Metraipolis › archive › dc_da8ac22b647f
LESSER: Post-Training Data Selection with Output-Layer Gradients — arXiv:2610.03702
- Source
- arXiv
- Original
- https://arxiv.org/abs/2610.03702
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-05T23:10:30.643Z
- SHA-256 of the archived text
cb099aee4ee4245dac3ac3a94e27bf16b6c20caf4ec8c36fb79b6156d53b097d- Stable URL of this page
https://metraipolis.com/documento/dc_da8ac22b647f
How to cite
LESSER: Post-Training Data Selection with Output-Layer Gradients — arXiv:2610.03702 — arXiv, https://arxiv.org/abs/2610.03702 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_da8ac22b647f (Metraipolis archive dc_da8ac22b647f, sha256 cb099aee4ee4245dac3ac3a94e27bf16b6c20caf4ec8c36fb79b6156d53b097d)
BibTeX
@misc{dcda8ac22b647f,
title = {LESSER: Post-Training Data Selection with Output-Layer Gradients — arXiv:2610.03702},
howpublished = {arXiv},
url = {https://arxiv.org/abs/2610.03702},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_da8ac22b647f; sha256 cb099aee4ee4245dac3ac3a94e27bf16b6c20caf4ec8c36fb79b6156d53b097d},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv, https://arxiv.org/abs/2610.03702. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
The choice of post-training data for large language models substantially affects downstream performance. Gradient-based data selection is a popular approach that ranks training data by how well their gradients align with those of a small validation set. However, ranking with full-parameter gradients requires an expensive backward pass on every sample, making computation intractable for large candidate pools. This raises a natural question: can we approximate full-gradient features at a fraction of the cost? Conveniently, we find that output-layer gradients suffice for effective data selection, yet require only the cheaper forward pass. We implement this as LESSER, a drop-in wrapper for selection methods that reduces the feature-extraction FLOP cost by $9.7\times$ for SFT and $3.0\times$ for RL benchmarks, while tracking full-gradient performance on downstream tasks. Empirically, we find that even when output-layer and full gradients rank individual samples differently, they select batches with aligned gradients. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.LG, nuovi lavori del giorno). Categorie arXiv: Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2610.03702 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-05T23:01:38.722Z dalla pagina https://arxiv.org/list/cs.LG/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).