Layer-Parallel Training for Transformers
This addresses scalability issues for large foundational models by enabling layer-parallel training, though it is incremental as it builds on existing neural ODE and parallel-in-time methods.
The paper tackles the challenge of parallelizing transformer training across layers using a multilevel parallel-in-time algorithm based on neural ODEs, achieving parallel acceleration for models like BERT and GPT2 while maintaining accuracy comparable to serial training during fine-tuning.
We present a new training methodology for transformers using a multilevel, layer-parallel approach. Through a neural ODE formulation of transformers, our application of a multilevel parallel-in-time algorithm for the forward and backpropagation phases of training achieves parallel acceleration over the layer dimension. This dramatically enhances parallel scalability as the network depth increases, which is particularly useful for increasingly large foundational models. However, achieving this introduces errors that cause systematic bias in the gradients, which in turn reduces convergence when closer to the minima. We develop an algorithm to detect this critical transition and either switch to serial training or systematically increase the accuracy of layer-parallel training. Results, including BERT, GPT2, ViT, and machine translation architectures, demonstrate parallel-acceleration as well as accuracy commensurate with serial pre-training while fine-tuning is unaffected.