How to forecast a single time series using a globally trained model?

[Confused-Forecaster]

Greetings,

I have been trying to train a global univariate model on multiple time series using future exogenous variables and then use it to predict individual time series. The fitting of the model seems to work fine, but when I try to predict a single time series I get:

ValueError: There are missing combinations of ids and times in `futr_df`.
You can run the `make_future_dataframe()` method to get the expected combinations or the `get_missing_future(futr_df)` method to get the missing combinations.

I'm using AutoLSTM for testing purposes which according to this list should be a univariate model supporting future exogenous variables.

config= dict(
            input_size=12,
            futr_exog_list=['Exo_1'],
)

models = [AutoLSTM(h= 12,
            loss=MAE(),
            valid_loss=MAE(),
            config=config,
            search_alg=BasicVariantGenerator(random_state=1),
            num_samples=10,
            refit_with_val=False,
            cpus=4,
            gpus=0,
            backend="ray")                
]

nf = NeuralForecast(models=models, freq='MS')
nf.fit(df=df)
prediction = nf.predict(futr_df=futr_df)

The fitting/training df looks like this:

	ds	y	unique_id	Exo_1
0	2019-09-01	6951	0	7294.666667
1	2019-10-01	7095	0	7294.666667
2	2019-11-01	1992	0	7294.666667
3	2019-12-01	1220	0	7294.666667
4	2020-01-01	3686	0	7294.666667
...	...	...	...	...
2054	2022-07-01	7028	48	12531.463415
2055	2022-08-01	5824	48	12531.463415
2056	2022-09-01	8959	48	12531.463415
2057	2022-10-01	4842	48	12531.463415
2058	2022-11-01	9019	48	12531.463415

The futr_df for the predict like this:

	ds	unique_id	Exo_1
0	2022-12-01	0	7294.666667
1	2023-01-01	0	7294.666667
2	2023-02-01	0	7294.666667
3	2023-03-01	0	7294.666667
4	2023-04-01	0	7294.666667
5	2023-05-01	0	7294.666667
6	2023-06-01	0	7294.666667
7	2023-07-01	0	7294.666667
8	2023-08-01	0	7294.666667
9	2023-09-01	0	7294.666667
10	2023-10-01	0	7294.666667
11	2023-11-01	0	7294.666667

Using nf.make_future_dataframe() on the fitted model returns this:

	unique_id	ds
0	0	2022-12-01
1	0	2023-01-01
2	0	2023-02-01
3	0	2023-03-01
4	0	2023-04-01
...	...	...
583	48	2023-07-01
584	48	2023-08-01
585	48	2023-09-01
586	48	2023-10-01
587	48	2023-11-01

Shouldn't a univariate global model be able to forecast a time series in isolation?