| # Wrapper: run job, stream full log to HF every 60s so crashes are always visible. | |
| cd /workspace | |
| up() { python -c " | |
| from huggingface_hub import HfApi | |
| a=HfApi(token='$HF_TOKEN'); a.create_repo('aixonlab/svg-slm-4b-lora',exist_ok=True) | |
| a.upload_file(path_or_fileobj='/workspace/all.log',path_in_repo='logs/all.log',repo_id='aixonlab/svg-slm-4b-lora')" >/dev/null 2>&1; } | |
| pip install -q huggingface_hub >/dev/null 2>&1 | |
| bash run_pod.sh > all.log 2>&1 & | |
| P=$! | |
| while kill -0 $P 2>/dev/null; do up; sleep 60; done | |
| echo "RUN_POD EXITED" >> all.log; up | |