Parameter-Efficient Transformer Embeddings

TL;DR

This paper introduces a parameter-efficient method for transformer embeddings that uses deterministic Fourier-based token vectors and a lightweight MLP, reducing parameters and training time while maintaining competitive NLP performance.

Contribution

It presents a novel embedding approach combining Fourier expansion and a small MLP, significantly decreasing model size and training time without sacrificing accuracy.

Findings

Achieves competitive NLP performance with fewer parameters

Trains faster and operates without dropout

Demonstrates potential for scalable, memory-efficient language models

Abstract

Embedding layers in transformer-based NLP models typically account for the largest share of model parameters, scaling with vocabulary size but not yielding performance gains proportional to scale. We propose an alternative approach in which token embedding vectors are first generated deterministically, directly from the token IDs using a Fourier expansion of their normalized values, followed by a lightweight multilayer perceptron (MLP) that captures higher-order interactions. We train standard transformers and our architecture on natural language inference tasks (SNLI and MNLI), and evaluate zero-shot performance on sentence textual similarity (STS-B). Our results demonstrate that the proposed method achieves competitive performance using significantly fewer parameters, trains faster, and operates effectively without the need for dropout. This proof-of-concept study highlights the potential for scalable, memory-efficient language models and motivates further large-scale experimentation based on our findings.