File size: 5,251 Bytes
8aa2acf
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
#!/usr/bin/env bash
set -uo pipefail
# ─────────────────────────────────────────────────────────────────────────────
# run_all_groot.sh — evaluate ALL 10 GR00T conflict checkpoints.
#
# One checkpoint per GPU: for each category we start a dedicated GR00T zmq
# inference server (gr00t venv) pinned to one GPU, then run the ManiSkill OOD
# sweep (ms venv) on the SAME GPU for GPU physics+render, then stop the server.
# Up to NUM_GPUS checkpoints run concurrently (processed in waves).
#
# Each checkpoint is evaluated on its OWN conflict experiment (category name).
# Every episode video is saved under
#   ${OUT_ROOT}/<category>/experiments/<run_name>/video/ep000{,_wrist}.mp4
#
# Usage:
#   bash run_all_groot.sh [seed] [total_episodes]
# Defaults: seed=42  total_episodes=200  (→ 400 runs per checkpoint)
# Override episode count: bash run_all_groot.sh 42 20
# ─────────────────────────────────────────────────────────────────────────────

SEED="${1:-42}"
TOTAL_EPISODES="${2:-200}"

ROOT=/workspace/groot_eval
HARNESS="${ROOT}/harness"
CKPT_ROOT="${ROOT}/checkpoints"
OUT_ROOT="${OUT_ROOT:-${ROOT}/results}"
LOG_DIR="${ROOT}/logs/run_$(date +%Y%m%d_%H%M%S)"
GROOT_VENV_PY="${ROOT}/gr00t_repo/codebase/.venv/bin/python"
GROOT_CODE="${ROOT}/gr00t_repo/codebase"
MS_PY="${ROOT}/.venv_ms/bin/python"
NUM_GPUS="${NUM_GPUS:-8}"
SIM_BACKEND="${SIM_BACKEND:-gpu}"
BASE_PORT="${BASE_PORT:-5600}"

CATEGORIES=(
  color_object color_size color_spatial size_object spatial_object
  spatial_size verb_color verb_object verb_size verb_spatial
)

mkdir -p "${LOG_DIR}" "${OUT_ROOT}"
echo "seed=${SEED} total_episodes=${TOTAL_EPISODES} sim_backend=${SIM_BACKEND}"
echo "logs: ${LOG_DIR}"
echo "results/videos: ${OUT_ROOT}"

run_one() {
  local cat="$1" gpu="$2" port="$3"
  local ckpt="${CKPT_ROOT}/${cat}"
  local slog="${LOG_DIR}/server_${cat}.log"
  local clog="${LOG_DIR}/client_${cat}.log"
  local exp_root="${OUT_ROOT}/${cat}/experiments"
  local results_txt="${OUT_ROOT}/${cat}/${cat}_ood_seed${SEED}.txt"
  mkdir -p "${exp_root}" "$(dirname "${results_txt}")"

  echo "[$(date +%H:%M:%S)] START ${cat}  gpu=${gpu} port=${port}"

  # ── Start GR00T inference server on this GPU ──
  ( cd "${GROOT_CODE}" && CUDA_VISIBLE_DEVICES="${gpu}" \
      HF_HOME="${ROOT}/.hf_cache" HF_TOKEN="$(cat ${ROOT}/.hf_token)" \
      NO_ALBUMENTATIONS_UPDATE=1 TOKENIZERS_PARALLELISM=false \
      "${GROOT_VENV_PY}" -m gr00t.eval.run_gr00t_server \
        --model-path "${ckpt}" \
        --embodiment-tag new_embodiment \
        --device cuda:0 \
        --host 127.0.0.1 \
        --port "${port}" ) > "${slog}" 2>&1 &
  local server_pid=$!

  # ── Wait until the server is listening (model load can take ~1-2 min) ──
  local ready=0
  for _ in $(seq 1 180); do
    if ! kill -0 "${server_pid}" 2>/dev/null; then
      echo "[${cat}] SERVER DIED during startup — see ${slog}"
      break
    fi
    if grep -q "Server ready\|Server is ready and listening" "${slog}" 2>/dev/null; then
      ready=1; break
    fi
    sleep 3
  done
  if [[ "${ready}" -ne 1 ]]; then
    echo "[${cat}] server not ready; killing. tail server log:"
    tail -n 20 "${slog}" || true
    kill "${server_pid}" 2>/dev/null || true
    wait "${server_pid}" 2>/dev/null || true
    return 1
  fi
  echo "[$(date +%H:%M:%S)] [${cat}] server ready (pid ${server_pid}), starting sweep"

  # ── Run the OOD sweep on the SAME GPU (ManiSkill GPU physics+render) ──
  CUDA_VISIBLE_DEVICES="${gpu}" \
  HOST=127.0.0.1 PORT="${port}" SIM_BACKEND="${SIM_BACKEND}" \
  EXPERIMENT_ROOT="${exp_root}" GROOT_MAIN="${HARNESS}/groot_main.py" MS_PY="${MS_PY}" \
  bash "${HARNESS}/run_ood_groot_inference.sh" \
      "${cat}" "${SEED}" "${TOTAL_EPISODES}" "${results_txt}" \
      > "${clog}" 2>&1
  local rc=$?

  # ── Stop the server ──
  kill "${server_pid}" 2>/dev/null || true
  wait "${server_pid}" 2>/dev/null || true
  echo "[$(date +%H:%M:%S)] DONE ${cat} rc=${rc}  results=${results_txt}"
  return ${rc}
}

# ── Process in waves of NUM_GPUS (one checkpoint per GPU) ──
launched=0
pids=()
for idx in "${!CATEGORIES[@]}"; do
  cat="${CATEGORIES[$idx]}"
  gpu=$(( launched % NUM_GPUS ))
  port=$(( BASE_PORT + gpu ))
  run_one "${cat}" "${gpu}" "${port}" &
  pids+=($!)
  launched=$(( launched + 1 ))
  if (( launched % NUM_GPUS == 0 )); then
    echo "[$(date +%H:%M:%S)] --- waiting for wave to finish ---"
    for p in "${pids[@]}"; do wait "${p}" || true; done
    pids=()
  fi
done
for p in "${pids[@]}"; do wait "${p}" || true; done

echo
echo "================ ALL DONE ================"
for cat in "${CATEGORIES[@]}"; do
  rt="${OUT_ROOT}/${cat}/${cat}_ood_seed${SEED}.txt"
  if [[ -f "${rt}" ]]; then
    echo "### ${cat}"
    grep -E "^overall_" "${rt}" 2>/dev/null || echo "  (no overall_ line — check logs)"
  else
    echo "### ${cat}: MISSING ${rt}"
  fi
done
echo "Videos under: ${OUT_ROOT}/<category>/experiments/<run_name>/video/"