Bonsai 2 Compresses Qwen 27B to 6GB but Falters on Long Tasks
PrismML's Bonsai 2 ternary compression shrinks Qwen 27B to a 6GB file while claiming 98% performance, but real-world testing shows it matches the full model only on short tasks and collapses on longer agent workflows.
2026-09-28 ~ 2026-09-28 · 2 related posts
- Episode 1: PrismML's Ternary Bonsai 2 27B Shrinks to 5.9GB, Retains 98.2% Performance(2026-09-18, 20 posts)
- Episode 2: Bonsai 2 Compresses Qwen 27B to 6GB but Falters on Long Tasks(2026-09-28, 2 posts)
- Bonsai 2 squeezes Qwen 27B into 6GB with ternary weights, but fails agentic tasks — Prompt Engineering · 2026-09-28
- Bonsai 2 tested: 98% of Qwen 27B in 6GB, but zero working agent builds — Prompt Engineering · 2026-09-28