discourse modeling

Sequential fine-tuning improves essay scoring

Sequential fine-tuning of LLaMA better captures essay task dependencies than independent or shuffled training.