Matryoshka Model Learning for Improved Elastic Student Models
This addresses the resource-intensive process of model development for industry applications with evolving constraints, though it appears incremental as it builds on existing teacher-student distillation methods.
The paper tackles the problem of efficiently developing multiple ML models to meet varying serving constraints by proposing MatTA, a framework that trains multiple accurate Student models from a single training run. The method demonstrated 20% improvement on a key metric in production A/B tests and achieved over 24% relative improvement on SAT Math and over 10% on LAMBADA with GPT-2 Medium.
Industry-grade ML models are carefully designed to meet rapidly evolving serving constraints, which requires significant resources for model development. In this paper, we propose MatTA, a framework for training multiple accurate Student models using a novel Teacher-TA-Student recipe. TA models are larger versions of the Student models with higher capacity, and thus allow Student models to better relate to the Teacher model and also bring in more domain-specific expertise. Furthermore, multiple accurate Student models can be extracted from the TA model. Therefore, despite only one training run, our methodology provides multiple servable options to trade off accuracy for lower serving cost. We demonstrate the proposed method, MatTA, on proprietary datasets and models. Its practical efficacy is underscored by live A/B tests within a production ML system, demonstrating 20% improvement on a key metric. We also demonstrate our method on GPT-2 Medium, a public model, and achieve relative improvements of over 24% on SAT Math and over 10% on the LAMBADA benchmark.