Saved in:
Bibliographic Details
Main Authors: Li, Yudong, Cai, Jiawei, Shen, Linlin
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2604.12397
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910128247144448
author Li, Yudong
Cai, Jiawei
Shen, Linlin
author_facet Li, Yudong
Cai, Jiawei
Shen, Linlin
contents Standard Large Language Model (LLM) pre-training typically treats corpora as flattened token sequences, often overlooking the real-world context that humans naturally rely on to contextualize information. To bridge this gap, we introduce Knowledge Coordinate Conditioning (KoCo), a simple method that maps every document into a three-dimensional semantic coordinate. By prepending these coordinates as textual prefixes for pre-training, we aim to equip the model with explicit contextual awareness to learn the documents within the real-world knowledge structure. Experiment results demonstrate that KoCo significantly enhances performance across 10 downstream tasks and accelerates pre-training convergence by approximately 30\%. Furthermore, our analysis indicates that explicitly modeling knowledge coordinates helps the model distinguish stable facts from noise, effectively mitigating hallucination in generated outputs.
format Preprint
id arxiv_https___arxiv_org_abs_2604_12397
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle KoCo: Conditioning Language Model Pre-training on Knowledge Coordinates
Li, Yudong
Cai, Jiawei
Shen, Linlin
Computation and Language
Standard Large Language Model (LLM) pre-training typically treats corpora as flattened token sequences, often overlooking the real-world context that humans naturally rely on to contextualize information. To bridge this gap, we introduce Knowledge Coordinate Conditioning (KoCo), a simple method that maps every document into a three-dimensional semantic coordinate. By prepending these coordinates as textual prefixes for pre-training, we aim to equip the model with explicit contextual awareness to learn the documents within the real-world knowledge structure. Experiment results demonstrate that KoCo significantly enhances performance across 10 downstream tasks and accelerates pre-training convergence by approximately 30\%. Furthermore, our analysis indicates that explicitly modeling knowledge coordinates helps the model distinguish stable facts from noise, effectively mitigating hallucination in generated outputs.
title KoCo: Conditioning Language Model Pre-training on Knowledge Coordinates
topic Computation and Language
url https://arxiv.org/abs/2604.12397