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

fix: remove screen, run training directly

Browse Source
This commit is contained in:
Christian Medina
2026-07-01 14:16:05 -04:00
parent 3bd51ec7db
commit 4287e42148
1 changed files with 2 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
train-on-this-server.sh
View File

@@ -78,19 +78,8 @@ echo "Press Ctrl+C to stop (model will be saved at checkpoint)."
echo "" echo ""
# Use torchrun for distributed training (2 GPUs) # Use torchrun for distributed training (2 GPUs)
# Add --check-only to validate without training: echo "Starting training..."
# torchrun --nproc_per_node=2 training/scripts/train.py --config training/configs/ornith-35b-lora.yaml --check-only
# 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 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 ==="