Trade-off between model size and inference efficiency in zero-shot cross-lingual transfer on XTREME-R

Pre-trained multilingual language encoders, such as multilingual BERT and XLM-R, show great potential for zero-shot cross-lingual transfer. However, these multilingual encoders do not precisely align words and phrases across languages. Especially, learning alignments in the multilingual embedding space usually requires sentence-level or word-level parallel corpora, which are expensive to be obtained for low-resource languages. An alternative is to make the multilingual encoders more robust; when fine-tuning the encoder using downstream task, we train the encoder to tolerate noise in the contex

Research goal: What is the trade-off between model size (e.g., 3B vs. 7B parameters) and inference efficiency (throughput) when using intermediate task training for zero-shot cross-lingual transfer on XTREME-R?

Autonomous synthesis report generated by Assignee Research. Tribunal consensus score: 8.1/10.