From 2eb7665960bb939fca63753ea796ec8cd5ae8b7c Mon Sep 17 00:00:00 2001
From: janEbert <janpublicebert@posteo.net>
Date: Tue, 9 Jul 2024 10:04:50 +0200
Subject: [PATCH] Integrate script updates into README
---
README.md | 8 +++++---
1 file changed, 5 insertions(+), 3 deletions(-)
diff --git a/README.md b/README.md
index 9205aef..2073c67 100644
--- a/README.md
+++ b/README.md
@@ -262,9 +262,10 @@ use
Instead of
```shell
-srun python -u -m torchrun_jsc \
- --nproc_per_node="$DEVICES_PER_NODE" \
+srun env -u CUDA_VISIBLE_DEVICES python -u -m torchrun_jsc \
+ --nproc_per_node=gpu \
--nnodes="$SLURM_JOB_NUM_NODES" \
+ --rdzv_id="$SLURM_JOB_ID" \
--rdzv_endpoint="$MASTER_ADDR":"$MASTER_PORT" \
--rdzv_backend=c10d \
"$curr_dir"/main.py "$@"
@@ -273,7 +274,7 @@ srun python -u -m torchrun_jsc \
use
```shell
-srun python -u "$curr_dir"/main.py "$@"
+srun env -u CUDA_VISIBLE_DEVICES python -u "$curr_dir"/main.py "$@"
```
Additionally, if using PyTorch Lightning, you may encounter issues
@@ -301,6 +302,7 @@ def patch_lightning_slurm_master_addr():
'juwelsbooster',
'juwels',
'jurecadc',
+ 'jusuf',
]:
return
--
GitLab