EXOIMAGE GPU PREVIEW MODEL PACK MODIFICATION NOTICE

Pack: exoimage.gpu-preview.bk-sdm-tiny
Pack version: 1.1.0
Source model: nota-ai/bk-sdm-tiny
Pinned source revision: 0364108e53b7f7f4d2585e817a0b7a83dc261cfa

EXOIMAGE changed the following files from the pinned source model for local
LiteRT inference:

- models/text-encoder.tflite: converted the pinned text-encoder weights to a
  static LiteRT model.
- models/denoiser.tflite: converted the pinned U-Net weights to a static LiteRT
  model, then applied the pinned same-size graph rewrite that replaces 56
  GATHER_ND operators with RESHAPE operators.
- models/decoder.tflite: converted the pinned VAE decoder weights to a static
  LiteRT model.

EXOIMAGE did not fine-tune or retrain the pinned source weights. The tokenizer
vocabulary and merges files are copied from the pinned source revision without
intentional modification. These notices do not replace or alter the CreativeML
Open RAIL-M terms or use restrictions in LICENSES/LICENSE.txt.