Deploying to Model Hub for Inference with custom tokenizer

@jodiak Did you ever find a solution to this?