choderalab/pymbar: 4.2.0 Disable JAX Acceleration with an Environmental Variable
Authors/Creators
- Michael Shirts1
- Kyle Beauchamp
- Levi Naden2
- John Chodera3
- jaimergp4
- Mike Henry5
- Stefano Martiniani6
- Chaya Stern7
- Matt Thompson8
- badisa
- Jennifer A Clark9
- Josh Fass
- Richard Gowers
- Robert T. McGibbon
- Bradley Dice10
- David L. Dotson11
- Irfan Alibay12
- Iván Pulido3
- Kazuya Ujihara
- Marc Schuh
- Tucker "Emme" Burgin
- 1. University of Colorado Boulder
- 2. @MolSSI at Virginia Tech
- 3. Memorial Sloan Kettering Cancer Center
- 4. @quansight
- 5. @choderalab & @OpenFreeEnergy
- 6. New York University
- 7. @Kekulai
- 8. Open Molecular Software Foundation
- 9. @openforcefield @omsf
- 10. @NVIDIA @rapidsai
- 11. Datryllic LLC
- 12. The Open Molecular Software Foundation
Description
This release allows a user to set PYMBAR_DISABLE_JAX=TRUE to disable JAX acceleration.
This enables disabling JAX at runtime.
Previously, the only way to control JAX was to install or uninstall the jax package, which made it difficult for users to actually disable JAX since some packages depended on the pymbar package which on conda-forge installs the jax package.
See the README.md for more details.
We will be improving accelerator selection in future releases.
conda-forge Package Notes
Previous versions of pymbar pinned numpy < 2.3 due to a change in numpy 2.3 where glibc implementations of some math functions are used where previously numpy's built in implementations were used. See https://github.com/choderalab/pymbar/issues/556 and https://github.com/alchemistry/alchemlyb/pull/431#pullrequestreview-2933773722 for some more details.
However, this pin prevents pymbar from working with python 3.14 since there are no numpy <2.3 builds that work with python 3.14. So we have decided to relax the numpy pin. Within our own testing framework we have not observed any differences between numpy version. While there are numerical differences, for our tests, those differences are within our tolerance. You can still install this version of pymbar with an older numpy version, it just won't work with python 3.14. For example conda install -c conda-forge "pymbar==4.2.0" "numpy<2.3" "python=3.13" will work.
Changelog for robots :point_down:
What's Changed
- switch from master to main by @mikemhenry in https://github.com/choderalab/pymbar/pull/567
- update README to reflect change in default branch name by @mikemhenry in https://github.com/choderalab/pymbar/pull/568
- disable jax if PYMBAR_DISABLE_JAX is set by @mikemhenry in https://github.com/choderalab/pymbar/pull/569
Full Changelog: https://github.com/choderalab/pymbar/compare/4.1.0...4.2.0
Files
choderalab/pymbar-4.2.0.zip
Files
(17.4 MB)
| Name | Size | Download all |
|---|---|---|
|
md5:863aa3445023d265902fad6b39739783
|
17.4 MB | Preview Download |
Additional details
Related works
- Is supplement to
- Software: https://github.com/choderalab/pymbar/tree/4.2.0 (URL)
Software
- Repository URL
- https://github.com/choderalab/pymbar