Guardado en:
Detalles Bibliográficos
Autores principales: Liu, Chen, Sun, Xingzhi, Xiao, Xi, Van Tassel, Alexandre, Xu, Ke, Reimann, Kristof, Liao, Danqi, Gerstein, Mark, Wang, Tianyang, Wang, Xiao, Krishnaswamy, Smita
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:https://arxiv.org/abs/2602.00217
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866911645485236224
author Liu, Chen
Sun, Xingzhi
Xiao, Xi
Van Tassel, Alexandre
Xu, Ke
Reimann, Kristof
Liao, Danqi
Gerstein, Mark
Wang, Tianyang
Wang, Xiao
Krishnaswamy, Smita
author_facet Liu, Chen
Sun, Xingzhi
Xiao, Xi
Van Tassel, Alexandre
Xu, Ke
Reimann, Kristof
Liao, Danqi
Gerstein, Mark
Wang, Tianyang
Wang, Xiao
Krishnaswamy, Smita
contents Large language models (LLMs) achieve remarkable performance through ever-increasing parameter counts, but scaling incurs steep computational costs. To better understand LLM scaling, we study representational differences between LLMs and their smaller counterparts, with the goal of replicating the representational qualities of larger models in smaller models. We observe a geometric phenomenon which we term $\textbf{embedding condensation}$, where token embeddings collapse into a narrow cone-like subspace in some language models. Through systematic analyses across multiple Transformer families, we show that small models such as $\texttt{GPT2}$ and $\texttt{Qwen3-0.6B}$ exhibit severe condensation, whereas larger models such as $\texttt{GPT2-xl}$ and $\texttt{Qwen3-32B}$ are more resistant to this phenomenon. Additional observations show that embedding condensation is not reliably mitigated by knowledge distillation from larger models. To fight against it, we formulate a dispersion loss that explicitly encourages embedding dispersion during training. Experiments demonstrate that it mitigates condensation, recovers dispersion patterns seen in larger models, and yields performance gains across 10 benchmarks. We believe this work offers a principled path toward improving smaller Transformers without additional parameters.
format Preprint
id arxiv_https___arxiv_org_abs_2602_00217
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Dispersion Loss Counteracts Embedding Condensation and Improves Generalization in Small Language Models
Liu, Chen
Sun, Xingzhi
Xiao, Xi
Van Tassel, Alexandre
Xu, Ke
Reimann, Kristof
Liao, Danqi
Gerstein, Mark
Wang, Tianyang
Wang, Xiao
Krishnaswamy, Smita
Machine Learning
Large language models (LLMs) achieve remarkable performance through ever-increasing parameter counts, but scaling incurs steep computational costs. To better understand LLM scaling, we study representational differences between LLMs and their smaller counterparts, with the goal of replicating the representational qualities of larger models in smaller models. We observe a geometric phenomenon which we term $\textbf{embedding condensation}$, where token embeddings collapse into a narrow cone-like subspace in some language models. Through systematic analyses across multiple Transformer families, we show that small models such as $\texttt{GPT2}$ and $\texttt{Qwen3-0.6B}$ exhibit severe condensation, whereas larger models such as $\texttt{GPT2-xl}$ and $\texttt{Qwen3-32B}$ are more resistant to this phenomenon. Additional observations show that embedding condensation is not reliably mitigated by knowledge distillation from larger models. To fight against it, we formulate a dispersion loss that explicitly encourages embedding dispersion during training. Experiments demonstrate that it mitigates condensation, recovers dispersion patterns seen in larger models, and yields performance gains across 10 benchmarks. We believe this work offers a principled path toward improving smaller Transformers without additional parameters.
title Dispersion Loss Counteracts Embedding Condensation and Improves Generalization in Small Language Models
topic Machine Learning
url https://arxiv.org/abs/2602.00217