smoke-test.sh

#!/bin/sh
set -eu

BASE_URL=${BASE_URL:-http://127.0.0.1:4000/v1}
API_KEY=${API_KEY:-${LITELLM_MASTER_KEY:-}}
TIMEOUT_SECONDS=${TIMEOUT_SECONDS:-120}

if [ -z "${API_KEY:-}" ]; then
  echo "set API_KEY or LITELLM_MASTER_KEY" >&2
  exit 1
fi

tmp_dir=$(mktemp -d)
trap 'rm -rf "$tmp_dir"' EXIT

/usr/bin/curl -fsS \
  -H "Authorization: Bearer $API_KEY" \
  "$BASE_URL/models" > "$tmp_dir/models.json"

jq -e '.data | length == 1 and .[0].id == "glm-5-nvidia"' "$tmp_dir/models.json" >/dev/null
jq '{data:[.data[] | {id,owned_by}]}' "$tmp_dir/models.json"

cat > "$tmp_dir/payload.json" <<'EOF'
{"model":"glm-5-nvidia","stream":false,"messages":[{"role":"user","content":"Reply with exactly: smoke test ok"}]}
EOF

/usr/bin/curl --max-time "$TIMEOUT_SECONDS" -fsS \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  "$BASE_URL/chat/completions" \
  --data-binary @"$tmp_dir/payload.json" > "$tmp_dir/completion.json"

jq -e '.model == "glm-5-nvidia" and .choices[0].message.content == "smoke test ok"' "$tmp_dir/completion.json" >/dev/null
jq '{model,choices:[.choices[] | {message:.message.content}],usage}' "$tmp_dir/completion.json"