Perplexity-Aware Data Scaling Law: Perplexity Landscapes Predict Performance for Continual Pre-training
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866918263782375424 |
|---|---|
| author | Liu, Lei Zhu, Hao Shen, Yue Chu, Zhixuan Wang, Jian Gu, Jinjie Ren, Kui |
| author_facet | Liu, Lei Zhu, Hao Shen, Yue Chu, Zhixuan Wang, Jian Gu, Jinjie Ren, Kui |
| contents | Continual Pre-training (CPT) serves as a fundamental approach for adapting foundation models to domain-specific applications. Scaling laws for pre-training define a power-law relationship between dataset size and the test loss of an LLM. However, the marginal gains from simply increasing data for CPT diminish rapidly, yielding suboptimal data utilization and inefficient training. To address this challenge, we propose a novel perplexity-aware data scaling law to establish a predictive relationship between the perplexity landscape of domain-specific data and the test loss. Our approach leverages the perplexity derived from the pre-trained model on domain data as a proxy for estimating the knowledge gap, effectively quantifying the informational perplexity landscape of candidate training samples. By fitting this scaling law across diverse perplexity regimes, we enable adaptive selection of high-utility data subsets, prioritizing content that maximizes knowledge absorption while minimizing redundancy and noise. Extensive experiments demonstrate that our method consistently identifies near-optimal training subsets and achieves superior performance on both medical and general-domain benchmarks. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2512_21515 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Perplexity-Aware Data Scaling Law: Perplexity Landscapes Predict Performance for Continual Pre-training Liu, Lei Zhu, Hao Shen, Yue Chu, Zhixuan Wang, Jian Gu, Jinjie Ren, Kui Machine Learning Computation and Language Continual Pre-training (CPT) serves as a fundamental approach for adapting foundation models to domain-specific applications. Scaling laws for pre-training define a power-law relationship between dataset size and the test loss of an LLM. However, the marginal gains from simply increasing data for CPT diminish rapidly, yielding suboptimal data utilization and inefficient training. To address this challenge, we propose a novel perplexity-aware data scaling law to establish a predictive relationship between the perplexity landscape of domain-specific data and the test loss. Our approach leverages the perplexity derived from the pre-trained model on domain data as a proxy for estimating the knowledge gap, effectively quantifying the informational perplexity landscape of candidate training samples. By fitting this scaling law across diverse perplexity regimes, we enable adaptive selection of high-utility data subsets, prioritizing content that maximizes knowledge absorption while minimizing redundancy and noise. Extensive experiments demonstrate that our method consistently identifies near-optimal training subsets and achieves superior performance on both medical and general-domain benchmarks. |
| title | Perplexity-Aware Data Scaling Law: Perplexity Landscapes Predict Performance for Continual Pre-training |
| topic | Machine Learning Computation and Language |
| url | https://arxiv.org/abs/2512.21515 |