openvino: "RuntimeError: Cannot get dims for non static shape" while inferencing openvino

Hello! I have a strange issue while inferencing openvino (version 2022.1.0). There is a simple class:

import openvino.runtime as ov
import numpy as np

class OpenVinoEngine():
    def __init__(self, model_fpath, num_workers):
        core = ov.Core()
        raw_model = core.read_model(model_fpath, "AUTO")
        self.model = core.compile_model(raw_model, "CPU", config={"INFERENCE_NUM_THREADS": str(num_workers)})
        self.infer_request = self.model.create_infer_request()  # line 1

    def process(self, batch):
        # self.infer_request = self.model.create_infer_request()  # line 2
        self.infer_request.infer([batch])
        output = [out.data[:] for out in self.infer_request.output_tensors]
        if len(output) > 1:
            return output
        return output[0]

eng = OpenVinoEngine('model.onnx', 8)
out = eng.process(np.random.rand(1, 3, 32, 256))

The engine is working when passing my model with dynamic input shape {?,3,?,?} (model.onnx.zip) and other models too. But there is a huge memory leak - my full pipeline (which has a couple of these engines with different models), which took about 3GB with onnxruntime inference, suddenly after 10 requests takes up to 5-6GB and keeps increasing. So I try to comment line 1 and uncomment line 2, but there is an error on the very first request:

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
/tmp/ipykernel_2333043/2507853659.py in <module>
----> 1 out = eng.process(np.random.rand(1, 3, 32, 256))

/tmp/ipykernel_2333043/1809003609.py in process(self, batch)
      9 
     10     def process(self, batch):
---> 11         self.infer_request = self.model.create_infer_request()
     12         self.infer_request.infer([batch])
     13         output_tensors = [out.data[:] for out in self.infer_request.output_tensors]

~/env/lib/python3.8/site-packages/openvino/runtime/ie_api.py in create_infer_request(self)
    157         :rtype: openvino.runtime.InferRequest
    158         """
--> 159         return InferRequest(super().create_infer_request())
    160 
    161     def infer_new_request(self, inputs: Union[dict, list] = None) -> dict:

RuntimeError: Cannot get dims for non static shape

The error says that something is wrong with dynamic shape but in case with line 1 it works perfectly with the same model. Maybe someone has faced similar problems? Also maybe there is a simpler way to deal with this memory issue? Any clues would be highly appreciated.