EvalYaks: Instruction Tuning Datasets and LoRA Fine-tuned Models for Automated Scoring of CEFR B2 Speaking Assessment Transcripts
Published in Computers and Education: Artificial Intelligence, 2024
Recommended citation: Scaria, N., Kennedy, S. J. J., Latinovich, T., & Subramani, D. N. EvalYaks: Instruction Tuning Datasets and LoRA Fine-tuned Models for Automated Scoring of CEFR B2 Speaking Assessment Transcripts. Computers and Education: Artificial Intelligence. https://arxiv.org/abs/2408.12226
EvalYaks is a family of six distinct models created to automatically evaluate and score CEFR B2 speaking assessments, achieving 96% average acceptable accuracy with a variation of 0.35 levels.