cmedina/agenx-lora-training
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: add screen support to bypass cgroup limits

Browse Source
This commit is contained in:
Christian Medina
2026-07-01 14:09:34 -04:00
parent ce758f7eb5
commit 3bd51ec7db
1 changed files with 11 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
train-on-this-server.sh
View File

@@ -80,7 +80,17 @@ echo ""
# Use torchrun for distributed training (2 GPUs) # Use torchrun for distributed training (2 GPUs)
# Add --check-only to validate without training: # Add --check-only to validate without training:
# torchrun --nproc_per_node=2 training/scripts/train.py --config training/configs/ornith-35b-lora.yaml --check-only # torchrun --nproc_per_node=2 training/scripts/train.py --config training/configs/ornith-35b-lora.yaml --check-only
torchrun --nproc_per_node=2 training/scripts/train.py --config training/configs/ornith-35b-lora.yaml
# Run in screen if available (bypasses cgroup limits)
if command -v screen &> /dev/null; then
echo "Running in screen session..."
screen -dmS lora_training bash -c "torchrun --nproc_per_node=2 training/scripts/train.py --config training/configs/ornith-35b-lora.yaml"
echo "Training started in screen session 'lora_training'"
echo "Watch with: screen -r lora_training"
else
echo "Screen not available, running directly..."
torchrun --nproc_per_node=2 training/scripts/train.py --config training/configs/ornith-35b-lora.yaml
fi
echo "" echo ""
echo "=== Training Complete ===" echo "=== Training Complete ==="