Gespeichert in:
| Hauptverfasser: | , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | https://arxiv.org/abs/2603.18534 |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866911529031434240 |
|---|---|
| author | Kim, Konwoo Kotha, Suhas Choi, Yejin Hashimoto, Tatsunori Haber, Nick Liang, Percy |
| author_facet | Kim, Konwoo Kotha, Suhas Choi, Yejin Hashimoto, Tatsunori Haber, Nick Liang, Percy |
| contents | Synthetic data augmentation has emerged as a promising solution when pre-training is constrained by data rather than compute. We study how to design synthetic data algorithms that achieve better loss scaling: not only lowering loss at finite compute but especially as compute approaches infinity. We first show that pre-training on web data mixed with synthetically generated rephrases improves i.i.d. validation loss on the web data, despite the synthetic data coming from an entirely different distribution. With optimal mixing and epoching, loss and benchmark accuracy improve without overfitting as the number of synthetic generations grows, plateauing near $1.48\times$ data efficiency at 32 rephrases per document. We find even better loss scaling under a new perspective: synthetic generations from the same document can form a single substantially longer megadocument instead of many short documents. We show two ways to construct megadocs: stitching synthetic rephrases from the same web document or stretching a document by inserting rationales. Both methods improve i.i.d. loss, downstream benchmarks, and especially long-context loss relative to simple rephrasing, increasing data efficiency from $1.48\times$ to $1.80\times$ at $32$ generations per document. Importantly, the improvement of megadocs over simple rephrasing widens as more synthetic data is generated. Our results show how to design synthetic data algorithms that benefit more from increasing compute when data-constrained. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_18534 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Data-efficient pre-training by scaling synthetic megadocs Kim, Konwoo Kotha, Suhas Choi, Yejin Hashimoto, Tatsunori Haber, Nick Liang, Percy Machine Learning Synthetic data augmentation has emerged as a promising solution when pre-training is constrained by data rather than compute. We study how to design synthetic data algorithms that achieve better loss scaling: not only lowering loss at finite compute but especially as compute approaches infinity. We first show that pre-training on web data mixed with synthetically generated rephrases improves i.i.d. validation loss on the web data, despite the synthetic data coming from an entirely different distribution. With optimal mixing and epoching, loss and benchmark accuracy improve without overfitting as the number of synthetic generations grows, plateauing near $1.48\times$ data efficiency at 32 rephrases per document. We find even better loss scaling under a new perspective: synthetic generations from the same document can form a single substantially longer megadocument instead of many short documents. We show two ways to construct megadocs: stitching synthetic rephrases from the same web document or stretching a document by inserting rationales. Both methods improve i.i.d. loss, downstream benchmarks, and especially long-context loss relative to simple rephrasing, increasing data efficiency from $1.48\times$ to $1.80\times$ at $32$ generations per document. Importantly, the improvement of megadocs over simple rephrasing widens as more synthetic data is generated. Our results show how to design synthetic data algorithms that benefit more from increasing compute when data-constrained. |
| title | Data-efficient pre-training by scaling synthetic megadocs |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2603.18534 |