Mitchell DeHaven

@busishengui Did you ever resolve this issue? Having similar issues on a different dataset.

@Giovani-Merlin Is that repo still active? The link is now dead.

@bzp83 The `--patience` flag essentially tells you how many epochs can elapse without an improvement in the best validation loss before it terminates training. I was using a patience of...

STFT is onnx exportable, you just need `return_complex=False` in the torch.stft definition (as onnx supports STFT, but not with complex values).

> Using the following repro: > > ```python > from speechbrain.pretrained import EncoderDecoderASR > import torch > > asr_model= EncoderDecoderASR.from_hparams(source="speechbrain/asr-wav2vec2-transformer-aishell", savedir="pretrained_models/asr-transformer-aishell") > asr_model.eval() > wavs = torch.rand(1,34492) > wav_lens =...

Has anyone gotten this code working that could give the steps required to get it training?

> `distil-small.en` is released here: https://huggingface.co/distil-whisper/distil-small.en > > It's quite hard to compress further than this without loosing WER performance: https://huggingface.co/distil-whisper/distil-small.en#why-is-distil-smallen-slower-than-distil-large-v2 Is there any way we can access the small...

> > Is there any way we can access the small 2-layer decoder variant? > > Yes, _c.f._ https://huggingface.co/distil-whisper/distil-small.en @sanchit-gandhi From https://huggingface.co/distil-whisper/distil-small.en: > While distil-medium.en and distil-large-v2 use two decoder...

If you don't mind editing the code and don't want to change versions for whatever reason, you can simply modify the device for that portion of the model to run...

The speed improvement from 4 decoder layers to 2 decoder layers would probably be negligible.