From c9d60c656c01059a9b10523ad66ddb80e9e0c8ab Mon Sep 17 00:00:00 2001 From: Thomas Roeblitz Date: Tue, 20 May 2025 10:45:57 +0200 Subject: [PATCH 1/3] {2023.06}[2023a,a64fx] PyTorch 2.1.2 originally built with EB 4.9.0 --- .../2023.06/a64fx/eessi-2023.06-eb-4.9.4-2023a.yml | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-) diff --git a/easystacks/software.eessi.io/2023.06/a64fx/eessi-2023.06-eb-4.9.4-2023a.yml b/easystacks/software.eessi.io/2023.06/a64fx/eessi-2023.06-eb-4.9.4-2023a.yml index a1dfd4a929..e9389b2211 100644 --- a/easystacks/software.eessi.io/2023.06/a64fx/eessi-2023.06-eb-4.9.4-2023a.yml +++ b/easystacks/software.eessi.io/2023.06/a64fx/eessi-2023.06-eb-4.9.4-2023a.yml @@ -170,12 +170,12 @@ easyconfigs: # # see https://github.com/easybuilders/easybuild-easyblocks/pull/3036 # include-easyblocks-from-pr: 3036 - LAMMPS-2Aug2023_update2-foss-2023a-kokkos.eb -## PR 19573 was included since EB 4.9.1 -## - PyTorch-2.1.2-foss-2023a.eb: -## options: -## # see https://github.com/easybuilders/easybuild-easyconfigs/pull/19573 -## from-pr: 19573 -# - PyTorch-2.1.2-foss-2023a.eb +# PR 19573 was included since EB 4.9.1 +# - PyTorch-2.1.2-foss-2023a.eb: +# options: +# # see https://github.com/easybuilders/easybuild-easyconfigs/pull/19573 +# from-pr: 19573 + - PyTorch-2.1.2-foss-2023a.eb - matplotlib-3.7.2-gfbf-2023a.eb # PR 19554 was included since EB 4.9.1 # - PyQt5-5.15.10-GCCcore-12.3.0.eb: From 2cd3d36122db544be98eb0dbee8037f9f8ed0a9a Mon Sep 17 00:00:00 2001 From: Thomas Roeblitz Date: Thu, 22 May 2025 10:05:48 +0200 Subject: [PATCH 2/3] use a quarter of the cores when building PyTorch --- eb_hooks.py | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) diff --git a/eb_hooks.py b/eb_hooks.py index adf4600daa..49c4b9f821 100644 --- a/eb_hooks.py +++ b/eb_hooks.py @@ -1248,7 +1248,7 @@ def post_module_hook(self, *args, **kwargs): } POST_MODULE_HOOKS = {} - +Torch # Define parallelism limit operations def divide_by_factor(parallel, factor): """Divide parallelism by given factor""" @@ -1272,6 +1272,9 @@ def set_maximum(parallel, max_value): 'MBX': { '*': (divide_by_factor, 2), }, + 'PyTorch': { + CPU_TARGET_A64FX: (divide_by_factor, 4), + }, 'TensorFlow': { '*': (divide_by_factor, 2), CPU_TARGET_A64FX: (set_maximum, 8), From ec5bec5ee6565549d93445b0af34af69180acf06 Mon Sep 17 00:00:00 2001 From: Thomas Roeblitz Date: Fri, 23 May 2025 12:46:12 +0200 Subject: [PATCH 3/3] fix error and pull in upstream changes --- eb_hooks.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/eb_hooks.py b/eb_hooks.py index d6a4a80d85..b2eb2fdc86 100644 --- a/eb_hooks.py +++ b/eb_hooks.py @@ -1252,7 +1252,7 @@ def post_module_hook(self, *args, **kwargs): } POST_MODULE_HOOKS = {} -Torch + # Define parallelism limit operations def divide_by_factor(parallel, factor): """Divide parallelism by given factor"""