Distill or Annotate? Cost-Efficient Fine-Tuning of Compact Models
Kang, Junmo, Xu, Wei, Ritter, Alan
–arXiv.org Artificial Intelligence
Fine-tuning large models is highly effective, however, inference can be expensive and produces carbon emissions. Knowledge distillation has been shown to be a practical solution to reduce inference costs, but the distillation process itself requires significant computational resources. Rather than buying or renting GPUs to fine-tune, then distill a large model, an NLP practitioner might instead choose to allocate the available budget to hire annotators and manually label additional fine-tuning data. In this paper, we investigate how to most efficiently use a fixed budget to build a compact model. Through extensive experiments on six diverse tasks, we show that distilling from T5-XXL (11B) to T5-Small (60M) is almost always a cost-efficient strategy compared to annotating more data to directly train a compact model (T5-Small). We further investigate how the optimal budget allocated towards computation varies across scenarios. We will make our code, datasets, annotation cost estimates, and baseline models available as a benchmark to support further work on cost-efficient training of compact models.
arXiv.org Artificial Intelligence
Jul-5-2023
- Country:
- North America
- Dominican Republic (0.04)
- United States
- Washington > King County
- Seattle (0.14)
- Minnesota > Hennepin County
- Minneapolis (0.14)
- Louisiana > Orleans Parish
- New Orleans (0.04)
- Washington > King County
- Europe
- Spain > Catalonia
- Barcelona Province > Barcelona (0.04)
- Ireland > Leinster
- County Dublin > Dublin (0.04)
- Spain > Catalonia
- Asia
- Bangladesh (0.04)
- Middle East > UAE
- Abu Dhabi Emirate > Abu Dhabi (0.04)
- North America
- Genre:
- Research Report (1.00)
- Industry:
- Technology: