FAQ

Common first-run questions about SFT vs RL, endpoint selection, and API key access.

Should I use SFT or RL?

Use SFT when you already know the target output and have labeled examples.

Use RL when you do not have one fixed target answer but you can score behavior with a reward, verifier, test suite, or environment feedback.

If you have both, you can combine them. A common pattern is SFT for the base behavior and RL for optimization, but that is not a required order for every task.

Does MinT support SFT?

Yes. MinT supports SFT directly.

The standard SFT path is:

• forward_backward(..., loss_fn="cross_entropy")
• optim_step(...)

Which domain should I use?

Choose by your network path:

• Mainland China -> https://mint-cn.macaron.xin/
• Outside Mainland China -> https://mint.macaron.xin/

If you are unsure, try the one that matches your region first. The practical goal is lower latency and stable connectivity.

Where do I get MINT_API_KEY?

MINT_API_KEY is currently issued by the Mind Lab team.

To request access:

• go to https://macaron.im/mindlab
• use Schedule a Demo
• or email contact@mindlab.ltd