Examples
The mixtrain-examples repository contains complete projects you can run in your own workspace. Each example is self-contained and includes its dependencies and setup instructions.
If this is your first time using Mixtrain, follow the quickstart to install the CLI and log in, then clone the repository:
git clone https://github.com/MixTrain/mixtrain-examples.git
cd mixtrain-examplesStart with continuous VLM post-training
Continuous VLM Post-Training follows a complete iteration of the model lifecycle. Adding labeled images starts LoRA fine-tuning, registers the resulting checkpoint, deploys it, and evaluates it against the base model. The walkthrough explains how dataset versions, workflows, routines, checkpoints, and evaluations fit together; the repository contains the complete runnable project.
Runnable examples
| Example | What you will build | Level | |
|---|---|---|---|
| Hello Workflow | A GPU workflow that reports its CUDA environment | Intro | Run with Mixtrain |
| Daily arXiv Digest | A routine that collects recent papers on a weekday schedule | Intro | Run with Mixtrain |
| Dataset Importer | A workflow that imports a Hugging Face dataset into Mixtrain | Intro | Run with Mixtrain |
| Parameter Sweep | A workflow that submits another workflow across a range of inputs | Intermediate | Run with Mixtrain |
| YOLO Object Detection | An object-detection model for images | Intermediate | Run with Mixtrain |
| YOLO Video Tracking | An object-tracking model with a custom container image | Intermediate | Run with Mixtrain |
| Continuous Multimodal Evaluation | A dataset-triggered routine that evaluates new VLM test cases | Intermediate | Run with Mixtrain |
| Continuous VLM Post-Training | A data-triggered loop that fine-tunes and evaluates a VLM | Advanced | Walkthrough |
More walkthroughs
- Simple Workflow explains workflow inputs, sandbox configuration, and the
run()method. - VLM RL Training introduces reinforcement learning for a vision-language model.
- Evaluations compares outputs from multiple generative models on the same inputs.