make wfl pool autoparallelization torch GPU ID aware - #363

Merged

bernstei merged 13 commits into

mainfrom

Nov 3, 2025

Commits on Oct 28, 2025

Enable assigning of specific GPUs to each autopara python subprocess with WFL_TORCH_DEVICE_IS_SUBPROCESS_ID
bernstei
committed
41bd2be
Switch all tests to ase LangevinBOAB, and fix missing gpu_id_queue when not running in pool. Also, workaround opt.converged ASE issue 1744
bernstei
committed
fa89458
Refactor clean_rundir and related
bernstei
committed
ab9e460

Commits on Oct 29, 2025

typo
bernstei
committed
409456e
Don't use forkserver for autopara pool unless WFL_TORCH_N_GPUS is set
bernstei
committed
c545aac
Fix convergence check in optimizers, and skip phono3py tests if not available
bernstei
committed
0c8222e
Fix autopara GPU test
bernstei
committed
1a2f51b
lint
bernstei
committed
8946a61
Back to opt.converged(None) for wfl.generate.optimize, since PreconLBFGS supports that but doesn't fully support irun
bernstei
committed
37499b2
test for wfl.calculator.utils.clean_rundir
bernstei
committed
f0d55e9
fix LangevinBAOAB import in test_md.py
bernstei
committed
4de5eab
Cleaner use of 'converged = opt.run()' in optimize and neb
bernstei
committed
b69e76f

Commits on Oct 30, 2025

Update expected number of pass/fail/etc in complete_pytest.tin
bernstei
committed
353a22d