We can try to convert the random forest model to `.onnx` model (using `skl2onnx`) and run it with onnx runner to boost the inference efficiency.