A Docker image for running local TensorFlow instance on Google Colab with GPU support. No need to figure out all those dependencies.

I run it on WSL2, but it should work on any Linux machine. Build by running:

docker build -t tf-colab-local-gpu .

Then run it with:

docker run --gpus all -u $(id -u):$(id -g) -it --rm -p 8888:8888 tf-colab-local-gpu

(I update instructions with mounting the directory, but I don't use it)

It will print a link to the Jupyter notebook, something like this:

https://d1af4f9a601c:8888/?token=...

Replace the Docker container ID with localhost and paste it into the Colab.

1. Publish on Docker Hub
2. Continuous build
3. Fix the printed link, if possible.

I may also install extra PIP packages and such as I am learning Keras and TensorFlow.