How I can eval model after each epoch end? #48
-
How I can eval model after each epoch end? Thanks |
Beta Was this translation helpful? Give feedback.
Replies: 3 comments
-
The evaluation code is implemented for a single GPU, without distributed data parallel. So, it wasn't included during training. However, all checkpoints after each epoch will be cached by default in the directory you specified. Each checkpoint is named as Fred. |
Beta Was this translation helpful? Give feedback.
-
It is wise not to include evaluation code in the training process. VCOCO's evaluation code is so slow that it takes me 20 minutes for each evaluation. I was wondering if VCOCO's evaluation method could be improved to speed things up. |
Beta Was this translation helpful? Give feedback.
Hi @SherlockHolmes221,
The evaluation code is implemented for a single GPU, without distributed data parallel. So, it wasn't included during training. However, all checkpoints after each epoch will be cached by default in the directory you specified. Each checkpoint is named as
ckpt_steps_epochs
. You can run evaluation on the checkpoints you are interested in.Fred.