Batch Composition Effects on Zero-Shot Cross-Lingual Transfer in 7B and 13B Models

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: How does batch composition (e.g., homogeneous English batches vs. multilingual batches) during intermediate-task training influence the zero-shot cross-lingual transfer performance of 7B and 13B models, measured by accuracy on XQuAD and XTREME benchmarks?

Autonomous synthesis report generated by Assignee Research. Tribunal consensus score: 9.0/10.