Scaling Embedding Layers in Language Models

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Yu, Da, Cohen, Edith, Ghazi, Badih, Huang, Yangsibo, Kamath, Pritish, Kumar, Ravi, Liu, Daogao, Zhang, Chiyuan
Format: Preprint
Publié: 2025
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866908607987056640
author Yu, Da
Cohen, Edith
Ghazi, Badih
Huang, Yangsibo
Kamath, Pritish
Kumar, Ravi
Liu, Daogao
Zhang, Chiyuan
author_facet Yu, Da
Cohen, Edith
Ghazi, Badih
Huang, Yangsibo
Kamath, Pritish
Kumar, Ravi
Liu, Daogao
Zhang, Chiyuan
contents We propose $SCONE$ ($S$calable, $C$ontextualized, $O$ffloaded, $N$-gram $E$mbedding), a new method for extending input embedding layers to enhance language model performance. 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. After training, embeddings are precomputed and stored in off-accelerator memory; during inference, querying them has minimal impact on latency due to the low complexity of embedding lookups. $SCONE$ enables two new scaling strategies: increasing the number of n-gram embeddings and scaling the model used to learn them, both while maintaining fixed accelerator usage during inference (in terms of FLOPS and memory). We show that scaling both aspects enables a model with 1B accelerator-resident parameters to outperform a 1.9B-parameter baseline across diverse corpora, while using only about half the FLOPS and accelerator memory during inference.
format Preprint
id arxiv_https___arxiv_org_abs_2502_01637
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Scaling Embedding Layers in Language Models
Yu, Da
Cohen, Edith
Ghazi, Badih
Huang, Yangsibo
Kamath, Pritish
Kumar, Ravi
Liu, Daogao
Zhang, Chiyuan
Computation and Language
Machine Learning
We propose $SCONE$ ($S$calable, $C$ontextualized, $O$ffloaded, $N$-gram $E$mbedding), a new method for extending input embedding layers to enhance language model performance. 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. After training, embeddings are precomputed and stored in off-accelerator memory; during inference, querying them has minimal impact on latency due to the low complexity of embedding lookups. $SCONE$ enables two new scaling strategies: increasing the number of n-gram embeddings and scaling the model used to learn them, both while maintaining fixed accelerator usage during inference (in terms of FLOPS and memory). We show that scaling both aspects enables a model with 1B accelerator-resident parameters to outperform a 1.9B-parameter baseline across diverse corpora, while using only about half the FLOPS and accelerator memory during inference.
title Scaling Embedding Layers in Language Models
topic Computation and Language
Machine Learning
url https://arxiv.org/abs/2502.01637