We might be able to strip some of the CONVERSION_RULES listed in malt/core/config.py (e.g. the tensorflow specific ones) which may improve performance, but I don't know if it is measurable.
Originally posted by @dime10 in https://github.com/PennyLaneAI/diastatic-malt/pull/2#discussion_r1476291437