How to make batch inference in pipeline Transformers

[khongtrunght]

I'm new to BentoML. When I read this example : https://github.com/bentoml/BentoML/tree/main/examples/inference_graph . I see that the model is not batched.
When I follow the tutorial Adaptive Batching, I do somthing like that:

import bentoml

bentoml.transformers.save_model(
    name="unmasker",
    pipeline=unmasker,
    signatures={
        "__call__": {
            "batchable": True,
            "batch_dim": 0,
        },
    },
)

But it not work and yield error when run, can anybody with experiments helps me?