How to read data with multiple ndarrays linked to one label?

When an image such as jpeg is linked to the label,I think the pipeline is like this

class DALIPipeline(Pipeline):
  def __init__(self):
     self.input=nvidia.dali.ops.readers.File(file_root='.', file_list='dali.txt')
     self.decode = nvidia.dali.ops.decoders.Image(device = "mixed", output_type = types.DALIImageType.RGB)
  def define_graph(self):
     images, labels = self.input(name="Reader")
     images = self.decode(images)

but, I don't know how to cope with the case that image1.npy &image2.npy are linked to a label(0 or 1 or 2...)

I have train.csv which has like below

target,file_path,extra_file_path
1,~~/id0.npy,~~~/id0.npy
0,~~/id1.npy,~~~/id1.npy
0,~~/id2.npy,~~~/id2.npy

Start with the shown DALIPipeline.define_graph entry point and the train.csv and dali.txt input formats. Check the documented reader and ndarray-loading entry points, then define what a supported example for multiple .npy files sharing one label should demonstrate.