Acknowledgments

本文为论文《SmolLM2：紧凑高效的小语言模型家族》「论文原文」板块的对应章节，按 arXiv 官方全文逐字保留，层级与「中文全文翻译」板块一一对应。

This work would not have been possible without the contributions and support of our collaborators and colleagues:

• •

We thank Nouamane Tazi, Phuc Nguyen, Ferdinand Mom, and Haojun Zhao for designing and building our training framework, Nanotron.

• •

We thank Guilherme Penedo and Hynek Kydlíček for building our data pipeline framework, Datatrove.

• •

We thank Clémentine Fourrier and Nathan Habib for developing our evaluation framework, LightEval.

• •

We thank all our colleagues who participated in discussions that contributed to the development and refinement of SmolLM2.

• •

We thank Muhammed Emin Baslak and Pierre-Carl Langlais for their work on enhancing SmolLM2 with Entropix.

We also extend our gratitude to the broader research community and open-source ecosystem for fostering collaboration and innovation, without which this project would not have been possible.