finetuned to korean with wikipedia dataset. batch size 32768 using gc trainer. base model is gte-multilingual-base.
