Verifying that minibatch is actually randomly sampling

Your plot looks like this after my bugfix:
image