Struggling to predict tabular

I’ve trained models easily with tabular.

However, when it comes to making predictions on new data… I have no clue.

The dataframe I am loading is in the exact same format as the dataframes the model trained on.

I’ve tried to follow the method in the fastai book but it’s not clear. Here is what I tried:

learn = tabular_learner(dls, metrics=rmse)
learn.fit_one_cycle(3)
epoch	train_loss	valid_loss	_rmse	time
0	0.793192	0.069080	0.262830	01:18
1	0.056668	0.051787	0.227568	01:18
2	0.023692	0.047442	0.217811	01:32
with open("./test_dfs/marks.pkl", 'rb') as f:
    test_df = pickle.load(f)
learn.predict(test_df)
AttributeError: 'DataFrame' object has no attribute 'to_frame'

What am I doing wrong?
Thanks