TFX docker image is heavy

Question

I was just wondering why tensorflow/tfx image is 8Gb while gpu tensorflow/tensorflow image is less than 3gB? At runtime, I expect tfx container to just load input artifacts, run some tensorflow code and then save output artifacts. So i can't figure out the weight delta.

Thanks and Regards

Youcef

Answer 1

Here is some explanation on why it's been so heavy:

The increase actually happened since 0.28, because we are now basing off a verified build which is compatible with GPU out of box.

You can also build the image yourself on top of a base python image.