Could not decode serialized type: np.ndarray. This could be because a required module is missing

[Rashmip-nd]

Description

I'm trying to generate a calibration cache file for post-training-quantizatio using Polygraphy. For which I created custom input json file referring to this [https://github.com/NVIDIA/TensorRT/blob/main/tools/Polygraphy/how-to/use_custom_input_data.md]. The input shape of the model is (1,3,384,640).

The command used is below - polygraphy convert model.onnx --int8 --load-inputs custom.json --calibration-cache custom_calib.cache -o model_trt.engine [I] Loading input data from custom.json [!] Could not decode serialized type: np.ndarray. This could be because a required module is missing.

Environment

TensorRT Version: 10.0.1.6-1

NVIDIA GPU: Tesla T4

NVIDIA Driver Version: 470.239.06

CUDA Version: 11.4

CUDNN Version:

Operating System: Ubuntu 20.04.6 LTS

Python Version (if applicable): 3.8.10

Tensorflow Version (if applicable):

PyTorch Version (if applicable):

Baremetal or Container (if so, version):

Relevant Files

Model link:

Steps To Reproduce

Commands or scripts: polygraphy convert model.onnx --int8 --load-inputs custom.json --calibration-cache custom_calib.cache -o model_trt.engine

Have you tried the latest release?:

Can this model run on other frameworks? For example run ONNX model with ONNXRuntime (polygraphy run <model.onnx> --onnxrt): Yes

[moraxu]

Have you serialized the NumPy arrays properly using JSON-compatible formats, i.e. like:

input_data = np.random.rand(1, 3, 384, 640).astype(np.float32)
input_data_list = input_data.tolist()
with open("custom.json", "w") as f:
    json.dump({"input": input_data_list}, f)

?

Do you have modules like numpy, json installed locally?

If yes, are you willing to share your custom.json content?

[lix19937]

Loading input data from custom.json [!] Could not decode serialized type: np.ndarray. This could be because a required module is missing.

Means your json has illegal type np.ndarray, you can use .tolist() to convert np.ndarray. @Rashmip-nd