True. The only problem with lower quantization though is that the model fails to understand long prompts.