Ternary-Bonsai-2-27B 2-bit MLX quantized model trends on Hugging Face

prism-ml · hf · 2026-09-18

prism-ml's Ternary-Bonsai-2-27B MLX 2-bit quantized model is trending on Hugging Face. It targets on-device text generation with ternary quantization, hybrid attention (prismhadamardqwen35), and CUDA/Metal support, squeezing a 27B-class model into 2-bit for local inference.

Related event: PrismML releases Ternary Bonsai 2 27B: 5.95GB, retains 98.2% performance(11 posts)→

Original post →

More from Infra

Infra channel →