How much GPU memory is needed for finetuning stt_en_conformer_ctc_small? #3252
Answered
by
titu1994
tareqalmuntasir7
asked this question in
Q&A
-
|
Hi, I am trying to finetune stt_en_conformer_ctc_small, both encoder and decoder with a RTX 3090 24GB GPU. But even when I use batch size 1, it's throwing OOM. Is this model too big or is it a bug? I see there are only 51 million parameters in the model. |
Beta Was this translation helpful? Give feedback.
Answered by
titu1994
Nov 26, 2021
Replies: 1 comment
-
|
What is your maximum audio length ? It should not be more than 16-20 seconds during training. We can fit about batch size 16 with 32 GB of RAM at fp32, and bs 32 at fp16 (but training is unstable with fp16) |
Beta Was this translation helpful? Give feedback.
0 replies
Answer selected by
tareqalmuntasir7
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
What is your maximum audio length ? It should not be more than 16-20 seconds during training. We can fit about batch size 16 with 32 GB of RAM at fp32, and bs 32 at fp16 (but training is unstable with fp16)