bug: adaptive batching not work in mlflow model (gensim word2vec)

[BangDaeng]

Describe the bug

gensim word2vec model

import os
import mlflow

from gensim.models.word2vec import Word2Vec


class ScappyWrapper(mlflow.pyfunc.PythonModel):
    def load_context(self, context):
        file_path = os.path.join(context.artifacts["model_path"], "scappy_base.bin")
        self.model = Word2Vec.load(file_path)

    def predict(self, _, model_input):
        model_input = model_input[0].tolist()
        return self.model.wv[model_input]

save_model

  bentoml.mlflow.import_model(
      tag,
      model_uri=version.source,
      signatures={"predict": {"batchable": True}},
  )

service.py

import bentoml
from bentoml.io import JSON, NumpyNdarray

from constant import BUILD_NAME, MODEL_NAME

scappy_model = bentoml.mlflow.get(MODEL_NAME)
_scappy_runnable = scappy_model.to_runnable()


class ScappyRunnable(_scappy_runnable):
    def __init__(self):
        super().__init__()

    @bentoml.Runnable.method(batchable=True)
    def predict(self, input_series):
        keywords = input_series["keywords"]
        output = super().predict(keywords)
        return output


scappy_runner = bentoml.Runner(ScappyRunnable)
svc = bentoml.Service(MODEL_NAME, runners=[scappy_runner])


@svc.api(
    input=JSON(),
    output=NumpyNdarray(),
    route=BUILD_NAME,
)
async def predict(input_series):
    output = await scappy_runner.predict.async_run(input_series)
    return output

With these codes, it was confirmed that the service and the reference are operating normally But adaptive batching not work in bentoml with mlflow,, why not??

To reproduce

No response

Expected behavior

No response

Environment

python: 3.10.12 bentoml: 1.0.22 mlflow: 2.5.0 gensim: 4.3.2