Label-Focused Tuning Pipeline

This record contains the source code, experiment configurations, aggregate metrics, and a collection of PEFT/LoRA adapters accompanying the article “Label-Focused Tuning Pipeline for Efficient Multitask Text Classification in a Low-Resource Language.” The implementation supports Joint LoRA and Dual Adapter training, inference, and evaluation for three Vietnamese multitask text-classification datasets: NEU-ESC, UIT-VSFC, and ViCTSD. The Vistral-7B base model and the original datasets are not redistributed. Information about the available checkpoints, experimental scope, environment setup, and usage instructions is provided in the accompanying documentation. GitHub repository:github.com