Supersonic Labs releases CPU-based Julia 1 decision model
Supersonic Labs has released Julia 1, a 144.3M-parameter open decision model designed to select among supplied alternatives rather than generate text. Users provide context, a question and between 2 and 20 options; the model returns one selection along with probabilities for every option. It supports choice classification, ordered scoring and yes-or-no truth estimates, while preserving caller-supplied IDs.
Julia 1 is based on JHU CLSP’s mmBERT-small encoder and is distributed with Apache 2.0 weights on Hugging Face. It runs locally with Python 3.11+ on a CPU or BF16-capable GPU, and an ONNX version can run in browsers through WebGPU. The model supports up to 8,192 combined tokens, although published benchmarks used 1,024. Training and experiments cost about R$540 (US$104.08), according to the lab.
In evaluations, Julia 1 exceeded Jev reference results on Typed Decisions, AG News and DAIR Emotion, but scored 64/100 on Banking77, below the 87% reference. Median latency was 33.15 ms per decision on an Apple M4. A hosted API is planned but is not yet open; Julia 2 is in development.
Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.