This is the code for the paper Donkii: Can Annotation Error Detection Methods Find Errors in Instruction-Tuning Datasets? Installation pip install -r requirements Reproduces results bash run_experiments.sh