Movatterモバイル変換

[0]ホーム
URL:

Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

 Sign up
Appearance settings

Estimation of running time?#541

Sannndy0000 started this conversation inGeneral
Discussion options

Hi,

I just set up ASR and SV downstreams, but I found that running evaluation (which involves training) on each task took more than 24 hours (with one V100 GPU). Is this normal? May I get an estimation on how it it usually takes to finish evaluation on the whole benchmark?

Or I could have missed something. I wrote my own upstream model. Do I need to freeze the base model weights somewhere, or will this library do this automatically?

Thanks!

You must be logged in to vote

Replies: 2 comments

Comment options

Yes, it is normal.

You must be logged in to vote
0 replies
Comment options

Hi. I also had a query regarding training time.

I am training on top of fbank features using default ASR task config described here:https://github.com/s3prl/s3prl/blob/main/s3prl/downstream/docs/superb.md#asr-automatic-speech-recognition

I am training on one A100 gpu.

It has been 24 hours since training, but the logs show that overall is only 18% complete. This seems very slow to me. Is this accurate?

I also noticed that for ASR, the default gradient_accumulate_steps=1. This would suggest that using DDP to speed training would produce behavior that differs from default.

You must be logged in to vote
0 replies
Sign up for freeto join this conversation on GitHub. Already have an account?Sign in to comment
Category
General
Labels
None yet
3 participants
@Sannndy0000@mahiluthra@raotnameh
[8]ページ先頭
©2009-2026 Movatter.jp