This message was deleted.

Looks like this works and turns threading off. In case someone finds this message via slack search, something like this should also work, based on an earlier thread here:

from bentoml._internal.runner.strategy import Strategy
from bentoml._internal.runner.runnable import Runnable
import typing as t


class SingleThreadStrategy(Strategy):
    @classmethod
    def get_worker_count(
        cls,
        runnable_class: t.Type[Runnable],
        resource_request: dict[str, t.Any],
    ) -> int:
        return 1

    @classmethod
    def get_worker_env(
        cls,
        runnable_class: t.Type[Runnable],
        resource_request: dict[str, t.Any],
        worker_index: int,
    ) -> dict[str, t.Any]:
        return {
            "BENTOML_NUM_THREAD": 1,
            "OMP_NUM_THREADS": 1,  # openmp
            "OPENBLAS_NUM_THREADS": 1,  # openblas,
            "MKL_NUM_THREADS": 1,  # mkl,
            "VECLIB_MAXIMUM_THREADS": 1,  # accelerate,
            "NUMEXPR_NUM_THREADS": 1,  # numexpr
            "RAYON_RS_NUM_CPUS": 1,
            "TF_NUM_INTEROP_THREADS": 1,
            "TF_NUM_INTRAOP_THREADS": 1,

model_runner = bentoml.Runner(bentoml.sklearn.get(m.tag).to_runnable(), scheduling_strategy=SingleThreadStrategy)