Overview

Train, deploy, and evaluate your own models for the tasks that matter to your product.

Mint Recursive brings open-model training and inference into one workspace. Fine-tune with your own examples, deploy the resulting model, and use evaluation results to guide the next iteration. Training and hosted inference run remotely, without a local GPU.

Start with your goal

Train your first model Start with a JSONL dataset and create your own model version.Read the guide Connect a model to your app Create a deployment, verify readiness, and obtain its API configuration.Read the guide Measure model quality Choose a benchmark and model, then inspect execution and outcomes.Read the guide Plan your costs Compare model rates and dedicated GPU options.Read the guide

From data to a usable model

1. Training data
2. SFT job
3. Checkpoint
4. Deployment
5. API calls & evaluation

Begin with a clear task and examples of the behavior you want. Use training to adapt the model, then test on held-out tasks before deciding which version to serve.

Training and deployment workspace · Screenshot placeholder · final product capture to follow

Understand your workspace

Resource	Purpose	Console location
Dataset	Training examples and validation results	My data (我的数据)
Training job	Model, dataset version, execution, and artifacts	My training (我的训练)
Model version	Training lineage, checkpoints, and deployment actions	My models (我的模型)
Deployment	An inference service for a trained version	Deploy (部署)
Evaluation run	Execution and results for a benchmark	My evaluations (我的评测)

For an existing model, open Model calls (模型调用) to obtain API access details. To customize a model, follow the Quickstart.