Obtain n-best hypotheses with canary

[abnerLing]

I'm trying to generate n-best top hypotheses but I can't figure out how.
I've tried increasing the beam_size and setting return_best_hypothesis to False in the cfg.decoding but I'm still running into some errors.

from nemo.collections.asr.models import EncDecMultiTaskModel
canary_model = EncDecMultiTaskModel.from_pretrained('nvidia/canary-1b')

decode_cfg = canary_model.cfg.decoding
decode_cfg.beam.beam_size=3
decode_cfg.beam.return_best_hypothesis=False
canary_model.change_decoding_strategy(decode_cfg)

transcription_results = canary_model.transcribe(
    audio=['test-audio.wav']
    )

AssertionError                            Traceback (most recent call last)
[<ipython-input-61-97bdde110836>](https://localhost:8080/#) in <cell line: 1>()
----> 1 transcription_results = canary_model.transcribe(
      2     audio=['test-audio.wav'])
      3     )

11 frames
[/usr/local/lib/python3.10/dist-packages/nemo/collections/asr/parts/submodules/multitask_beam_decoding.py](https://localhost:8080/#) in format_hypotheses(self, packed_result, decoder_input_ids)
    222         if decoder_input_ids is not None:
    223             assert (
--> 224                 len(packed_result) == decoder_input_ids.shape[0]
    225             ), f"Mismatching number of examples {len(packed_result)=} {decoder_input_ids.shape[0]=}"
    226             decoder_input_ids = decoder_input_ids.detach().cpu()

AssertionError: Mismatching number of examples len(packed_result)=3 decoder_input_ids.shape[0]=1