Mind Lab Toolkit (MinT)
Get Started

Migrate from Community to Enterprise

MinT Community and Enterprise share the same SDK and API, so code migrates at zero cost. Migration is essentially just switching the endpoint and key — the training script needs no changes.

Three differences that matter for migration

Of all the differences between Community and Enterprise, three relate to code from a migration point of view:

  1. Endpoint. Community uses the public mint.macaron.xin; Enterprise uses a dedicated cluster (cloud / on-premise).
  2. API key. Enterprise keys are issued through an onboarding session and are not interchangeable with the self-service keys of Community.
  3. Available base models. Enterprise additionally offers commercial models such as GLM / Kimi / DeepSeek (see Supported Models).

Everything else — the SDK package name, module paths, method signatures, data structures, and the training-loop pattern — stays the same.

Migration steps

1. Confirm the SDK version

Enterprise and Community share mindlab-toolkit:

pip install --upgrade git+https://github.com/MindLab-Research/mindlab-toolkit.git
python -c "import mint; print(mint.__version__)"

2. Switch the endpoint and key

Point the client's environment variables to the Enterprise address (the mainland region still uses mint-cn; private deployments use the domain Mind Lab provides at delivery):

# Community (old)
# export MINT_BASE_URL=https://mint.macaron.xin/
# export MINT_API_KEY=sk-community-...

# Enterprise (new)
export MINT_BASE_URL=https://<your-enterprise-endpoint>/
export MINT_API_KEY=sk-enterprise-...
export TINKER_BASE_URL=$MINT_BASE_URL
export TINKER_API_KEY=$MINT_API_KEY

For more detailed credential management, see Configure & Manage API Keys.

3. (Optional) Switch to an Enterprise-only base model

If your Community script hard-codes Qwen/Qwen3-..., it works directly on Enterprise. To upgrade to an Enterprise-only commercial model, change one string:

training_client = service.create_lora_training_client(
    base_model="GLM-5.1",        # Enterprise-only
    rank=16,
)

Post-migration checklist

  • python -c "import mint; print(mint.__version__)" outputs a non-empty value.
  • MINT_BASE_URL points to the enterprise tenant endpoint; verify TCP connectivity with curl -v.
  • MINT_API_KEY is an Enterprise key (starts with sk-).
  • Your original SFT script runs on Enterprise on the first try: ServiceClient works with no exceptions, and forward_backward returns futures normally.

Recommendation

For a training job that already works on Community, we recommend running one regression pass with "same base model + same hyperparameters" after migration to confirm the metric trajectory matches, before switching to an Enterprise-only commercial base model or a larger model.

On this page