QLoRA Supervised Fine-Tuning for LLaMA-2
Fine-tuned LLaMA-2-7B-Chat using QLoRA on a single T4 GPU, achieving a 43% relative ROUGE-L improvement on long-form cover letter generation.
Fine-tuned LLaMA-2-7B-Chat using QLoRA on a single T4 GPU, achieving a 43% relative ROUGE-L improvement on long-form cover letter generation.
Identifying how socioeconomic and demographic factors drive chronic absenteeism in NYC public high schools using interpretable regression and ensemble models.
Identifying what predicts young people returning to NYC’s Summer Youth Employment Program, using longitudinal administrative and survey data with cluster-robust logistic regression.