Tensor Programs V: Tuning Large Neural Networks via Zero-Shot Hyperparameter Transfer
This addresses the problem of expensive hyperparameter tuning for researchers and practitioners working with billion-parameter models, offering a novel paradigm that reduces computational costs.
The paper tackles the high cost of hyperparameter tuning for large neural networks by introducing muTransfer, a method that leverages Maximal Update Parametrization to enable zero-shot transfer of optimal hyperparameters from smaller to larger models, achieving results like outperforming BERT-large and GPT-3 with significantly reduced tuning costs.
Hyperparameter (HP) tuning in deep learning is an expensive process, prohibitively so for neural networks (NNs) with billions of parameters. We show that, in the recently discovered Maximal Update Parametrization (muP), many optimal HPs remain stable even as model size changes. This leads to a new HP tuning paradigm we call muTransfer: parametrize the target model in muP, tune the HP indirectly on a smaller model, and zero-shot transfer them to the full-sized model, i.e., without directly tuning the latter at all. We verify muTransfer on Transformer and ResNet. For example, 1) by transferring pretraining HPs from a model of 13M parameters, we outperform published numbers of BERT-large (350M parameters), with a total tuning cost equivalent to pretraining BERT-large once; 2) by transferring from 40M parameters, we outperform published numbers of the 6.7B GPT-3 model, with tuning cost only 7% of total pretraining cost. A Pytorch implementation of our technique can be found at github.com/microsoft/mup and installable via `pip install mup`.