Managed fine-tuning platform
Onrup vs Predibase
Should I use Onrup or Predibase?
Predibase has been acquired and predibase.com now redirects to Rubrik. If you are an existing customer, confirm continuity terms directly before comparing anything else.
A fine-tuning platform built around serving many LoRA adapters from a single GPU. Acquired — predibase.com now redirects to Rubrik.
Where they differ
| Onrup | Predibase | |
|---|---|---|
| Availability | Generally available | Site redirects to the acquirer |
| Multi-adapter serving | Adapters co-resident against a shared base | Their core strength, historically |
| Evaluation gate | Blinded, blocking, before deploy | Not part of the product |
| Pricing | Published rate card per GPU class | No reachable pricing page |
On price
No standalone pricing page is reachable: predibase.com returns a 301 redirect to rubrik.com as of the date checked, following the acquisition. Existing customers should confirm continuity terms directly.
Rather than estimate a comparable figure, we leave it blank. Their pricing page, checked 2026-08-06.
The longer answer
As of the date this page was verified, predibase.com returns a permanent redirect to rubrik.com. That is the most important fact for anyone evaluating them, and it is why the comparison below is shorter than the others.
Their technical contribution was real and worth acknowledging. The work on serving many adapters from a single GPU — dynamically swapping them rather than running separate deployments — was the best in the category, and the open-source server that came out of it is still worth reading if you are building anything similar.
We do the same thing in principle: adapters are small, they share one resident base model, and the marginal cost of an additional variant is a fraction of a full model. We do not claim to have taken it as far as they did.
If you are an existing customer, the question is continuity rather than features, and it is one only the acquirer can answer.
Where Predibase wins
Their multi-adapter serving work was the best in the category, and the open-source server that came out of it is still worth reading if you are serving hundreds of adapters against one base model.
Choose them if
- You are an existing customer with continuity terms in place
- You want the open-source multi-adapter serving work specifically
On ownership
Weights on Onrup are downloadable from every finished run and publishable to a model hub in one call. On Predibase: Adapters were exportable; confirm post-acquisition terms.
Frequently asked questions
What happened to Predibase?
The domain now issues a 301 redirect to rubrik.com, which indicates an acquisition. We verified this directly rather than reporting it second-hand; the date is on this page.
Do you support multi-adapter serving?
Yes. Adapters are co-resident against a shared base model within your account, so serving several task variants costs one base plus a small amount per adapter rather than a full model each.
Can I migrate my adapters directly?
Only if they were trained against a base model and version in our catalogue. Otherwise the training data migrates and the model is retrained.
Researching Predibase alternatives more broadly? →
Last verified 6 August 2026. Predibase figures come from their own pricing page on the date checked.
Start with the free tier
A magic link creates your account, your tenant and your first API key. No card until you ask for compute.