PrismML releases Bonsai 27B ternary model

Bonsai 27B, a ternary 3-bit model based on Qwen3.6, achieves near fp16 precision while running on just 10GB of VRAM. It uses PrismML's ternary quantization technique similar to BitNet.