Skip to content

Conversation

@danielzgtg
Copy link
Contributor

@danielzgtg danielzgtg commented Aug 2, 2025

Tested working with kokoro -t Hello -o test.wav. This is needed for Ubuntu 25.04.

See also: hexgrad/kokoro#244

Tested working with `kokoro -t Hello -o test.wav`. This is needed for
Ubuntu 25.04.
@unicodeveloper
Copy link

Any idea when this will be merged in?

@danielzgtg
Copy link
Contributor Author

@hexgrad This misaki PR should be merged so that the identically-named kokoro PR you merged can start having effect.

@hexgrad hexgrad merged commit fba1236 into hexgrad:main Aug 11, 2025
@cclauss
Copy link

cclauss commented Aug 30, 2025

@danielzgtg
Copy link
Contributor Author

danielzgtg commented Aug 30, 2025

That cython-blis PR is not evidence that "Blis seems to be not yet ready for Py3.13". The last part that "Changes accepted Python versions to <3.15" merely enables Python 3.13 testing. Unlike this PR, cython-blis already had 3.13 enabled.

The first half of that cython-blis PR adds "free-threading support". According to https://docs.python.org/3/howto/free-threading-python.html , "The free-threaded mode is experimental and work is ongoing to improve it: expect some bugs and a substantial single-threaded performance hit." Most machine learning people aren't enabling it in production, if not, at all. They are using the "the default GIL-enabled build". Even if I wanted to use free-threading, I couldn't find the build for Ubuntu, or upstream in Debian. Most importantly, since cython-blis is declaring incompatibility with free-threading, so my this PR will allow installation on normal Python 3.13, while I think the cython-blis dependency will block installation when using the experimental option.

@cclauss
Copy link

cclauss commented Aug 30, 2025

@danielzgtg
Copy link
Contributor Author

That is because something is pinning Blis to 0.7: https://github.com/cclauss/misaki/actions/runs/17343241531/job/49239886842#step:5:237 I will send a PR here fixing this via spaCy in a few minutes.

danielzgtg added a commit to danielzgtg/misaki that referenced this pull request Aug 30, 2025
spaCy 4 hasn't been released. 4.0.0.dev{1,2} have been yanked,
questioning the stability. More importantly, 4.0.0.dev3 was published
2024-04-22 while the normal 3.8.7 is newer, being published 2025-05-23,
a whole year later. The outdated 2024 spaCy depends on an outdated
thinc, which depends on an outdated blis whose build fails on Python
3.13.

I fixed this by restricting spaCy to stable 3.x versions. The spaCy 4
apparently wasn't even being used in the final installation that Kokoro
sees, somehow only appearing in a misconfigured build step.

Fixes: hexgrad#85 (comment)
Reported-by: @cclauss
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants