PrismML Shrinks Qwen3.5 VL to Ternary Weights, 9B Model Fits in 2GB
PrismML released Ternary Bonsai 2, converting Qwen3.5 VL models (2B/4B/9B) to ternary weights of -1, 0, +1, shrinking the 9B model to just 2GB while retaining about 90% quality across 19 benchmarks.
2026-10-20 ~ 2026-10-20 · 2 related posts
- Episode 1: PrismML Open-Sources Ternary Bonsai 2: 9B in 2GB, 2B Hits 500 tok/s in Browser(2026-10-16, 10 posts)
- Episode 2: Bonsai 2 Tri-value Small Models Tested: 4B Hits the Sweet Spot(2026-10-16, 2 posts)
- Episode 3: PrismML Shrinks Qwen3.5 VL to Ternary Weights, 9B Model Fits in 2GB(2026-10-20, 2 posts)
- PrismML's Ternary Bonsai 2 squeezes Qwen3.5 VL models to ternary weights, 9B at 2GB — lmoroney · 2026-10-20
1 near-duplicate retellings: lmoroney