This message was deleted.

Any help will be greatly appreciated.

Hi Yash, pipeline is the recommended inference api in transformers. We support the other methods only because there are some use cases where pipeline is not applicable

When I save a pipeline as bentoml model for multi label classification with top_k=None even though the pipeline returns all the labels, the model when served returns only the top label. code for pipeline:

bento_model = pipeline('text-classification', model=model, tokenizer=tokenizer, top_k=None)

bentoml.transformers.save_model(name="bento-model", pipeline=bento_model)

code for service:

runner = bentoml.transformers.get("bento-model:latest").to_runner()

svc = bentoml.Service("bento-model", runners=[runner])

@svc.api(input=Text(), output=JSON())
def bento_model(input_series: str):
    return runner.run(input_series)