@james0001 welcome to the KNIME forum. I can offer this information. Especially on Apple silicon chips the keras and tensor flow integrations need some improvements.
Currently you will have to work with specific combination of older packages and for the M chips you need a workaround.