Timezone: »

 
DoReMi: Optimizing Data Mixtures Speeds Up Language Model Pretraining
Sang Michael Xie · Hieu Pham · Xuanyi Dong · Nan Du · Hanxiao Liu · Yifeng Lu · Percy Liang · Quoc Le · Tengyu Ma · Adams Wei Yu

The mixture proportions of pretraining data domains (e.g., Wikipedia, books, web text) greatly affect language model (LM) performance. In this paper, we propose Domain Reweighting with Minimax Optimization (DoReMi), which first trains a small proxy model using group distributionally robust optimization (Group DRO) over domains to produce domain weights (mixture proportions) without knowledge of downstream tasks. We then resample a dataset with these domain weights and train a larger, full-sized model. In our experiments, we use DoReMi on a 280M-parameter proxy model to find domain weights for training an 8B-parameter model (30x larger) more efficiently. On The Pile, DoReMi improves perplexity across all domains, even when it downweights a domain. DoReMi improves average few-shot downstream accuracy by 6.5% points over a baseline model trained using The Pile's default domain weights and reaches the baseline accuracy with 2.6x fewer training steps. On the GLaM dataset, DoReMi, which has no knowledge of downstream tasks, even matches the performance of using domain weights tuned on downstream tasks.

Author Information

Sang Michael Xie (Stanford University)
Hieu Pham (Google)
Xuanyi Dong (University of Technology Sydney)
Nan Du (Google Deepmind)
Hanxiao Liu (Google Brain)
Yifeng Lu (Google)
Percy Liang (Stanford University)
Quoc Le (Google Brain)
Tengyu Ma (Stanford)
Adams Wei Yu (Google Brain)

More from the Same Authors