Transfer learning with collaborative filtering

I’ve created a recommendation model using collab_learner
data = CollabDataBunch.from_df(df, seed=42, valid_pct=0.1, item_name=‘title’)
learn = collab_learner(data, n_factors=40, y_range=y_range, wd=1e-1)

I want to add a new user with their own ratings to the model and retrain but am struggling to get it to work.
I’ve trained the model and exported it. Leaded it with learn.load('dotprod');.

My understanding is to change the data using
learn.data = CollabDataBunch.from_df(new_df, seed=42, valid_pct=0.1, item_name='title'), where new_df is a dataframe with just the new user data, in the same format as the trained dataframe. Trying to fit this though gives me RuntimeError: element 0 of tensors does not require grad and does not have a grad_fn.

Should I just add the new_df to the trained one and retrain it all again?