How to Compile a ONNX Models that export with dynamic_axes?

#1

Hi

        feat_dim = configs['model_args'].get('feat_dim', 80)
        if 'feature_args' in configs:  # deprecated IO
            num_frms = configs['feature_args'].get('num_frms', 200)
        else:  
            num_frms = configs['dataset_args'].get('num_frms', 200)

        dummy_input = torch.ones(1, num_frms, feat_dim)
        torch.onnx.export(
            model, dummy_input,
            args.output_model,
            do_constant_folding=True,
            verbose=False,
            opset_version=14,
            input_names=['feats'],
            output_names=['embs'],
            dynamic_axes={'feats': {0: 'B', 1: 'T'}, 'embs': {0: 'B'}})

How should the onnx model generated by the conversion above be compiled with tvm to support dynamic input? Is there any example? Thanks！