Closed Kirin-kun closed 5 years ago
Yeah, Tensorflow 1.13 doesn't work with the tensorboard callback, which means we can't log any of the loss data. This is potentially a Keras issue (It hasn't been updated since tf 1.13 came out).
We currently recommend TF 1.12 as our main supported version (it's also quicker), but will bump if/when this gets fixed.
TF 1.13.1 is now fully supported.
I deployed a new install on a brand new machine (Redhat 7.2) and it mostly works, but the graph and analysis in the gui don't seem to work.
The graph doesn't show up and when I'm trying to load the current state.json, It errors:
Using CUDA 10.0.
Maybe linked to tensorflow-1.13.1 which changed something and it's not compatible anymore?
It's not blocking (I can live without analysis and graph), but I'll just leave this issue open.