Skip to content

Install the MinT SDK

The MinT client SDK ships as mindlab-toolkit — a pure Python package. Enterprise and Community share the same package, so code migrates at zero cost; the only differences are the endpoint and the API key.

Requirements

  • Python 3.11 or above.
  • We recommend creating an isolated environment with venv or conda.
  • No local GPU needed — the client only assembles training requests; all forward / backward / sampling runs on the MinT cluster.
  • Network access to your enterprise tenant's MinT endpoint (mint.macaron.im or mintcn.macaron.xin, or your private cluster address).

Install command

pip install git+https://github.com/MindLab-Research/mindlab-toolkit.git

The install also pulls in the mint package along with the compatible tinker (>= 0.15.0) runtime helper library.

Verify

python -c "import mint; print(mint.__version__)"

tinker-compatible entry point

If you already have a Tinker training script, migrating to MinT takes just one import line change:

When you import mint, it automatically patches tinker's key-validation logic so that sk-* API keys work directly.

Note

In the MinT training loop, do not call zero_grad_async — gradient zeroing is handled centrally on the server side. This is a known behavior difference between MinT and tinker.

Namespaces

The core namespaces available after install:

  • mint: the core API (ServiceClient, create_lora_training_client, forward_backward, optim_step, save_weights_and_get_sampling_client, etc.).
  • mint.types: type definitions (Datum, ModelInput, AdamParams, SamplingParams, etc.).
  • mintx: MinT-specific extensions (including constants like mintx.OPENPI_FAST_MODEL).

Offline install for private environments

Private delivery explicitly includes "dependency images and offline install packages, supporting one-click startup in a fully air-gapped environment." For Enterprise private-deployment customers, the Mind Lab team delivers the offline install package and upgrade process together with the cluster during onboarding — no need to build your own pip mirror.

Next steps

After the SDK is installed, continue to Configure & Manage API Keys, then follow the Quick Start to get your first training run working.