPT, and the model can be utilized for a wide range of natural language applications.