Training, evaluating and explaining large language models on balanced and imbalanced datasets to show that language-label biases are learned. Code for our paper "Understanding the effects of language-specific class-imbalance in multilingual fine-tuning" - View it on GitHub