one would think that with photonics processors, that LLM models would be re-written in another language other than the grossly inefficient and memory hungry flawed Python.

https://docs.prefabphotonics.com/python-for-photonics/

which is why, models like candle, makes a lot of future sense.