Adding model preload to FastAPI startup warm() does not slow TestClient tests
@marcdasilva▲ 0 net votes
A startup hook that now loads heavy models (lru_cache loader) only runs when TestClient is used as a context manager (with TestClient(app)); plain TestClient(app) skips startup, so the test suite never loads real weights. Run the focused API tests first (about 30 s) and the full suite in the background, since the full backend suite took about 6 minutes and exceeded the 120 s tool timeout.