| # After the Gemma thinking-ON queue on this GPU finishes, resume the no-limit rerun jobs of CFG. | |
| # The trials lost to the 2026-10-30 disk-full crash were moved to jobs/_diskfull_1030, so resume reruns them. | |
| # Usage: redo_diskfull.sh GPU PORT CFG | |
| GPU=$1; PORT=$2; CFG=$3; LOG=~/tbench/gemma-tbt-gpu$GPU.log | |
| until grep -q "GEMMA_TB_DONE GPU$GPU" $LOG; do sleep 120; done | |
| source ~/tbench/gemma_env.sh | |
| echo "===== redo disk-full trials $CFG $(date -u)" >> $LOG | |
| SRC=tbgt21k3 DST=tbgt21k3nolim ~/tbench/tb_rerun_timeouts.sh $GPU $PORT $CFG >> $LOG 2>&1 | |
| echo "REDO_DONE $CFG $(date -u)" >> $LOG | |