Benchmarks & research
Supersonic Labs' Julia 1 is a 144M-parameter multilingual decision model that picks among 2 to 20 answers you supply, small enough to run on a CPU or in a browser.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
It adds a decision head to the mmBERT-small encoder and needs its own Python runtime rather than standard Transformers loading. Its makers say knowledge and multi-step reasoning are not established, and on the broad Decision Index it sits near the bottom of the board. It is unrelated to the JevSDK.jl Julia library.
Julia 1 suits sorting short messages into labels you write yourself, such as which team should handle a customer request, in many languages and without a graphics card. Keep the options few, clear and distinct: its makers say long lists of similar labels are where it goes wrong most, so test it on your own messages first.