I was able to solve my problem by getting the inputs from the model itself:
predictor = torch.load(‘models/deepAR.pth’)
feat_static_cat = predictor.prediction_net.example_input_array[‘feat_static_cat’]
feat_static_real = predictor.prediction_net.example_input_array[‘feat_static_real’]
past_time_feat = predictor.prediction_net.example_input_array[‘past_time_feat’]
past_target = predictor.prediction_net.example_input_array[‘past_target’]
past_observed_values = predictor.prediction_net.example_input_array[‘past_observed_values’]
future_time_feat = predictor.prediction_net.example_input_array[‘future_time_feat’]
example_inputs = (feat_static_cat, feat_static_real, past_time_feat,
past_target, past_observed_values, future_time_feat)
model = predictor.prediction_net.model
traced_script_module = torch.jit.trace(model, example_inputs)