Reproduce “Simple, Scalable Adaptation for Neural Machine Translation” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Ankur Bapna, Naveen Arivazhagan, Orhan Firat

The authors adapt translation models by freezing a pretrained model and adding task-specific adapters. On IWSLT’15 English–French, adapters score 44.63 BLEU versus 44.59 for fine-tuning, using less than 0.11% additional parameters. Adapter capacity varies by task: JRC-Acquis requires about 13.5% capacity to come within 0.5 BLEU of fine-tuning. Multilingual refinement improves results but leaves a small regression in high-resource translations into English.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.