scripts/launch_remote_gpu.sh

#!/usr/bin/env bash
set -euo pipefail

REMOTE=${PCMT_REMOTE:?Set PCMT_REMOTE, for example user@host}
PORT=${PCMT_REMOTE_PORT:-22}
REMOTE_DIR=${PCMT_REMOTE_DIR:-/workspace/pcmt}
REMOTE_COMMAND=${PCMT_REMOTE_COMMAND:-make run-small}
TELEMETRY_INTERVAL=${PCMT_TELEMETRY_INTERVAL:-1}

rsync -az --delete \
  --exclude '.git' \
  --exclude '.DS_Store' \
  --exclude 'old_backup_do_not_touch_stay_away' \
  --exclude 'runs' \
  --exclude 'figures' \
  ./ -e "ssh -p ${PORT} -o StrictHostKeyChecking=no" "${REMOTE}:${REMOTE_DIR}/"

ssh -p "${PORT}" -o StrictHostKeyChecking=no "${REMOTE}" "
  set -euo pipefail
  cd '${REMOTE_DIR}'
  source /venv/main/bin/activate || true
  python3 - <<'PY' >/dev/null 2>&1 || uv pip install numpy matplotlib pytest opencv-python-headless
import numpy, matplotlib, cv2
PY
  python3 - <<'PY' >/dev/null 2>&1 || uv pip install torch --index-url https://download.pytorch.org/whl/cu128
import torch
assert torch.cuda.is_available()
PY
  python3 -m pcmt.cli queue-init --path runs/gpu_jobs.jsonl
  python3 -m pcmt.cli queue-add --path runs/gpu_jobs.jsonl --name run-small-gpu --command '${REMOTE_COMMAND}'
  (nvidia-smi --query-gpu=timestamp,name,utilization.gpu,memory.used,memory.total --format=csv -l ${TELEMETRY_INTERVAL} > runs/gpu_telemetry.csv 2>&1 & echo \$! > runs/gpu_telemetry.pid)
  python3 -m pcmt.cli queue-run-next --path runs/gpu_jobs.jsonl --log-dir runs/gpu_queue_logs
  kill \$(cat runs/gpu_telemetry.pid) 2>/dev/null || true
  tail -n 5 runs/gpu_telemetry.csv || true
"

mkdir -p runs/remote_gpu
rsync -az -e "ssh -p ${PORT} -o StrictHostKeyChecking=no" "${REMOTE}:${REMOTE_DIR}/runs/" runs/remote_gpu/