How to load compressed model ?

[WYYAHYT]

The compressed model can't be loaded with the origin model shape, so the compressed model can't be applied, it can only be used to check accuracy after training.

Maybe the shape of compressed model can be saved for building corresponding model class.

[Tianyi Chen]

Thanks for reaching out. You may find the solution, the compressed model can be directly loaded in most cases

compressed_model = torch.load(compressed_model_path)

Please note that there are some occasional cases that torch can not directly load model. In such case, we provide ckpt_format='onnx' to yield compressed model in onnx format.

oto.construct_subnet(
            out_dir=OUT_DIR,
            ckpt_format='onnx'
        )

See test_yolo_v5.