bug: We can NOT achieve json input about bentoml HOT 10 CLOSED

try if this works for you:

import bentoml

@bentoml.service(
    traffic={"timeout": 10},
)
class Hello:
    @bentoml.api(input_spec=dict)
    def predict(self, **data: Any) -> dict:
        print("data:", data)
        return data

from bentoml.

From the api_endpoint method of the ServiceAppFactory class, it call the input_spec.from_http_request method with two arguments passed, but in bentoml.io JSON class, the from_http_request method only take 1 argument !

from bentoml.

You are mixing new APIs with legacy ones. Here is the working code:

import bentoml
from bentoml.io import JSON

@bentoml.service(
    traffic={"timeout": 10},
)
class Hello:
    @bentoml.api
    def predict(self, data: dict) -> dict:
        print("data:", data)
        return data

We no longer needs to specify input= and output= since they are moved to the function signature

from bentoml.

@frostming thanks for quickly reply
But I do not want t wrapper of the request body, I have tried the way you just mentioned, but it requires the request body like this: {"data": {"f1": 1, "f2": 2}}, I do not want a nested data field, I want request body in this way: {"f1": 1, "f2": 2}

from bentoml.

our model depends on hundreds of features, and it's managed by the company metadata service, the upstream would pass the parameter in the way above, we just hope to replace the model serving part to Bentoml, that's why I need the parameter in the way above.

from bentoml.

Hi @frostming that does not work, the swagger ui can not be opened and the predict call also failed

from bentoml.

@cceasy needs a little workaround:

import bentoml
from pydantic import RootModel


class Input(RootModel[dict]):
    pass


@bentoml.service(
    traffic={"timeout": 10},
)
class Hello:
    @bentoml.api(input_spec=Input)
    def predict(self, root: dict) -> dict:
        print("data:", root)
        return root

from bentoml.

@frostming I am afraid the workaround not work

from bentoml.

@cceasy Make sure you are using the latest patch release. It is working on my side.

from bentoml.

Hi @frostming Awesome, after upgrade from 1.2.4 to 1.2.9. it works.
Thank you !

from bentoml.