It seems that your GPU memory is already containing an old model. You are probably running 2 notebook in parallel, so you have to close all the notebooks before loading this model.

It is also possible that you have loaded 2 big models in the same notebook which the the GPU memory can't handle. Try to restart your notebook in this case and run your model again.

If neither of these 2 scenarios is correct, you might need to reduce your sequence length/batch size or use a smaller model.