[jdthedisciple]

If for testing then why not just mock the whole thing for ultimate performance ... ?

[nkozyra]

Probably faster to use off the shelf model with llama.cpp than to mock it