Tensorboard inside a docker on a remote server

Question

Context

I'm working on a remote server and training a dataset inside a tensorflow docker container. After training the dataset, I want to view training logs with the help of tensorboard.

What I've tried

I tried to connect my local computer to the remote server with

ssh -L 16006:127.0.0.1:6006 my-name@IP_Address_remote_server

then run the docker with

docker run --gpus=all -it -p 16006:6006 tensorflow/tensorflow: latest-devel-gpu-py3

And then

tensorboard --logdir log --port=6006

And then try to access to the link that tensorboard gave me on my local computer on google chrome but I can't access to the local host,the link does not work. I tried as well

docker run --gpus=all -it -p 16006:6006 -p 6006:6006 tensorflow/tensorflow: latest-devel-gpu-py3

Same results. Can someone help me? Thank you

Answer 1

I'm having the same issues when I try to reproduce your code.

Something that I personally use and seems to do the trick:

1. Launch tensorboard with the --bind_all flag.
2. Go to IP_Address_remote_server:16006 (in a Chrome browser if you want all the tensorboard functionality such as tracing)