| set -euo pipefail | |
| if [ $# -ne 1 ]; then | |
| echo "Usage: bash launch_and_log.sh <train_script>" | |
| exit 1 | |
| fi | |
| TRAIN_SCRIPT="$1" | |
| RUN_ID="$(date +%Y%m%d_%H%M%S)" | |
| RUN_DIR="/raid/data/weifeng/Pre_train/swift_runs/runs/${RUN_ID}" | |
| if command -v python >/dev/null 2>&1; then | |
| PYTHON_BIN=python | |
| elif command -v python3 >/dev/null 2>&1; then | |
| PYTHON_BIN=python3 | |
| else | |
| echo "Error: neither python nor python3 was found in PATH." | |
| exit 1 | |
| fi | |
| mkdir -p "${RUN_DIR}" | |
| cp "${TRAIN_SCRIPT}" "${RUN_DIR}/command.sh" | |
| env | sort > "${RUN_DIR}/env.txt" | |
| nvidia-smi > "${RUN_DIR}/nvidia-smi.txt" | |
| "${PYTHON_BIN}" -m pip freeze > "${RUN_DIR}/pip_freeze.txt" | |
| echo "Run directory: ${RUN_DIR}" | |
| echo "Logging to: ${RUN_DIR}/train.log" | |
| bash "${TRAIN_SCRIPT}" 2>&1 | tee "${RUN_DIR}/train.log" | |