| set -euo pipefail | |
| workspace="/mnt/pvc/users/simon/agentptb/runs/sol-max-s1/workspace" | |
| output="$workspace/outputs/sft-agent-mix-clean-full-v1" | |
| weights="$output/weights/step_100" | |
| checkpoint="$output/checkpoints/step_100/trainer/.metadata" | |
| while [[ ! -f "$weights/STABLE" || ! -f "$checkpoint" ]]; do | |
| sleep 30 | |
| done | |
| python "$workspace/scripts/check_checkpoint.py" \ | |
| "$weights" "$output/metrics.jsonl" --expected-step 100 \ | |
| >"$workspace/logs/checkpoint-health-step100.json" | |
| touch "$workspace/state/step100-healthy" | |