Preserving the order of predictions with multi gpu

Part 1 (2019)
1

hello, i am predicting on my image dataset across multiple gpus. with my dataset, the order in which the images are predicted matters. on a single GPU, the order is preserved. across multiple gpus, the order is not preserved. how may i fix this?

# load file path to saved models

files = [
    Path(file.parent, file.stem) 
    for file 
    in (path/'models').iterdir() 
    if file.name.endswith('.pth')
]

# make predictions

predictions = []
for file in files:
    print(file)
    learn = learn.load(file, with_opt=False)
    with learn.distrib_ctx():
        prediction, _ = learn.get_preds(dl=dataset.train)
    prediction = pd.Series(prediction)
    prediction = prediction.map(lambda x: float(x[0]))
    predictions.append(prediction)

predictions = pd.concat(predictions, axis=1, ignore_index=False)
predictions['probability'] = predictions.mean(axis=1)