Tensorflow 1.5.0 release canidate includes “eager execution” | Hacker News Reader

Tensorflow 1.5.0 release canidate includes “eager execution”

github.com

42 pointssupermdguy12 comments

Good to see it, been running tf-nightly, cuda 9.1 nvidia gpu drivers with cuda 8. The entire dependency issues between drivers/apps/libraries have been incredibly annoying.

Some Linux distro package modern versions of tensorflow, cuda, cudnn and nvidia which all work in tandem which makes this a whole lot less annoying to deal with.

Good news for me: "Starting from 1.6 release, our prebuilt binaries will use AVX instructions. This may break TF on older CPUs."

I just this morning gave up trying to compile Tensorflow from sources with AVX. I'm sure I could do it given the time but after 5 minutes it was clearly a more complicated project. Shame it may break older CPUs though, I wonder what the challenge is in releasing two versions?

Here is how I build architecture optimized TensorFlow 1.4 binaries for specific AWS instance types:

https://gist.github.com/daniel-j-h/b410dd21d4949a54c83d0c386...

The build script needs slight adaptions between major TensorFlow releases; otherwise works great for me so far.

During boot time you can then figure out the instance architecture you are and fetch the appropriate wheel. Hope it helps.

What is eager execution?

https://github.com/tensorflow/tensorflow/tree/r1.5/tensorflo...

> Eager execution is an experimental interface to TensorFlow that provides an imperative programming style (à la NumPy). When you enable eager execution, TensorFlow operations execute immediately; you do not execute a pre-constructed graph with Session.run().

This sounds like it would make mucking about with TensorFlow in the REPL far easier.

And debuging. Usually with tensorflow project I found my self at debuging hell where I really dont have any clue where the buggy line is.

This is great! No more constant `eval` spam :)

Eager execution is an experimental interface to TensorFlow that provides an imperative programming style (à la NumPy). When you enable eager execution, TensorFlow operations execute immediately; you do not execute a pre-constructed graph with Session.run().

https://github.com/tensorflow/tensorflow/tree/r1.5/tensorflo...

I definitely recommend reading the documentation for that info: https://github.com/tensorflow/tensorflow/tree/r1.5/tensorflo...

It's actually pretty clear (to anyone that has used TensorFlow)

Here’s the announcement from October, with examples and links to more documentation. It feels like a response to PyTorch.

https://research.googleblog.com/2017/10/eager-execution-impe...

Reply on news.ycombinator.com