I don't know if this issue is specific to Gemma3:27b finetuning, but when running

    model.save_pretrained_gguf(
        f"{sft_model_path}.GGUF",
        tokenizer,
        quantization_method=["q4_k_m", "q8_0"]
    )

After finetuning, the script runs into save_to_gguf_generic() got an unexpected keyword argument 'quantization_method'.