AsianPlayer/TriDrive / code /train /run_c4_evals_loop.sh
AsianPlayer's picture
download
raw
1.85 kB
#!/usr/bin/env bash
# cwm3/train/run_c4_evals_loop.sh — C4 lane driver (pre-registration docs/decisions.md 2026-09-17T19:44:28Z): every 5 min run the idempotent pairing script for the
# finished C4 runs (2 pairings in parallel, MemAvailable gate 16 GB so the trainer launches keep their 28 GB gate), until all 9 run dirs are complete OR every queue has
# ended (QUEUE DONE / EXHAUSTED in the A, B and C logs); then the final render (results/tables/latent_aux_c4.md) and the alert-timing section (--phead-runs).
set -u
cd "$(dirname "$0")/.."
PY=python
export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4
all_done() { local n=0; for v in lat01 lat03 lat0; do for s in 0 1 2; do [[ -f results/runs/I1_all_minus_road2drv_${v}_s${s}_final_ecan/PRED_DONE ]] && n=$((n+1)); done; done; [[ $n -eq 9 ]]; }
queues_ended() { for q in A B C; do lg=$(ls -t results/runs/r2_c4_${q}_*.log 2>/dev/null | head -1); [[ -n "$lg" ]] && grep -q "QUEUE DONE\|QUEUE EXHAUSTED" "$lg" || return 1; done; }
while ! all_done && ! queues_ended; do
MIN_AVAIL_GB=16 bash eval/run_joint_evals_c4.sh "0 1 2" 2 0 2>&1 | grep -vE "^skip" | tail -n 40
sleep 300
done
echo "[$(date -u +%H:%M:%SZ)] training phase ended (all_done=$(all_done && echo yes || echo no)); final pairings + render"
MIN_AVAIL_GB=12 bash eval/run_joint_evals_c4.sh "0 1 2" 3 1 2>&1 | grep -vE "^skip"
echo "[$(date -u +%H:%M:%SZ)] alert timing (--phead-runs) for the finished arms"
PH="ecan=I1_all_minus_road2drv_s{seed}_final_ecan"
for v in lat01 lat03 lat0; do ok=1; for s in 0 1 2; do [[ -f results/runs/I1_all_minus_road2drv_${v}_s${s}_final_ecan/PRED_DONE ]] || ok=0; done; [[ $ok -eq 1 ]] && PH="$PH ${v}=I1_all_minus_road2drv_${v}_s{seed}_final_ecan"; done
echo "phead-runs: $PH"
$PY eval/alert_timing_eval.py --seeds 0 1 2 --phead-runs $PH 2>&1 | tail -n 15
echo "[$(date -u +%H:%M:%SZ)] C4_LANE_EVALS_DONE"

Xet Storage Details

Size:
1.85 kB
·
Xet hash:
b120995e29953fc9e8dab6cf9de237f8859cc0048438d7b5f62157673e874c6b

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