mlx-community/Boogu-Image-0.1-Base-bf16
Text-to-Image • Updated • 2
MLX conversions of Boogu-Image-0.1 (OmniGen2-lineage T2I/edit, Apache-2.0) for Apple Silicon.
Note Reference precision (bf16); parity baseline. WIP.
Note int4 (group=32); cosine 0.99896 vs bf16. ~7.4 GB.
Note 4-step Decoupled-DMD distilled (guidance 1.0) — ~6x faster. int8 ~12.5 GB.
Note 4-step Decoupled-DMD distilled (guidance 1.0) — ~6x faster. bf16 reference.
Note Instruction-based image editing. int4 ~7.9 GB.
Note Instruction-based image editing. bf16 reference.