vehicle-detection-0202 responds to queries when running on CPU, gives an error when running on GPU

I am using the model vehicle-detection-0202 in its openvino_ir format running in a OVMS environment, however when I try to send requests to the model running on a CPU I can get back the expected response, e.g.

{
    "model_name": "my-model",
    "model_version": "1",
    "outputs": [{
            "name": "detection_out",
            "shape": [1, 1, 200, 7],
            "datatype": "FP32",
            "data": [0.0, ..., data, ..., 0.0]
        }]
}

If I instead try to send a request to the model running on a GPU I keep getting error messages like the following:

{"code":2,"message":"json: unsupported value: -Inf"}

(Sometimes instead of -Inf the value is NaN). I don't quite understand what I'm doing wrong, as the requests are in the same format and the data has all the same attributes. Any idea what's going on?

No files, tests, or entry points are named. Start by reproducing the same vehicle-detection-0202 request in OVMS with the openvino_ir model on CPU and GPU, then compare the output values and server logs. Done means identifying why GPU inference produces -Inf or NaN and confirming valid responses for both devices.