KoAlpaca is an open-source language model designed to understand and respond to Korean instructions. The model is a fine-tuned version of Polyglot-ko and LLaMA architectures, supporting parameter-efficient fine-tuning methods like QLoRA and LoRA through PEFT. Model weights are distributed via HuggingFace, and training examples including tensor parallel multi-GPU fine-tuning are provided as Colab notebooks.