SupersonicLabs/Julia-1 — new model trending #30 on Hugging Face
Supersonic Labs released Julia 1, a 144.3M-parameter model that picks among supplied answers for classification and routing.
Supersonic Labs released Julia 1, the first Julia-family model and a test of its training system, trending on Hugging Face. The 144.3M-parameter checkpoint starts from JHU CLSP's mmBERT-small and adds a decision head that selects among 2-20 supplied options for classification, routing, ordered scores, or Boolean decisions. A 2026-09-24 H200 BF16 evaluation reported 73.15% typed-decision accuracy (1,463/2,000), 94/100 on AG News, 86/100 on DAIR Emotion, 64/100 on a Banking77 shortlist, and 71.50% MASSIVE scenario accuracy across 52 locales (110,573/154,648). The card says fact recall, algebra, and long calculations are not established by these tests.
- Julia 1 has 144.3M parameters and is based on mmBERT-small.
- It maps a state, question, and 2-20 options to one decision.
- Typed decisions scored 73.15% on 1,463 of 2,000 cases.
- Pilots: AG News 94/100, emotion 86/100, Banking77 64/100.
- MASSIVE macro accuracy was 71.50% across 52 locales.
- Vendors
- Supersonic Labs
- Organizations
- Supersonic LabsJHU CLSP
- AI models
- Julia 1mmBERT-small
From the original article
8 min read at huggingface.cotext-classification · 274 downloads · 212 likes · license apache-2. 0
By SupersonicLabs · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.