File size: 8,992 Bytes
881f9f2 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 | #!/usr/bin/env bash
# Canonical GCP context-window closeout launcher.
set -euo pipefail
REPO_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
cd "$REPO_ROOT"
COHORT_TSV="${COHORT_TSV:-configs/gcp_context_closeout.tsv}"
BATCH_ID="${SMARTGRID_BATCH_ID:-gcp_context_$(date -u +%Y%m%dT%H%M%SZ)}"
STATE_FILE="${STATE_FILE:-}"
MANIFEST_TSV="${MANIFEST_TSV:-}"
MANIFEST_JSONL="${MANIFEST_JSONL:-}"
PYTHON_BIN="${PYTHON_BIN:-python3}"
DRY_RUN="${DRY_RUN:-0}"
RUN_JUDGE=1
ROWS_FILTER=""
usage() {
cat <<'EOF'
Usage: scripts/run_gcp_context_batch.sh [--batch-id ID|--resume-batch ID] [--rows A,B] [--dry-run] [--no-judge]
Runs the canonical seven-row GCP context closeout cohort with stable run IDs,
SMARTGRID_RESUME=1, per-row manifest/state files, and idempotent judge scoring.
Set SMARTGRID_SKIP_LOCAL_MODEL_PREFLIGHT=1 only for hosted-model cohorts
where every row has LAUNCH_VLLM=0.
EOF
}
while [ "$#" -gt 0 ]; do
case "$1" in
--batch-id)
BATCH_ID="$2"; shift 2 ;;
--resume-batch)
BATCH_ID="$2"; shift 2 ;;
--rows)
ROWS_FILTER=",$2,"; shift 2 ;;
--dry-run)
DRY_RUN=1; shift ;;
--no-judge)
RUN_JUDGE=0; shift ;;
-h|--help)
usage; exit 0 ;;
*)
echo "ERROR: unknown argument: $1" >&2
usage >&2
exit 2 ;;
esac
done
[ -n "$STATE_FILE" ] || STATE_FILE="logs/gcp_${BATCH_ID}_state.tsv"
[ -n "$MANIFEST_TSV" ] || MANIFEST_TSV="logs/gcp_${BATCH_ID}_manifest.tsv"
[ -n "$MANIFEST_JSONL" ] || MANIFEST_JSONL="logs/gcp_${BATCH_ID}_manifest.jsonl"
mkdir -p logs results/metrics results/judge_logs
cell_dir_name() {
case "$1" in
A) echo "cell_A_direct" ;;
B) echo "cell_B_mcp_baseline" ;;
C) echo "cell_C_mcp_optimized" ;;
Y) echo "cell_Y_plan_execute" ;;
Z) echo "cell_Z_hybrid" ;;
*) echo "cell_${1}" ;;
esac
}
state_value() {
local label="$1" field="$2"
[ -f "$STATE_FILE" ] || return 0
awk -F'\t' -v label="$label" -v field="$field" '
NR == 1 {
for (i = 1; i <= NF; i++) idx[$i] = i
next
}
$1 == label && idx[field] { value = $idx[field] }
END {
if (value != "") print value
}
' "$STATE_FILE"
}
append_state() {
if [ ! -f "$STATE_FILE" ]; then
printf 'label\tconfig\trun_id\tstatus\tstarted_at\tfinished_at\n' >"$STATE_FILE"
fi
printf '%s\t%s\t%s\t%s\t%s\t%s\n' "$1" "$2" "$3" "$4" "$5" "$6" >>"$STATE_FILE"
}
append_manifest() {
local label="$1" config="$2" run_id="$3" run_dir="$4" run_rc="$5" judge_rc="$6" status="$7" started="$8" finished="$9"
if [ ! -f "$MANIFEST_TSV" ]; then
printf 'label\tconfig\trun_id\trun_dir\trun_rc\tjudge_rc\tstatus\tstarted_at\tfinished_at\tprovider\tzone\tinstance\tgpu_type\n' >"$MANIFEST_TSV"
fi
printf '%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\n' \
"$label" "$config" "$run_id" "$run_dir" "$run_rc" "$judge_rc" "$status" "$started" "$finished" \
"${SMARTGRID_COMPUTE_PROVIDER:-gcp}" "${SMARTGRID_COMPUTE_ZONE:-}" "${SMARTGRID_COMPUTE_INSTANCE:-$(hostname)}" "${GPU_TYPE:-}" \
>>"$MANIFEST_TSV"
"$PYTHON_BIN" - "$MANIFEST_JSONL" "$label" "$config" "$run_id" "$run_dir" "$run_rc" "$judge_rc" "$status" "$started" "$finished" <<'PY'
import json
import os
import pathlib
import sys
path, label, config, run_id, run_dir, run_rc, judge_rc, status, started, finished = sys.argv[1:]
payload = {
"schema_version": 1,
"label": label,
"config": config,
"run_id": run_id,
"run_dir": run_dir,
"run_rc": None if run_rc == "" else int(run_rc),
"judge_rc": None if judge_rc == "" else int(judge_rc),
"status": status,
"started_at": started,
"finished_at": finished,
"compute_provider": os.environ.get("SMARTGRID_COMPUTE_PROVIDER", "gcp"),
"compute_zone": os.environ.get("SMARTGRID_COMPUTE_ZONE"),
"compute_instance": os.environ.get("SMARTGRID_COMPUTE_INSTANCE") or os.uname().nodename,
"gpu_type": os.environ.get("GPU_TYPE"),
}
pathlib.Path(path).parent.mkdir(parents=True, exist_ok=True)
with pathlib.Path(path).open("a", encoding="utf-8") as fh:
fh.write(json.dumps(payload, sort_keys=True) + "\n")
PY
}
validate_run_artifacts() {
local run_dir="$1" scenario_glob="$2" trials="$3" run_name="$4"
local validation_rc=0
local -a require_latency_args=()
if [ "${SMARTGRID_RESUME_REQUIRE_LATENCY:-1}" = "1" ]; then
require_latency_args+=(--require-latency)
fi
local validation_output
validation_output="$(
"$PYTHON_BIN" scripts/gcp_resume_state.py validate-run-shell \
--run-dir "$run_dir" \
--scenario-glob "$scenario_glob" \
--trials "$trials" \
--run-name "$run_name" \
"${require_latency_args[@]}"
)" || validation_rc=$?
eval "$validation_output"
echo "Artifact validation for $run_name: ${VALIDATION_COMPLETE:-0}/${VALIDATION_EXPECTED:-0} complete, ${VALIDATION_MISSING:-0} missing"
if [ "$validation_rc" -ne 0 ]; then
echo "ERROR: incomplete trajectory artifacts for $run_name: ${VALIDATION_REASON:-unknown}" >&2
echo "Missing sample: ${VALIDATION_MISSING_SAMPLE:-[]}" >&2
fi
return "$validation_rc"
}
preflight_runtime() {
[ "$DRY_RUN" = "1" ] && return 0
command -v "$PYTHON_BIN" >/dev/null
"$PYTHON_BIN" - <<'PY'
import sys
if sys.version_info < (3, 11):
raise SystemExit("Python >=3.11 required")
PY
if [ "${SMARTGRID_SKIP_LOCAL_MODEL_PREFLIGHT:-0}" = "1" ]; then
return 0
fi
command -v g++ >/dev/null
command -v cc1plus >/dev/null
if command -v nvidia-smi >/dev/null 2>&1; then
nvidia-smi -L >/dev/null
fi
"$PYTHON_BIN" - <<'PY'
import importlib.util
for module in ("torch", "vllm"):
if importlib.util.find_spec(module) is None:
raise SystemExit(f"{module} is not importable")
PY
}
preflight_boot_disk_autodelete() {
[ "$DRY_RUN" = "1" ] && return 0
[ -n "${SMARTGRID_COMPUTE_INSTANCE:-}" ] || return 0
[ -n "${SMARTGRID_COMPUTE_ZONE:-}" ] || return 0
command -v gcloud >/dev/null 2>&1 || return 0
local autodelete
autodelete="$(
gcloud compute instances describe "$SMARTGRID_COMPUTE_INSTANCE" \
--zone "$SMARTGRID_COMPUTE_ZONE" \
--format='value(disks[0].autoDelete)' 2>/dev/null || true
)"
if [ "$autodelete" = "True" ] || [ "$autodelete" = "true" ]; then
echo "ERROR: boot disk autoDelete=true for $SMARTGRID_COMPUTE_INSTANCE; set autoDelete=false before capture." >&2
return 1
fi
}
preflight_runtime
preflight_boot_disk_autodelete
tail -n +2 "$COHORT_TSV" | while IFS=$'\t' read -r label config; do
[ -n "$label" ] || continue
if [ -n "$ROWS_FILTER" ] && [[ "$ROWS_FILTER" != *",$label,"* ]]; then
continue
fi
if [ ! -f "$config" ]; then
echo "ERROR: config missing for $label: $config" >&2
exit 1
fi
existing_run_id="$(state_value "$label" run_id || true)"
existing_status="$(state_value "$label" status || true)"
# shellcheck disable=SC1090
source "$config"
run_id="${existing_run_id:-${BATCH_ID}_${label}_${EXPERIMENT_NAME}}"
run_dir="benchmarks/$(cell_dir_name "$EXPERIMENT_CELL")/raw/$run_id"
started="$(date -u +%Y-%m-%dT%H:%M:%SZ)"
if [ "$existing_status" = "complete" ]; then
if validate_run_artifacts "$run_dir" "$SCENARIOS_GLOB" "$TRIALS" "$run_id"; then
echo "Skipping $label: already complete in $STATE_FILE"
append_manifest "$label" "$config" "$run_id" "$run_dir" "" "" "skipped_complete" "$started" "$started"
continue
fi
echo "WARNING: $label was marked complete but artifact validation failed; rerunning with SMARTGRID_RESUME=1" >&2
fi
append_state "$label" "$config" "$run_id" "started" "$started" ""
if [ "$DRY_RUN" = "1" ]; then
echo "DRY_RUN $label: SMARTGRID_RUN_ID=$run_id bash scripts/run_experiment.sh $config"
append_state "$label" "$config" "$run_id" "dry_run" "$started" "$started"
append_manifest "$label" "$config" "$run_id" "$run_dir" "" "" "dry_run" "$started" "$started"
continue
fi
run_rc=0
SMARTGRID_BATCH_ID="$BATCH_ID" \
SMARTGRID_RUN_ID="$run_id" \
SMARTGRID_RESUME=1 \
SMARTGRID_COMPUTE_PROVIDER="${SMARTGRID_COMPUTE_PROVIDER:-gcp}" \
SMARTGRID_COMPUTE_INSTANCE="${SMARTGRID_COMPUTE_INSTANCE:-$(hostname)}" \
bash scripts/run_experiment.sh "$config" || run_rc=$?
artifact_rc=0
validate_run_artifacts "$run_dir" "$SCENARIOS_GLOB" "$TRIALS" "$run_id" || artifact_rc=$?
judge_rc=0
if [ "$RUN_JUDGE" = "1" ]; then
"$PYTHON_BIN" scripts/judge_trajectory.py \
--run-dir "$run_dir" \
--scenario-dir data/scenarios \
--out results/metrics/scenario_scores.jsonl \
--log-dir results/judge_logs || judge_rc=$?
fi
finished="$(date -u +%Y-%m-%dT%H:%M:%SZ)"
status="complete"
[ "$run_rc" -ne 0 ] && status="run_failed"
[ "$run_rc" -eq 0 ] && [ "$artifact_rc" -ne 0 ] && status="artifact_failed"
[ "$judge_rc" -ne 0 ] && status="judge_failed"
append_state "$label" "$config" "$run_id" "$status" "$started" "$finished"
append_manifest "$label" "$config" "$run_id" "$run_dir" "$run_rc" "$judge_rc" "$status" "$started" "$finished"
done
|