• 论文信息与摘要
• 1 Introduction
• 2 Background
• 3 Pretraining datasets
• 4 Pretraining
• 5 Post-training
• 6 SmolLM2 135M and 360M
• 7 Conclusion
• Acknowledgments
• Impact Statement
• References
• Appendix A Training setup
• Appendix B English web ablations
• Appendix C FineMath
• Appendix D Stack-Edu
• Appendix E Detailed pretraining results
• Appendix F Post-training
• Appendix G Long context evaluations