Releasing a model without benchmarks seems to say the model is probably bad...