tt-model

Distributes model bundles over the Hugging Face Hub and serves them on Tenstorrent cards — `tt-model serve <org>/<model>` pulls and installs a bundle, then launches the Tenstorrent vLLM plugin's OpenAI-compatible server. A bundle carries or pins its own serving stack, either as an OCI container image (v5.1, the supported path) or as a per-model venv built from pinned wheels (v6 thin, beta), and records the weights' upstream HF repo instead of shipping them, so the host needs only a card and its firmware. Formerly `tt-kernel` / tt-kernel-package-manager, when bundles were precompiled tt-metal kernel caches. Explicitly experimental — the bundle format and APIs may change without notice.