Scaling Laws for Neural Language Models
| dc.contributor.author | Kaplan, Jared | |
| dc.contributor.author | McCandlish, Sam | |
| dc.contributor.author | Henighan, Tom | |
| dc.contributor.author | Brown, Tom B | |
| dc.contributor.author | Chess, Benjamin | |
| dc.contributor.author | Child, Rewon | |
| dc.contributor.author | Gray, Scott | |
| dc.contributor.author | Radford, Alec | |
| dc.contributor.author | Wu, Jeffrey | |
| dc.contributor.author | Amodei, Dario | |
| dc.date.accessioned | 2025-06-02T13:13:50Z | |
| dc.date.available | 2025-06-02T13:13:50Z | |
| dc.date.issued | 2020-01-23 | |
| dc.description | 이 논문은 언어모델 성능이 파라미터 수, 데이터 크기, 연산량에 대해 power-law 패턴으로 증가함을 수치적으로 분석합니다. ©2020 OpenAI | |
| dc.description.abstract | We study empirical scaling laws for language model performance on the cross-entropy loss. The loss scales as a power-law with model size, dataset size, and the amount of compute used for training, with some trends spanning more than seven orders of magnitude. Other architectural details such as network width or depth have minimal effects within a wide range. Simple equations govern the dependence of overfitting on model/dataset size and the dependence of training speed on model size. These relationships allow us to determine the optimal allocation of a fixed compute budget. Larger models are significantly more sample-efficient, such that optimally compute-efficient training involves training very large models on a relatively modest amount of data and stopping significantly before convergence. | |
| dc.description.sponsorship | OpenAI | |
| dc.identifier.uri | https://arxiv.org/abs/2001.08361 | |
| dc.identifier.uri | http://data.inu.ac.kr/handle/123456789/1956 | |
| dc.language.iso | en_US | |
| dc.publisher | arXiv | |
| dc.subject | Scaling Laws | |
| dc.subject | Language Models | |
| dc.subject | Parameter Scaling | |
| dc.subject | Deep Learning | |
| dc.title | Scaling Laws for Neural Language Models | |
| dc.type | Article |