Hi Godfrey, thx for the message! Some notes in no particular order :-)
First: YES, I should totally write up an update with the current TF version and address the new challenges — but you know Hofstadter’s law, right?
Second, luckily for us, there’s some optimization that can be done! When happened to me, I followed general rules (like the ones here, for example: https://medium.com/@maebert/machine-learning-on-aws-lambda-5dc57127aee1#.hj8rkzrmh) to save some space, remove example folders and unused TF folders and then upload my zip. Let me know how that goes!
Third, Google announced Tensorflow lite some time ago, but I can’t find anything on the website today — I found TF mobile though, which seems a less bloated version that could fit this use case. Have you tried that?