Deepspeech on amd gpu

[JRMeyer]

>>> fabien4455
[May 24, 2019, 7:59am]

github.com/mozilla/DeepSpeech

#### Issue: AMD Support ?

opened by fabien4455 on
2019-05-23

closed by lissyx on
2019-05-24

{.content style='white-space: pre-wrap;'}<br/> For support and discussions, please use our Discourse forums.<br/> If you've found a bug, or have a feature request, then please create...<br/>

[invalidstyle='display:inline-block;margin-top:2px;background-color: #B8B8B8;padding: 2px;border-radius: 4px;color: #fff;margin-left: 3px;'}

Here is my github ticket to have informations...

I have AMD GPU WX9100 16GB HBM2 RAM.

I try to make deepspeech work. Actually only CPU is working...

They suggest me to increase batch size.

The file that i need to change is util/flags.py ?

[This is an archived TTS discussion thread from discourse.mozilla.org/t/deepspeech-on-amd-gpu]

[JRMeyer]

>>> lissyx
[May 24, 2019, 8:01am]



> Here is my github ticket to have informations...

Still lacking the information about the dataset that was asked on that
issue.



> The file that i need to change is util/flags.py ?

No, please read the documentation and use
--{train,dev,test}_batch_size X when starting training

[Archived Post]

[JRMeyer]

>>> fabien4455
[May 24, 2019, 11:20am]

We use this DATASET : https://voice.mozilla.org/fr/datasets

[Archived Post]

[JRMeyer]

>>> lissyx
[May 24, 2019, 2:15pm]

Looks like you are unblocked, so to update context, I've linked you the
Dockerfile for french model that includes other dataset, and now the
goal is to have a Dockerfile for AMD with feature parity to the
NVIDIA-based Docker file

[Archived Post]