SilentSwarm or Petals?¶
Petals had a different aim: a public, BitTorrent-style swarm of other people's GPUs. Its last release was in 2023, so check that it still works for you before planning around it. SilentSwarm is a private pool you run yourself.
Checked against Petals's own documentation on 2026-11-05. Projects change, so check their current pages too. "Not documented" means we did not find it in what we read, not that it does not exist. We have not run either project against the other, so this page has no speed comparison.
| Petals | SilentSwarm | |
|---|---|---|
| Built for | Running and fine-tuning large models across other people's GPUs, BitTorrent-style | Serving and fine-tuning one model across mismatched machines you control |
| How a model is split | Transformer blocks spread across servers, assigned by throughput | Contiguous layer stages, sized to each card's memory and speed |
| Hardware | NVIDIA, AMD, Apple M1/M2 and WSL | Linux with NVIDIA GPUs |
| Fine-tuning | Yes: adapters and trainable prompts, kept on the client | LoRA fine-tuning across the same split |
| API and interface | Python client, plus a separate chat app. OpenAI compatibility not found | OpenAI-compatible API and a web interface |
| Network | Public swarm by default, or a private one. Its docs say not to use the public swarm for sensitive data | Workers connect out to your own coordinator: direct links where possible, its relay otherwise, across buildings and routers |
| Encryption between machines | No explicit claim in its docs | None of its own between machines. A WireGuard mesh encrypts traffic when it is up |
| Licence | MIT, open source | Source not public, terms agreed during the evaluation |
| Status | Last release 2023-09-06, last commit 2024-08-25. Looks unmaintained, though not archived | In pre-evaluation |
Where Petals is the better choice¶
- Nothing to operate, if a public swarm is available: you need no machines of your own.
- Wider hardware, including AMD and Apple.
- Open source, under MIT.
- Direct access from PyTorch, including hidden states, for research.
Where SilentSwarm fits¶
- Prompts must not run on strangers' machines.
- You want an OpenAI-compatible API.
- You want software that is being actively worked on.
SilentSwarm has gaps of its own: no sandbox, and no encryption of its own between machines or at rest. Data and security lists them, and our benchmarks show what it does on old cards.
Sources¶
SilentSwarm is in pre-evaluation. We are looking for companies, institutions and individuals who want to try it on their own machines: Apply for the evaluation.