One more 0.5.x release?

[hauntsaninja]

Given 0.6 has several changes (see #798 (comment) ), it would be nice if magika could make another 0.5 release including #591

[reyammer]

Hello @hauntsaninja, (happy new year!). That's a good idea. We may have a new (faster) model for 0.6.x so we may postpone a bit more to allow for more testing.

Any other changes you think would be good to include in a new 0.5.x release?

[reyammer]

More thoughts on this: I just remembered that I did try to fix #591 for the 0.5.x versions, but at the time we were still relying on poetry, and poetry had issues with it (see the discussion in #591 itself). So, unclear what would be beneficial to include in another 0.5.x version. And it would be relatively tricky, as we would need to pick up again whatever release pipeline we had pre poetry => uv migration, with likely many details I have forgotten about.

That being said, we mainly heard two concerns about the new 0.6.x version:

1. increased inference time. standard_v2_1 was indeed slower than standard_v1. However, we made progress on this aspect, and the upcoming standard_v3_0 should be even faster than standard_v1 (see Start using standard_v3_0 in rust CLI & test #868 and Start using standard_v3_0 in rust CLI & test #868 for more context).

2. the python package is not pure python anymore. This is indeed annoying, but having an actual binary CLI (with no python-related bootstrap overhead has its significant advantages). A discussion on potential alternative is here: Discussion on having a source distribution that is pure python #870.

Any feedback is appreciated!

[reyammer]

Hello again!

we just released a new -rc version, which other than shipping the latest rust client and the last model (which should be ~3x faster than standard_v2_1 and 20% faster than standard_v1), is also available as a pure-python wheel for platforms we don't support yet with rust.

https://pypi.org/project/magika/0.6.1rc0/

See details in the changelog.

You should be able to install and test with pip install --pre magika.

This should address the two concerns we heard about the new version: 1) slower inference speed (now fixed), and 2) the new wheels were not pure-python anymore, limiting the potential install base. With all this done, I believe a new version for 0.5.x should be less critical.

If you think that would still be beneficial, let us know, we are happy to work through any issues you may encounter :-)

Thanks!

[hauntsaninja]

Sorry, been a little busy. But that sounds great, thank you!