Pre-evaluation
Evaluate SilentSwarm on your own machines
SilentSwarm serves and fine-tunes open language models across ordinary machines, so that together they hold a model none of them could hold alone. It is in pre-evaluation: the source code is not public, and we are looking for companies, institutions and individuals who want to try it and tell us what they find.
What evaluating means
You run SilentSwarm on machines you choose, with a model and a workload that matter to you, and we work through the setup with you. Then you tell us what held up and what did not.
What you get
- The software on your own machines. With your own coordinator, your prompts and data stay there.
- A direct line to the people building it. Questions go to them, not to a ticket queue.
- Numbers from your hardware. How fast it serves on your machines, measured the way we measure our benchmarks.
- A say in what comes next. What evaluators run into shapes what we build.
What we ask of you
- Honest feedback. What worked, what did not, and what you would need to use it for real.
- One contact person we can reach during the evaluation.
- Machines of your own. Tell us what you have, and we will say honestly whether it fits.
- Terms agreed first. We agree the terms with you before you start.
Who it suits
- Research groups and universities with workstations or GPUs they already own.
- Cities and public institutions that want their data to stay in their own network.
- Companies that need a model larger than one of their GPUs can hold, or want to run it in-house.
- Heating and hardware makers looking at heat from compute.
- Individuals with GPUs of their own. You do not need an organisation behind you.
What exists, and what does not yet
We would rather name the gaps now than have you find them.
Available today
- Serving through an OpenAI-compatible API, across one machine or several.
- LoRA fine-tuning split across machines, with checkpoints and exported models.
- Published benchmarks with hardware and method, on the benchmark page.
Not yet
- Service-level guarantees. Expect rough edges while we stabilize.
- A sandbox. The worker runs with the rights of whoever installed it, so run it on a machine where that is acceptable. The data and security page has the details.
- Payment for contributing a machine. Contribution is measured, but does not pay money today.
Questions before you apply
- Is the source code public?
- No. SilentSwarm is in pre-evaluation and its source code is not public. A source review under NDA is available on request.
- What does an evaluation cost?
- We agree the terms with each evaluator before the start, so ask us about your case in the form.
- Do I need to be a company?
- No. Companies, public institutions, research groups and individuals can all apply.
- What happens after I apply?
- We read every application ourselves and reply by email. We may not be able to start with everyone at once, and we will tell you if so.
Apply for the evaluation
A few lines are enough. We answer every message ourselves.
We use your details only to answer you - see the privacy policy.