TensorX
返回文献探索

Paper · arXiv 2502.01637

Scaling Embedding Layers in Language Models

Da Yu, Edith Cohen, Badih Ghazi, Yangsibo Huang, Pritish Kamath, Ravi Kumar, Daogao Liu, Chiyuan Zhang

24 upvotesFebruary 3, 2025arXiv 预印本
AI 摘要

SCONE extends language model performance by using contextualized n-gram embeddings, precomputed and stored off-accelerator, to minimize inference costs while maintaining fixed FLOPS.

n-gram embeddingscontextualized representationoff-accelerator memoryinference-time FLOPS

Abstract

We propose SCONE (Scalable, Contextualized, Offloaded, N-gram Embedding), a method for extending input embedding layers to enhance language model performance as layer size scales. To avoid increased decoding costs, SCONE retains the original vocabulary while introducing embeddings for a set of frequent n-grams. These embeddings provide contextualized representation for each input token and are learned with a separate model during training. During inference, they are precomputed and stored in off-accelerator memory with minimal impact on inference speed. SCONE enables two new scaling strategies: increasing the number of cached n-gram embeddings and scaling the model used to learn them, all while maintaining fixed inference-time FLOPS. We show that scaling both aspects allows SCONE to outperform a 1.9B parameter baseline across diverse corpora, while using only half the inference-time FLOPS.

北京市昌平区探索星信息技术及软件开发工作室

京ICP备2026059466号