Scaling Language Model Size for Zero-Shot Cross-Lingual Transfer Performance on XTREME

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: Does scaling up the size of the language model during intermediate-task training (e.g., from 1B to 10B parameters) improve zero-shot cross-lingual transfer performance on XTREME, and how does it compare to smaller models in terms of F1 scores and inference efficiency?

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