🌳 1-bit Bonsai 27B

Full 27B-class reasoning with binary weights — 3.9 GB deployed (14.2× smaller than FP16), ~90% of FP16 intelligence retained, 262K-token context, at a true 1.125 bits/weight. This Space runs the Q1_0 GGUF build with the PrismML llama.cpp fork; the original model is prism-ml/Bonsai-27B-mlx-1bit.

⏳ Loading…

Learn more: Whitepaper · Ternary Bonsai 27B · Prism ML · Discord

Note: this Space runs on free CPU hardware, so replies stream slowly — patience appreciated. It also sleeps when idle; the first message after a wake-up may take a few minutes while the model reloads.