AsianPlayer/TriDrive / code /train /run_f1_joint_after_kpf.sh
AsianPlayer's picture
download
raw
1.21 kB
#!/usr/bin/env bash
# cwm3/train/run_f1_joint_after_kpf.sh — LANE F1 (2026-09-17; pre-registration docs/decisions.md 2026-09-17T19:33:12Z): wait until the three seed-matched
# forecasters results/runs/KPF_mb_full_s{0,1,2} have PRED_DONE, then run the joint distillation queue train/queue_f1_joint.txt through the gated
# train/run_queue_cmds.sh (28 GB RAM available, 14 GB VRAM free, < 4 trainers, disk >= 60 GB). Polls every 60 s; gives up after --max-wait-min (default 240).
# setsid nohup bash train/run_f1_joint_after_kpf.sh [max_wait_min] > results/logs/f1_joint_waiter.out 2>&1 &
set -u
cd "$(dirname "$0")/.."
MAXW="${1:-240}"
t0=$(date +%s)
while true; do
ok=1
for s in 0 1 2; do [[ -f results/runs/KPF_mb_full_s${s}/PRED_DONE ]] || ok=0; done
[[ $ok -eq 1 ]] && break
if (( ($(date +%s) - t0) / 60 > MAXW )); then echo "[$(date +%H:%M:%S)] KPF_mb_full runs not finished after ${MAXW} min; giving up"; exit 2; fi
echo "[$(date +%H:%M:%S)] waiting for KPF_mb_full_s{0,1,2}/PRED_DONE"; sleep 60
done
echo "[$(date +%H:%M:%S)] forecasters ready; starting the joint queue"
bash train/run_queue_cmds.sh f1_joint train/queue_f1_joint.txt 28 14 3
echo "[$(date +%H:%M:%S)] F1_JOINT_QUEUE_EXIT $?"

Xet Storage Details

Size:
1.21 kB
·
Xet hash:
024c238ed00cb1a5279c8522d65e183a581d699cad65cbda47852ad1a3af9e60

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.