Minibase v0.2: Feature Updates
Synthetic data generation, new AI models, and a marketplace.
Today we released a bunch of features for Minibase, including a public marketplace, improved training insights, and several tools to make your workflow smoother. Remember that Minibase is in public beta and is currently free to use. Join our Discord community to learn more.
Model & Dataset Marketplace
We’ve launched our new Model & Dataset Marketplace, a community hub where you can share models and datasets. Users can set the privacy for each item (Individual, Team, or Public). Public models will appear in the marketplace, where others can view them and add them to their own workspace. Datasets remain secure—others cannot download your raw data locally.
Redesigned Model Detail Pages
Model and dataset detail pages have been redesigned. Whether in the marketplace or under your “Datasets” and “Models” tabs, you can now see much more information at a glance.
Training Performance Metrics
After training a model on Minibase, you’ll now receive automatic evals and tests by email. These metrics give you deeper insight into the quality of your model, and we’ll be expanding them over time.
Post-Training Editing
You can now edit your model name and description after training, and the same goes for datasets. This makes it easier to keep your workspace organized as your projects evolve.
Synthetic Data Generation
We’ve added a tool for synthetic data generation. You can describe the problem you’re solving, provide seed examples, and set a target number of datapoints. Minibase will generate examples continuously until your desired dataset size is reached.
Continue Training
You can now continue training a model after the initial run, even if it has already been deployed and tested. This makes it easier to fine-tune using additional datasets without starting from scratch.
Reliability Improvements
We’ve fixed several issues related to dataset uploads, training, inference, and more, making Minibase more stable overall.
Better Base Models
Finally, we’ve improved the quality of our base models, so you’ll see higher-quality outputs across all training workflows.
Full HF Fidelity
We’ve added full HF fidelity for all models, including chat models. This means you can now deploy models in full fidelity without compromise. We’ve also updated the inference server to better handle large files, making deployments smoother and more reliable.
Tooltips and UI Improvements
We’ve started rolling out tooltips, beginning with the model detail view. Many of these tooltips are linked to our Help Center (help.minibase.ai) articles, which we continue to expand on a weekly basis. Alongside this, we made numerous small improvements across the product—updated copy, clarified text, and polished flows—to make the interface easier to use.
We’ve been hard at work building new features. And we think this release is a big step toward making Minibase faster, more flexible, and more collaborative. We can’t wait to see what you build.
— The Minibase Engineers