Ternary Bonsai
by Prism ML · Model · Language & code models
Bonsai 2 from Prism ML is a family of vision models built for ternary weights – smaller models with close to full-precision quality.
AccessOpen weightsNeeds~5.9 GB; smallest build ~0.9 GBDownload5.9 GBSize27B parametersLicenseapache-2.0
Evidence
- Sep 20Ternary Bonsai 2 27B compresses Qwen 3.8 27B from 55.6 GB to 5.9 GB while keeping near-original benchmark scores
- Jul 19Ternary (5.9 GB) and 1-bit (3.9 GB) compressions of Qwen 3.6 27B small enough for an iPhone 17 Pro
- Jul 19Ternary retains 95% of baseline quality (80.5% vs 85% average benchmark score); 1-bit retains 90%
- Jul 16Fully ternary (-1/0/+1) 27B from Prism ML, 128-weight blocks sharing one scale, about 1.71 bits per weight