Thank you!
Its working and I can even use onnxruntime-directml (package) to run this on my AMD GPU! For that - the provider of ort_session_A and ort_session_C needs to be forced to ['CPUExecuti...
I noticed the function convert_char_to_pinyin wasn't correctly tokenizing my text according to my vocab. This only affects custom vocabs, but the function was returning every word as a token, even ...