Extremely High GPU VRAM Usage During Conversion

[chadrockey]

I've observed an issue where occasionally my VM sessions will crash and I tracked it down to VRAM usage during the conversion.

Here is a typical amount of VRAM usage:

And here is the jump when the conversion begins:

I was able to avoid this by adding this before the conversion. Then the VRAM isn't increased and even the normal ram isn't increased at all.

    import tensorflow as tf
    tf.config.experimental.set_visible_devices([], 'GPU')