Scaling Intermediate-Task Training Data for Cross-Lingual Code Benchmark Performance

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 the intermediate-task training dataset size for English code benchmarks (e.g., expanding beyond HumanEval/MBPP) lead to proportional improvements in zero-shot cross-lingual transfer performance on XTREME-R, as measured by BLEU and correctness scores?

Autonomous synthesis report generated by Assignee Research. Tribunal consensus score: 8.5/10.