Repository navigation
TPU Training Time #970
Answered
by
rwightman
MukundVarmaT
asked this question in
Q&A
|
Hi @rwightman I have been recently trying to set up a training experiment using torch and tpus and I came across the bits branch in this repository. It would be really helpful if you could share the training time (at least approximate) for imagenet classification using standard training procedures for models like resnets, efficientnet, etc. Thanks |
Answered by
rwightman
Nov 15, 2021
Replies: 1 comment 1 reply
|
@MukundVarmaT something you'll need to measure as it varies based on the nework and your hparams. Typical resnet50 I manage around 2800-3000 img/sec so you can calculate for imagenet and your target # of epochs. It can go higher if I lower the precision but that isn't well supported right now due to inability to force some parts of the model to stay in float32. |
1 reply
Answer selected by
MukundVarmaT
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
@MukundVarmaT something you'll need to measure as it varies based on the nework and your hparams. Typical resnet50 I manage around 2800-3000 img/sec so you can calculate for imagenet and your target # of epochs. It can go higher if I lower the precision but that isn't well supported right now due to inability to force some parts of the model to stay in float32.