multiobject time series into dataset map

[fedhere]

Hello, I have a dataset of multiple univariate time series. I want to adapt this notebook https://colab.research.google.com/drive/1uvTmh-pe1zO5TeaaRVDdoEWJ5dFDI-pA?usp=sharing from huggingface (https://huggingface.co/time-series-foundation-models/Lag-Llama) which uses the m4_weekly dataset. The dataset is read in via get_dataset into a gluonts.dataset.common.TrainDatasets

I know how to read my dataset in with PandasDataset.from_long_dataframe as a long dataframe (with an additional variable to id the time series) or as a wide dataset (see screenshots) and I can split into training and testing as shown here https://ts.gluon.ai/stable/tutorials/data_manipulation/dataset_splitting_example.html but the resulting dataset is not quite in the format i need it to run the notebook code on it

conversely, with this code

from gluonts.dataset.pandas import PandasDataset
def get_gluonts_format(df: pd.DataFrame, target, freq, prediction_length, test_length) -> TrainDatasets:
       
    meta = MetaData(freq=freq, prediction_length=prediction_length)
    train = PandasDataset(df[: -test_length], target=target)
    test = PandasDataset(df, target=target)
    df_gluonts = TrainDatasets(metadata=meta, train=train, test=test)
    return df_gluonts

get_gluonts_format(df_resampled[[0]], 0, "2d", 13, 100)

I can put one series into the correct format, but only one.