Pytorch pt and tensorflow pb model comparison

Is pytorch pt model just the weight file or the whole graph network?

If you store the state_dict as described here as the recommended way, the file would only contain all parameters and buffers.
The not recommended way would be to store the model directly, which wouldn’t store the computation graph or the architecture, but the Python objects with all parameters, so that you would have to restore the file and folder structure. would save a scripted model, which can be used in another process without redefining the model structure.