Saved in:
Bibliographic Details
Main Authors: Yao, Yiqun, fan, Siqi, Huang, Xiusheng, Fang, Xuezhi, Li, Xiang, Ni, Ziyi, Jiang, Xin, Meng, Xuying, Han, Peng, Shang, Shuo, Liu, Kang, Sun, Aixin, Wang, Yequan
Format: Preprint
Published: 2023
Subjects:
Online Access:https://arxiv.org/abs/2304.06875
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866929303887806464
author Yao, Yiqun
fan, Siqi
Huang, Xiusheng
Fang, Xuezhi
Li, Xiang
Ni, Ziyi
Jiang, Xin
Meng, Xuying
Han, Peng
Shang, Shuo
Liu, Kang
Sun, Aixin
Wang, Yequan
author_facet Yao, Yiqun
fan, Siqi
Huang, Xiusheng
Fang, Xuezhi
Li, Xiang
Ni, Ziyi
Jiang, Xin
Meng, Xuying
Han, Peng
Shang, Shuo
Liu, Kang
Sun, Aixin
Wang, Yequan
contents As language models scale up, it becomes increasingly expensive to verify research ideas because conclusions on small models do not trivially transfer to large ones. A possible solution is to establish a generic system that accurately predicts certain metrics for large models without training them. Existing scaling laws require hyperparameter search on the largest models, limiting their predicative capability. In this paper, we present an approach (namely μScaling) to predict the pre-training loss, based on our observations that Maximal Update Parametrization (μP) enables accurate fitting of scaling laws close to common loss basins in hyperparameter space. With μScaling, different model designs can be compared on large scales by training only their smaller counterparts. Further, we introduce nanoLM: an affordable LLM pre-training benchmark that facilitates this new research paradigm. With around 14% of the one-time pre-training cost, we can accurately forecast the loss for models up to 52B. Our goal with nanoLM is to empower researchers with limited resources to reach meaningful conclusions on large models. We also aspire for our benchmark to serve as a bridge between the academic community and the industry. Code for μScaling is available at https://github.com/cofe-ai/Mu-scaling. Code for nanoLLM will be available later.
format Preprint
id arxiv_https___arxiv_org_abs_2304_06875
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle nanoLM: an Affordable LLM Pre-training Benchmark via Accurate Loss Prediction across Scales
Yao, Yiqun
fan, Siqi
Huang, Xiusheng
Fang, Xuezhi
Li, Xiang
Ni, Ziyi
Jiang, Xin
Meng, Xuying
Han, Peng
Shang, Shuo
Liu, Kang
Sun, Aixin
Wang, Yequan
Computation and Language
Machine Learning
As language models scale up, it becomes increasingly expensive to verify research ideas because conclusions on small models do not trivially transfer to large ones. A possible solution is to establish a generic system that accurately predicts certain metrics for large models without training them. Existing scaling laws require hyperparameter search on the largest models, limiting their predicative capability. In this paper, we present an approach (namely μScaling) to predict the pre-training loss, based on our observations that Maximal Update Parametrization (μP) enables accurate fitting of scaling laws close to common loss basins in hyperparameter space. With μScaling, different model designs can be compared on large scales by training only their smaller counterparts. Further, we introduce nanoLM: an affordable LLM pre-training benchmark that facilitates this new research paradigm. With around 14% of the one-time pre-training cost, we can accurately forecast the loss for models up to 52B. Our goal with nanoLM is to empower researchers with limited resources to reach meaningful conclusions on large models. We also aspire for our benchmark to serve as a bridge between the academic community and the industry. Code for μScaling is available at https://github.com/cofe-ai/Mu-scaling. Code for nanoLLM will be available later.
title nanoLM: an Affordable LLM Pre-training Benchmark via Accurate Loss Prediction across Scales
topic Computation and Language
Machine Learning
url https://arxiv.org/abs/2304.06875