You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
xFormers is a library used by machine-learning practitioners to speed-up their workloads. It contains specific and optimised low-level components.
It has been active for 3 years now, but gained a lot of exposure within the last few years - it's a critical project in Pypi, and has ~50k downloads per day.
We would like to be able to keep ~30 major releases on pypi - aka 30GB of storage.
(note: we were previously granted 20GB of storage #2907)
How large is each release?
Each binary is ~150MB (because we ship CUDA binaries for multiple GPU architectures), and each release contains 8 binaries ({win, linux} x {cp38, cp39, cp310, cp311}). So each release is about 1GB
Binary size reduction: we're working on further reducing the binary size, but this effort depends on a modification to PyTorch which will make it to PyTorch 2.4, scheduled for September. The space limitation is currently a blocking issue that prevents further releases in the short term (unless we remove older binaries).
How frequently do you make a release?
Roughly every month, matching the PyTorch release schedule
Code of Conduct
I agree to follow the PSF Code of Conduct
The text was updated successfully, but these errors were encountered:
Project URL
https://pypi.org/project/xformers/
Does this project already exist?
New limit
30
Update issue title
Which indexes
PyPI
About the project
xFormers is a library used by machine-learning practitioners to speed-up their workloads. It contains specific and optimised low-level components.
It has been active for 3 years now, but gained a lot of exposure within the last few years - it's a critical project in Pypi, and has ~50k downloads per day.
We would like to be able to keep ~30 major releases on pypi - aka 30GB of storage.
(note: we were previously granted 20GB of storage #2907)
How large is each release?
Each binary is ~150MB (because we ship CUDA binaries for multiple GPU architectures), and each release contains 8 binaries (
{win, linux} x {cp38, cp39, cp310, cp311}
).So each release is about 1GB
Binary size reduction: we're working on further reducing the binary size, but this effort depends on a modification to PyTorch which will make it to PyTorch 2.4, scheduled for September. The space limitation is currently a blocking issue that prevents further releases in the short term (unless we remove older binaries).
How frequently do you make a release?
Roughly every month, matching the PyTorch release schedule
Code of Conduct
The text was updated successfully, but these errors were encountered: