Summary
Gensyn releases OPEN-1B, a 1.61B-parameter model trained on 400B tokens with a state hash for every optimizer step. Its fixed-order kernels, data stream and topology-independent reductions reproduce the same bits across CPU, CUDA and Metal. The release includes the full corpus, 810 checkpoints, training code and an audit-replay harness. A volunteer can verify one step on commodity hardware instead of rerunning the full training job.
Why it matters
For teams evaluating open models, this adds a verifiable tier above published weights and recipes. It can expose undeclared data or training changes without requiring an auditor to fund a full retraining run. The 1B scale and custom deterministic kernels make it a research prototype rather than a drop-in recipe for frontier training.
Technical details
| Model | 1.61B parameters; 1.08B non-embedding parameters |
|---|---|
| Training | 400,004,481,024 tokens; 80,957 optimizer steps; 48 H100 GPUs |
| Audit | bitwise replay on CPU, CUDA or Metal with chained BLAKE2b state hashes |
| Artifacts | full corpus, 810 checkpoints, code, audit kits and browser audit app |
| License | Apache-2.0 code repository |
| Limitation | 1B-scale demonstration; replay trusts pinned compiled kernels and artifacts |
Tags
open-modeltrainingreproducibilityauditmodel-provenance