#!/usr/bin/env bash
# Re-occupy GPU3: relaunch the JSL-Med-9B-v3.2 vLLM server (port 8010).
set -uo pipefail
SERVE_SCRIPT=/mnt/data/zainulabideen/aws/serve_jsl_9b_v3_2.sh
LOG=/mnt/data/zainulabideen/Data/RV3/fine_tune/logs/serve_jsl_9b_v3_2_resume.log
echo ">> checking GPU3 is free before starting..."
USED=$(nvidia-smi -i 3 --query-gpu=memory.used --format=csv,noheader,nounits)
if [ "$USED" -gt 5000 ]; then
  echo ">> GPU3 has ${USED} MiB in use — someone else may be using it. Run free-gpu3.sh first if that's stale."
  exit 1
fi
echo ">> launching JSL-Med-9B-v3.2 on GPU3, port 8010..."
nohup bash "$SERVE_SCRIPT" > "$LOG" 2>&1 &
disown
echo ">> launched, PID $!"
echo ">> waiting for health check..."
for i in $(seq 1 40); do
  r=$(curl -s -o /dev/null -w "%{http_code}" http://localhost:8010/v1/models -H "Authorization: Bearer jsl-1" 2>/dev/null)
  if [ "$r" == "200" ]; then
    echo ">> HEALTHY after ${i}0s"
    exit 0
  fi
  sleep 10
done
echo ">> WARNING: not healthy after 400s, check $LOG"