Error trying to extract model file deepspeech-gpu-0.6.0-alpha.2.tgz

DeepSpeech
#1

I just downloaded the latest (gpu version) model file with
wget https://github.com/mozilla/DeepSpeech/releases/tag/v0.6.0-alpha.2/deepspeech-gpu-0.6.0-alpha.2.tgz, but now am unable to extract it.

tar xvfz deepspeech-gpu-0.6.0-alpha.2.tgz throws the following error:

gzip: stdin: not in gzip format
tar: Child returned status 1
tar: Error is not recoverable: exiting now

Also, deleting, re-downloading and using my linux cinnamon gui file explorer to “extract here” also gives an error and does not unzip.

Is there a problem with this file?

Thanks!
Max

#2

I was able to download deepspeech-gpu-0.6.0-alpha.2.tgz and extract it on OS X.

#3

Thanks for confirming the file is fine, kdavis.

Ok weird - so downloading via “point and click,” I get a download which takes about 20 seconds and then I can also extract it fine.

But when I wget it, the download “finishes” within a couple seconds, I don’t get any log errors, but then I get an error when I try to extract it. That’s what I did when posting this issue originally, and just tried again and confirm it still fails for me.

Ok, I’ll just proceed by making sure to manually download the pretrained model files

#4

deepspeech-gpu-0.6.0-alpha.2.tgz is the nodejs cuda-enabled version, not the pre-trained model

#5

@mepstein68 and your link is invalid, it should be https://github.com/mozilla/DeepSpeech/releases/download/v0.6.0-alpha.2/deepspeech-gpu-0.6.0-alpha.2.tgz

#6

Hey thanks a lot lissyx, apparently I’m confused on a couple aspects here. So, two follow up questions on this:

  1. Is the 0.5.1 pretrained english model (downloadable via the command wget https://github.com/mozilla/DeepSpeech/releases/download/v0.5.1/deepspeech-0.5.1-models.tar.gz) still the recommended pretrained model to start retraining from (for transfer learning) even with version 0.6.0-alpha.2 of the code/repo?
  2. If using deepspeech-gpu, is there a separate -gpu version of the pretrained model I should be using? Or is it the same pretrained model file either way?

Thanks again for all the help!
Max

#7

Yes, model changed with 0.6, so please use 0.5.1 for transfer learning.

No, I don’t even understand how you can think it might be the case.