TF Nightly: Testing Docker-Based Builds starting Nov. 29

['Austin Anderson' via TensorFlow Announcements]

Hi everyone,

On Monday the 29th, TensorFlow DevInfra is planning to start building the TF Nightly packages on PyPI with TF SIG Build's Docker images, which are published to SIG Build on DockerHub. This change will be in-place and you should not have to change your workflow at all, but it's possible that the packages will act differently.

• If you use TensorFlow regularly, we'd like your feedback on whether or not the new packages are functionally different from the previous ones.
• If you build or contribute to TensorFlow, we'd like your feedback on whether you can use the Docker images to build TensorFlow as easily as we've intended.

There are more details and sample packages on this TensorFlow Community Forum thread. Please head there to give feedback or ask questions.

Thank you,

Austin

--
You received this message because you are subscribed to the Google Groups "TensorFlow Announcements" group.
To unsubscribe from this group and stop receiving emails from it, send an email to announce+u...@tensorflow.org.
To view this discussion on the web visit https://groups.google.com/a/tensorflow.org/d/msgid/announce/CAFQYGSpHD7gzXqXYQ4PzidFo7JbDXpfkNN%2BPeqHO8vnN20yFhQ%40mail.gmail.com.

['Austin Anderson' via TensorFlow Announcements]

Hi all,

I switched TensorFlow's tf-nightly build process over last night, and the resulting tf-nightly packages were built with Docker. Please let us know if you encounter any functional differences in the packages over on this TensorFlow Community Forum thread.

Thank you,

Austin

To view this discussion on the web visit https://groups.google.com/a/tensorflow.org/d/msgid/announce/CAFQYGSpEa6PhCsMQ%3Df6soTkvkxgNh%2BSj8sFEMd9R2zTXHOUC%2BQ%40mail.gmail.com.

[Daniel Wexler]

Is there any way to reduce the size of the nightly-gpu image? The CUDA drivers are *huge* and take a long time to copy and start on a VM. See my question on this forum.