PhotoPrism Installation Packages about photoprism-debian HOT 7 CLOSED

Thanks for the heads-up! That's good news, I'll test it out and update the guide.

I'm assuming the builds include the "regular" (non-AVX) version of Tensorflow?

Do you have plans to publish the builds on Github? That would make linking to the most recent build more convenient.

from photoprism-debian.

Understood, thanks for the information!

from photoprism-debian.

Oh, that's a great question! Since I've built this on a modern AMD Ryzen CPU, it could be that the AVX2 TensorFlow libraries are bundled. However, the performance difference to the non-AVX version is not big, so we could bundle that instead. Which do you prefer? We will add the packages to the next stable release on GitHub, since we created the current packages after the Docker release (so the build number does not match).

from photoprism-debian.

Note that you can also manually replace the TensorFlow libraries in /opt/photoprism/lib/ with a different version.

from photoprism-debian.

Since I've built this on a modern AMD Ryzen CPU, it could be that the AVX2 TensorFlow libraries are bundled. However, the performance difference to the non-AVX version is not big, so we could bundle that instead. Which do you prefer?

I've seen issues where people are running PhotoPrism on an older CPU that doesn't support AVX2, so probably the non-AVX version would be the safest bet, especially if there's not a lot of difference in terms of performance.

FWIW, if you're planning to use Github Actions for the builds, then currently the non-AVX version should get bundled (though that might change if Github moves their runners to a newer architecture). I know because I've taken a stab at building PhotoPrism for Linux that way. I'll stop that, of course, now that you're publishing official builds.

from photoprism-debian.

I've seen issues where people are running PhotoPrism on an older CPU that doesn't support AVX2, so probably the non-AVX version would be the safest bet, especially if there's not a lot of difference in terms of performance.

I just checked our build scripts to confirm that the standard TensorFlow libraries (without AVX) are bundled. The install-tensorflow.sh script tests CPU support for AVX/AVX2 only if you run it with "auto" as argument, which we do not do in this case.

FWIW, if you're planning to use Github Actions for the builds, then currently the non-AVX version should get bundled (though that might change if Github moves their runners to a newer architecture).

Due to previous security incidents related to GitHub and running a public CI infrastructure, we currently use a bare-metal server with no direct connection to GitHub to create binaries and run tests (with the exception of the docs repository, which uses GitHub Actions to generate HTML from Markdown). This also allows us to perform manual testing before an official release.

from photoprism-debian.

The installation package docs have been added to our public repository, so contributors can send pull requests with improvements:

• https://github.com/photoprism/photoprism/tree/develop/setup/pkg/linux

from photoprism-debian.