Which Tokens Should SFT Actually Learn? A Token-Trimming Perspective on Mathematical Reasoning
arXiv:2609.09707v1 Announce Type: new Abstract: Supervised fine-tuning (SFT) applies a uniform cross-entropy loss to all target tokens, even though different tokens provide unequal learning signals for mathematical reasoning. This uniform treatment can over-sharpen already…
- 01.arXiv:2609.09707v1 Announce Type: new Abstract: Supervised fine-tuning (SFT) applies a uniform cross-entropy loss to all target tokens, even though different tokens provide unequal learning signals for mathematical reasoning
- 02.This uniform treatment can over-sharpen already…
This material is provided as an extended analytical summary.
Read full article at sourceRelated Intelligence
All stories →Anthropic IPO 文件中的五个奇怪细节
Anthropic 最终向 SEC 提交了 S-1 招股说明书，该文件旨在让投资者了解其业务的幕后情况，包括其高管的收入、IPO 花费数十亿美元的计划，以及可能破坏其雄心壮志的风险。在超过…
Anthropic’s IPO filing reveals its strategy for keeping enterprise customers hooked
Anthropic’s S-1 shows more than half of its $100,000-plus customers now use both the Claude Platform and a first-party app, a cross-selling playbook reminiscent of Amazon that’s making enterprise customers stickier.
FairStress: тест на демографическую справедливость моделей
📌 FairStress: тест на демографическую справедливость моделей Институт Аллена опубликовал набор FairStress для проверки того, меняет ли языковая модель своё решение о человеке из-за его расы, пола, религии, возраста, инвалидности или ориентации в сферах найма, кредитования,…