Qianli Scott Zhu
Qianli Scott Zhu
> I like the delegate approach. I have been able to call .compile and .fit on it although there are some errors. Training complains that some variables do not have...
The stack trace I got from the code lab is like below, and i think it is probably a GPU runtime error, since the layer it complains is actually the...
The runtime log I got from colab is like below, which suggest the tf version and cudnn version doesn't match. Btw, I was using tf-nightly and tf-models-nightly for testing. ```...
@martin-gorner, this is quite weird. I was able to train the model e2e on a GPU instance on GCP without issue. The training logic is similar as yours in the...
See a demo of training result in https://colab.corp.google.com/drive/1UeTn-MJ9DW9ujy5HKVf4yK5DQ9JmK_Re#scrollTo=PEoCoZymu1nE.
Is this PR ready to merge?
Chatted with @tanzhenyu offline for this. We will park this PR for now since this is only going to be used by the Retina-net and not blocking the segmentation model....
Triage notes: This is a long stand performance issue, and we should have someone look into this.
@fchollet, This should be cherrypicked to 2.11 release before final.