Espnet: Is there any plan to support multi-machine multi-GPU ASR training?
Hello, I notice that espnet's PyTorch version uses data_parallel to support multi-GPU, but to my understanding, this can only be used in single-machine. Is there any plan to support multi-machine multi-GPU ASR training? So that making 10k+ hours training feasible in the industry. Thx.
kaituoxu
👍1
All 3 comments
We have been working internally with several projects and it's working with some minor changes, but I want to move in this direction once we fix a new training abstraction design (e.g., #1372). We're thinking of a (super) major update for this.
sw005320
on 20 Nov 2019
👍3
This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.
![stale[bot] picture](https://avatars3.githubusercontent.com/in/1724?v=4&s=40)
stale[bot]
on 4 Jan 2020
This issue is closed. Please re-open if needed.
stale[bot]
on 3 Feb 2020
Was this page helpful?
0 / 5 - 0 ratings
Related issues
Syrup274
路
4Comments
thrfdth
路
4Comments
ShigekiKarita
路
3Comments
enamoria
路
4Comments
marcosnetopires
路
4Comments
Most helpful comment
We have been working internally with several projects and it's working with some minor changes, but I want to move in this direction once we fix a new training abstraction design (e.g., #1372). We're thinking of a (super) major update for this.