What Language Model to Train if You Have One Million GPU Hours?
Authors:
- Teven Le ScaoTSTeven Le Scao
- Thomas WangTWThomas Wang
- Daniel HesslowDHDaniel Hesslow
- Lucile SaulnierLSLucile Saulnier
- Stas BekmanSBStas Bekman
- Niklas MuennighoffNMNiklas Muennighoff
- Colin RaffelCRColin Raffel
Preprint
Abstract
The crystallization of modeling methods around the Transformer architecture has been a boon for practitioners. Simple, well-motivated architectural variations can transfer across tasks and scale, increasing the impact of modeling research. However, with the emergence of state-of-the-art 100B+ parameters models, large language models are increasingly expensive to accurately design and train. Notably, it can be difficult to evaluate how modeling decisions may impact emergent capabilities, given that these capabilities arise mainly from sheer scale alone. In the process of building BLOOM--the Big Science Large Open-science Open-access Multilingual language model--our goal is to identify an architecture and training setup that makes the best use of our 1,000,000 A100-GPU-hours budget. Specifically, we perform an ablation study at the billion-parameter scale comparing different modeling practices and their impact on zero-shot generalization. In addition, we study the impact of various popular pre-training corpora on zero-shot generalization. We also study the performance of a multilingual model and how it compares to the English-only one. Finally, we consider the scaling behaviour of Transformers to choose the target model size, shape, and training setup. All our models and code are open-sourced at https://huggingface.co/bigscience .Keywords
All Related Versions
This publication has 16 references indexed in Scilit:
- RoFormer: Enhanced transformer with Rotary Position EmbeddingNeurocomputing, 2023
- Predictability and Surprise in Large Generative Models2022
- BitFit: Simple Parameter-efficient Fine-tuning for Transformer-based Masked Language-models2022
- The Power of Scale for Parameter-Efficient Prompt Tuning2021
- MathQA: Towards Interpretable Math Word Problem Solving with Operation-Based Formalisms2019
- BoolQ: Exploring the Surprising Difficulty of Natural Yes/No Questions2019
- “Going on a vacation” takes longer than “Going for a walk”: A Study of Temporal Commonsense Understanding2019
- Can a Suit of Armor Conduct Electricity? A New Dataset for Open Book Question Answering2018
- RACE: Large-scale ReAding Comprehension Dataset From Examinations2017
- SQuAD: 100,000+ Questions for Machine Comprehension of Text2016