Launched this week
tritify
Deterministic AI inference across multiple machines
2 followers
Deterministic AI inference across multiple machines
2 followers
tritify is an open-source deterministic AI inference runtime from Pelloran. It lets one AI model run across multiple machines while keeping execution reproducible and bit-verifiable. Same model. Same input. Same execution schedule. Same bits. Built for reproducibility, auditability, and distributed inference — not as a raw throughput claim. Open source under Apache 2.0.
Free
Launch Team
CloudflareDeploy serverless functions globally in seconds
Promoted
Hi,
We built tritify because we kept coming back to one question:
If the same AI model runs twice, why shouldn’t we be able to reproduce the exact same computation?
Most inference stacks optimize primarily for throughput. tritify explores a different axis: determinism.
It uses exact integer execution so that, under the same model, input, and execution schedule, the run can reproduce the same bits.
That also extends to distributed inference.
One model can be split across multiple machines, with each machine holding only part of it, while the distributed result remains reproducible and can be checked against a reference run by hash.
We demonstrated this with GLM-4.5-Air — roughly 106.9B parameters and ~427 GB of integer codes — distributed across three machines.
This is not a speed claim.
The goal is reproducibility, auditability, and verifiable execution.
tritify is open source under Apache 2.0.
GitHub:
https://github.com/pelloran-inc/tritify
Paper:
https://doi.org/10.5281/zenodo.23986737
We’d especially welcome independent tests, criticism, and reproducibility results.