prism-ml/bonsai-image-binary-4B-unpacked

Bonsai Image · Binary 4B — Unpacked FP16 Safetensors

FP16 safetensors (HuggingFace diffusers format) of the 1-bit Bonsai Image 4B model. This repo exists for users who want to run Bonsai Image with stock diffusers or other frameworks that don't yet support our low-bit packs natively. The 1-bit kernels are currently in our forks of MLX and the gemlite low-bit GEMM stack — once they're broadly available, this unpacked version will no longer be needed.

We strongly recommend using the optimized low-bit packs instead. The 1-bit format is where the Bonsai Image gains come from — an 8.3× transformer footprint reduction, sub-iPhone deployment, and ~5× faster inference vs the stock FP16 pipeline on Apple Silicon. This unpacked FP16 version is full-size and provides none of those advantages.

For the optimized 1-bit release packs (recommended):

• bonsai-image-binary-4B-mlx-1bit — 1-bit MLX for Apple Silicon (Mac, iPhone, iPad)
• bonsai-image-binary-4B-gemlite-1bit — 1-bit gemlite/HQQ for NVIDIA GPUs (Linux + Windows)

For the higher-quality variant:

• bonsai-image-ternary-4B-unpacked — Ternary FP16 safetensors (and the matching MLX 2-bit + gemlite INT2 packs)

See the Bonsai Image Demo repo for one-command setup of either variant on Mac, Linux, or Windows.