lc0 / tf2up Goto Github PK

View Code? Open in Web Editor NEW

13.0 3.0 7.0 589 KB

👨‍🎨A tiny web service to help you upgrade your TensorFlow code

Home Page: http://tf2up.ml/

License: MIT License

Dockerfile 0.72% Makefile 3.52% Shell 0.30% Python 50.17% Jupyter Notebook 2.65% CSS 7.24% HTML 35.40%

tensorflow tensorflow-tool tensorflow2 gde poweredbytf

tf2up's Introduction

tf2up.ml - TensorFlow 2.0 upgrader service

Idea is to make upgrade process to TensorFlow 2.0 of your Jupyter notebooks even easier!

Bookmarklet

javascript:(function(){ window.open(window.location.toString().replace(/.*github\.com\/(.*)/, 'http://tf2up.ml/\$1')); })()

Demo

How it works

This project is standing on the shoulders of giants:

TensorFlow's tf_upgrade_v2 with brand new support of inline jupyter notebook conversion
nbdime for generating nice diffs for before and after
Kubernetes + Helm for deploying this tiny service

Contributing

In order to run locally:

Main part: we need to build a docker image and run with local NBDIME_URL

make build -e

Start nbdime container in a separate terminal

make nbbuild nbrun

Now you can run the docker image with

make run

Kudos

Thanks to all amazing people, that in one or another way helped this project:

@DynamicWebPaige for supporting and making TF community awesome
Daria Korkuna for creating the logo
Jun Santos for frontend revamp
Martin Wicke for TF2, cake, and feedback
Jerry Kurata for testing the early versions of the upgrader

Citation

@misc{tf2up,
    author = {Sergii Khomenko},
    title = {tf2up.ml: TensorFlow 2.0 upgrader service},
    howpublished = {\url{https://github.com/lc0/tf2up}}
}

GitHub @lc0 · Twitter @lc0d3r · Code and Gradients

tf2up's People

Contributors

Stargazers

Watchers

Forkers

johndpope mhaghighat igsp7 nandu-h mommoren bootrewy criskgr

tf2up's Issues

Bring back health check for containers

Ensure proper render for mobile

Bunch of people are going to open this first on mobile.
We need to ensure, that they do not get lost

simplify run instructions by using as default NBDIME_URL=http://host.docker.internal:8081/d/

handle multi-cell bash magic

Sometimes users uses something like

%bash

echo 'something'
id

Example of notebook, that has many if such examples is https://github.com/tarrade/proj_DL_models_and_pipelines_with_GCP/blob/master/notebook/08-Mnist_keras_estimator.ipynb

Force refresh and give user an info about it

Currently we cache such requests and serve from local cache.

if user would refresh a file, we are not going to see this.

We be more explicit on this, like adding FAQ section.
Additionally, we can add a param for URL.

Setup CI

CI/CD
basic python lint to makefile
run pylint test as a part of CI
add pytest to run tests
install package in order to simplify imports https://github.com/lc0/tf2up/pull/33/files#diff-0c81b9c94c68ea34c497c1fa52daee56R17

handle line splits with `\`

Improve UI/UX

Mostly about fixing CSS and make it a bit more consistent across the website

Might also include #7

Update `nbdime`, given the merged fix, that should resolve the bug

details:

jupyter/nbdime#457

Add links to upgrade errors

Idea is to create a small page for the most common errors, so users can find some suggested way to go around this issue

/notebooks/9aa78c878d63a515c666342dff10ed47/original.ipynb:115:16: ERROR: Using member tf.contrib.image.rotate in deprecated module tf.contrib. tf.contrib.image.rotate cannot be converted automatically. tf.contrib will not be distributed with TensorFlow 2.0, please consider an alternative in non-contrib TensorFlow, a community-maintained repository, or fork the required code.