Capability Salience Vector: Fine-grained Alignment of Loss and Capabilities for Downstream Task Scaling Law

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Ge, Qiming, Xing, Shuhao, Gao, Songyang, Zhou, Yunhua, Zou, Yicheng, Zhang, Songyang, Chen, Zhi, Yan, Hang, Zhang, Qi, Guo, Qipeng, Chen, Kai
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908408728256512
author Ge, Qiming
Xing, Shuhao
Gao, Songyang
Zhou, Yunhua
Zou, Yicheng
Zhang, Songyang
Chen, Zhi
Yan, Hang
Zhang, Qi
Guo, Qipeng
Chen, Kai
author_facet Ge, Qiming
Xing, Shuhao
Gao, Songyang
Zhou, Yunhua
Zou, Yicheng
Zhang, Songyang
Chen, Zhi
Yan, Hang
Zhang, Qi
Guo, Qipeng
Chen, Kai
contents Scaling law builds the relationship between training computation and validation loss, enabling researchers to effectively predict the loss trending of models across different levels of computation. However, a gap still remains between validation loss and the model's downstream capabilities, making it untrivial to apply scaling law to direct performance prediction for downstream tasks. The loss typically represents a cumulative penalty for predicted tokens, which are implicitly considered to have equal importance. Nevertheless, our studies have shown evidence that when considering different training data distributions, we cannot directly model the relationship between downstream capability and computation or token loss. To bridge the gap between validation loss and downstream task capabilities, in this work, we introduce Capability Salience Vector, which decomposes the overall loss and assigns different importance weights to tokens to assess a specific meta-capability, aligning the validation loss with downstream task performance in terms of the model's capabilities. Experiments on various popular benchmarks demonstrate that our proposed Capability Salience Vector could significantly improve the predictability of language model performance on downstream tasks.
format Preprint
id arxiv_https___arxiv_org_abs_2506_13216
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Capability Salience Vector: Fine-grained Alignment of Loss and Capabilities for Downstream Task Scaling Law
Ge, Qiming
Xing, Shuhao
Gao, Songyang
Zhou, Yunhua
Zou, Yicheng
Zhang, Songyang
Chen, Zhi
Yan, Hang
Zhang, Qi
Guo, Qipeng
Chen, Kai
Computation and Language
Scaling law builds the relationship between training computation and validation loss, enabling researchers to effectively predict the loss trending of models across different levels of computation. However, a gap still remains between validation loss and the model's downstream capabilities, making it untrivial to apply scaling law to direct performance prediction for downstream tasks. The loss typically represents a cumulative penalty for predicted tokens, which are implicitly considered to have equal importance. Nevertheless, our studies have shown evidence that when considering different training data distributions, we cannot directly model the relationship between downstream capability and computation or token loss. To bridge the gap between validation loss and downstream task capabilities, in this work, we introduce Capability Salience Vector, which decomposes the overall loss and assigns different importance weights to tokens to assess a specific meta-capability, aligning the validation loss with downstream task performance in terms of the model's capabilities. Experiments on various popular benchmarks demonstrate that our proposed Capability Salience Vector could significantly improve the predictability of language model performance on downstream tasks.
title Capability Salience Vector: Fine-grained Alignment of Loss and Capabilities for Downstream Task Scaling Law
topic Computation and Language
url https://arxiv.org/abs/2506.13216