Make sure to use the correct quantized mmproj when downloading the model in LM Studio, for example.
Chat template
8-bit
16-bit