Weight Squeezing: Reparameterization for Knowledge Transfer and Model Compression
This work addresses model efficiency for NLP practitioners by enabling faster and more effective compression of large pre-trained models, though it is incremental as it builds on existing knowledge transfer techniques.
The paper tackles the problem of knowledge transfer and model compression by introducing Weight Squeezing, a method that learns a mapping from teacher model weights to smaller student model weights, applied to BERT models on the GLUE benchmark, resulting in better performance and faster training compared to other methods.
In this work, we present a novel approach for simultaneous knowledge transfer and model compression called Weight Squeezing. With this method, we perform knowledge transfer from a teacher model by learning the mapping from its weights to smaller student model weights. We applied Weight Squeezing to a pre-trained text classification model based on BERT-Medium model and compared our method to various other knowledge transfer and model compression methods on GLUE multitask benchmark. We observed that our approach produces better results while being significantly faster than other methods for training student models. We also proposed a variant of Weight Squeezing called Gated Weight Squeezing, for which we combined fine-tuning of BERT-Medium model and learning mapping from BERT-Base weights. We showed that fine-tuning with Gated Weight Squeezing outperforms plain fine-tuning of BERT-Medium model as well as other concurrent SoTA approaches while much being easier to implement.