Meta-rater: A Multi-dimensional Data Selection Method for Pre-training Language Models
Paper • 2504.14194 • Published
id stringlengths 20 20 | score int64 1 5 | normalized_score float64 0.2 1 | content stringlengths 217 3.74M | sub_path stringclasses 1
value |
|---|---|---|---|---|
BkiUd47xK7FjYAb_7pWN | 5 | 1 | \section{Introduction} \label{seintro}
Let $\mathcal{N}$ be a non-Archimedean ordered field extension of $\ensuremath{\mathbb{R}}$ that is real closed and
complete in the order topology and whose Hahn group $S_\mathcal{N}$ is Archimedean, i.e. (isomorphic to) a subgroup of $\ensuremath{\mathbb{R}}$. Recall that $S_{\m... | train/arxiv |
BkiUfuXxK0wg05VB91M5 | 5 | 1 | \section{INTRODUCTION}
Motivated by the Callan-Rubakov effect in the context of magnetic
monopoles \cite{callan}, studies have been carried out recently
on the possibility that cosmic strings can also catalyze
baryon-number violation with strongly enhanced cross sections.
It has been shown that the wave function of a f... | train/arxiv |
BkiUdB7xK7IDF1Ddtq1d | 5 | 1 | \section{\protect\large \bf Introduction}
\hspace{2em}Since the discovery of high-T$_c$ superconductivity,$^1$
intensive theoretical work has been carried out to understand its properties.
Much of this effort was devoted to the analysis of two dimensional
electronic models,$^2$ in particular, the Hubbard$^3$ and
$t - ... | train/arxiv |
BkiUarzxK2li-LM1PiPP | 5 | 1 | "\\section{Introduction}\n\nFor any linear map $\\Phi:M_{d_1}(\\mathbb C) \\to M_{d_2}(\\mathbb C)$,(...TRUNCATED) | train/arxiv |
BkiUddU4uzlhrI4X8QHl | 5 | 1 | "\\section{\\hspace{-10pt}}\n\\section{The hardest logic puzzle ever}\n\\begin{defn}[The Hardest Log(...TRUNCATED) | train/arxiv |
BkiUdizxaKgQGdQApUFb | 5 | 1 | "\\section{Introduction}\\label{intro}\n\nPolar Ring/Disk Galaxies (PRGs) are multi-spin systems. Th(...TRUNCATED) | train/arxiv |
BkiUc_Y5qhLA5_F8juOy | 5 | 1 | "\\section{Introduction}\n\n\nLet $\\Gamma$ be \na finitely generated group \nand let $S$ be \na fin(...TRUNCATED) | train/arxiv |
BkiUbMU5qWTD6essZRcR | 5 | 1 | "\\section{Introduction}\nThis demo file is intended to serve as a ``starter file''\nfor IEEE confer(...TRUNCATED) | train/arxiv |
BkiUddQ4uBhi4In9Ya_Y | 5 | 1 | "\\section{Heavy hadron lifetimes}\n\nLifetimes are fundamental properties of particles, which conne(...TRUNCATED) | train/arxiv |
BkiUdwHxK0fkXQzmMBqp | 5 | 1 | "\\section{Introduction}\n\nIn 1975, T. Koornwinder (\\cite{Koor75}) introduced a non--trivial metho(...TRUNCATED) | train/arxiv |
End of preview. Expand in Data Studio
Top 30B token SlimPajama Subset selected by the Cleanliness rater
This repository contains the dataset described in the paper Meta-rater: A Multi-dimensional Data Selection Method for Pre-training Language Models.
Code: https://github.com/opendatalab/Meta-rater
Dataset Description
This dataset contains the top 30B tokens from the SlimPajama-627B corpus, selected using the Cleanliness dimension of the PRRC (Professionalism, Readability, Reasoning, Cleanliness) framework. Each document in this subset is scored and filtered by a ModernBERT-based rater fine-tuned to assess the formatting, completeness, and absence of noise or irrelevant content in the text.
- Source: SlimPajama-627B Annotated Dataset
- Selection: Top 30B tokens by PRRC-Cleanliness score
- Quality metric: Cleanliness (0–5 scale, see below)
- Annotation coverage: 100% of selected subset
Dataset Statistics
- Total tokens: 30B (subset of SlimPajama-627B)
- Selection method: Top-ranked by PRRC-Cleanliness ModernBERT rater
- Domains: Same as SlimPajama (CommonCrawl, C4, GitHub, Books, ArXiv, Wikipedia, StackExchange)
- Annotation: Each document has a cleanliness score (0–5)
Cleanliness Quality Metric
Cleanliness evaluates the formatting, completeness, and absence of noise or irrelevant content in the text. Higher scores indicate well-formatted, complete, and clean data, while lower scores reflect noisy, incomplete, or poorly formatted content.
- 0–1: Serious or obvious issues affecting fluency or completeness
- 2–3: Some problems, but not seriously affecting reading
- 4–5: Minor or no problems; text is clean and well-formatted
Scores are assigned by a ModernBERT model fine-tuned on Llama-3.3-70B-Instruct annotations, as described in the Meta-rater paper.
Annotation Process
- Initial annotation: Llama-3.3-70B-Instruct rated 500k+ SlimPajama samples for cleanliness
- Model training: ModernBERT fine-tuned on these annotations
- Scoring: All SlimPajama documents scored by ModernBERT; top 30B tokens selected
Citation
If you use this dataset, please cite:
@article{zhuang2025meta,
title={Meta-rater: A Multi-dimensional Data Selection Method for Pre-training Language Models},
author={Zhuang, Xinlin and Peng, Jiahui and Ma, Ren and Wang, Yinfan and Bai, Tianyi and Wei, Xingjian and Qiu, Jiantao and Zhang, Chi and Qian, Ying and He, Conghui},
journal={arXiv preprint arXiv:2504.14194},
year={2025}
}
License
This dataset is released under the same license as the original SlimPajama dataset. See the original SlimPajama repository for details.
Contact
- Project Lead: Ren Ma (maren@pjlab.org.cn)
- Corresponding Author: Conghui He (heconghui@pjlab.org.cn)
- Issues: GitHub Issues
Made with ❤️ by the OpenDataLab team
- Downloads last month
- 88