Skip to content

fix: correct environment setup in container SLURM template #22

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
Neonkraft merged 4 commits into from
Apr 30, 2026
Merged

fix: correct environment setup in container SLURM template #22

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
f65cdf6
fix: remove CUDA_DEVICE_MAX_CONNECTIONS, fix PYTHONPATH, add WANDB_DI…
Neonkraft Apr 28, 2026
e50ee7e
Merge branch 'main' into fix/slurm-container-template
Neonkraft Apr 29, 2026
7bd7efa
fix: set PYTHONPATH to repo src/ in LlamaFactory container template
Neonkraft Apr 30, 2026
f077f78
fix: resolve run_dir to absolute path in container SLURM templates
Neonkraft Apr 30, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion src/post_training/slurm/job_llamafactory.sh.jinja
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -70,7 +70,7 @@ srun --export=ALL --wait=60 --kill-on-bad-exit=1 \

# Prevent host Python/PATH from interfering with container
export PATH=\"/usr/local/bin:/usr/bin:/bin\"
export PYTHONPATH=\"\"
export PYTHONPATH=\"{{ repo_dir }}/src\"
export PYTHONNOUSERSITE=1

export NODE_RANK=\"\$SLURM_NODEID\"
Expand Down
7 changes: 5 additions & 2 deletions src/post_training/slurm/job_trl_container.sh.jinja
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -35,7 +35,8 @@ export WORLD_SIZE=$(( SLURM_NNODES * GPUS_PER_NODE ))
# NCCL tuning for multi-node stability
export NCCL_IB_TIMEOUT=120
export NCCL_DEBUG=INFO
export CUDA_DEVICE_MAX_CONNECTIONS=1
# CUDA_DEVICE_MAX_CONNECTIONS=1 is a Megatron-LM tensor-parallel flag that
# serializes CUDA streams and hurts ZeRO-2 overlap_comm — do not set it.
export OMP_NUM_THREADS=1

echo "=========================================="
Expand Down Expand Up @@ -71,7 +72,7 @@ srun --export=ALL --wait=60 --kill-on-bad-exit=1 \

# Prevent host Python/PATH from interfering with container
export PATH=\"/usr/local/bin:/usr/bin:/bin\"
export PYTHONPATH=\"\"
export PYTHONPATH=\"{{ repo_dir }}/src\"
export PYTHONNOUSERSITE=1

export NODE_RANK=\"\$SLURM_NODEID\"
Expand All @@ -87,6 +88,8 @@ srun --export=ALL --wait=60 --kill-on-bad-exit=1 \

cd {{ repo_dir }}

export WANDB_DIR={{ run_dir }}

accelerate launch \
--num_machines \$NNODES \
--num_processes $WORLD_SIZE \
Expand Down
4 changes: 2 additions & 2 deletions src/post_training/slurm/launcher.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -114,7 +114,7 @@ def render_trl_container_slurm_script(
wall_time=config.slurm.wall_time,
signal_time_seconds=config.slurm.signal_time_seconds,
max_failures=config.slurm.max_failures,
run_dir=str(run_dir),
run_dir=str(run_dir.resolve()),
config_path=config_path,
# Accelerate flags
mixed_precision=config.accelerate.mixed_precision,
Expand Down Expand Up @@ -162,7 +162,7 @@ def render_llamafactory_slurm_script(
wall_time=config.slurm.wall_time,
signal_time_seconds=config.slurm.signal_time_seconds,
max_failures=config.slurm.max_failures,
run_dir=str(run_dir),
run_dir=str(run_dir.resolve()),
# Container
container_image=config.container.image,
bind_mounts=config.container.bind_mounts,
Expand Down
Loading