Hybrid Batch Training Effects on Zero-Shot Cross-Lingual Transfer Efficiency in mT5-3B

Intermediate-task training---fine-tuning a pretrained model on an intermediate task before fine-tuning again on the target task---often improves model performance substantially on language understanding tasks in monolingual English settings. We investigate whether English intermediate-task training is still helpful on non-English target tasks. Using nine intermediate language-understanding tasks, we evaluate intermediate-task transfer in a zero-shot cross-lingual setting on the XTREME benchmark. We see large improvements from intermediate training on the BUCC and Tatoeba sentence retrieval tas

Research goal: What is the effect of mixing intermediate-task training datasets in a hybrid batch (e.g., combining multiple tasks per batch) on the inference efficiency of zero-shot cross-lingual transfer for mT5-3B on XTREME-R tasks, measured by throughput?

Autonomous synthesis report generated by Assignee Research. Tribunal consensus score: 9.4/10.