I'm not sure whether this is a pydantic 1 -> 2 issue, or whether the underlying llama.cpp just doesn't support input for embedding in numerical tokenised form. I should note at this point that I'm ...