Ternary Bonsai 2 squeezes Qwen3.5 VL models to -1/0/+1 weights, 9B keeps 90% quality
lmoroney · x · 2026-10-20
PrismML's Ternary Bonsai 2 converts Qwen3.5 vision-language models at 2B, 4B and 9B into ternary weights (-1, 0, +1), with files ranging from 0.43 GB to 2 GB. According to PrismML, the 9B version retains over 90% of its quality across 19 benchmarks, released under Apache 2.0. Laurence Moroney breaks down how ternary weights work and what the benchmark table shows.
Related event: PrismML Shrinks Qwen3.5 VL to Ternary Weights, 9B Model Fits in 2GB(2 posts)→
More from Models
- Fable 5.5 rumored to launch tomorrow with better reasoning and root-cause fixes — daniel_mac8 · 2026-10-20
- AI predicts cancer death risk from a single face photo, Lancet study shows — Scobleizer · 2026-10-20
- OpenAI says latest models boost COBOL, Fortran and VB migration for legacy systems — ahelkky · 2026-10-20
- Top agent solves only 33% of Embodied First Exam; GPT-6 Astra took 40 min per bolt — ManlingLi_ · 2026-10-20
- Fable 5.5 delayed again — even the scoop itself got a patch note pushing past October — imjustnewatai · 2026-10-20
- DeepSeek-V4.1-Flash ranks 6th on CrusaderBench, questions its own existence mid-game — Hesamation · 2026-10-20