# NOTICE

super-decider-mlx. Copyright 2026 Singularity Compute. Licensed under the Apache License,
Version 2.0 (see LICENSE).

## Base model
- google/diffusiongemma-26B-A4B-it (Google DeepMind), Apache-2.0
  (https://ai.google.dev/gemma/docs/gemma_4_license).
- nvidia/diffusiongemma-26B-A4B-it-NVFP4, NVIDIA's ModelOpt NVFP4 checkpoint of the above,
  Apache-2.0.
- Singularity-Compute/super-decider, the fine-tune whose weights are converted here,
  Apache-2.0.

## Runtime
- `runtime/structured_server.py` is derived from mmastrac/djev (https://github.com/mmastrac/djev,
  Apache-2.0) and modified by Singularity Compute.
- Ported to MLX (https://github.com/ml-explore/mlx, MIT); `superdecider/` is an independent
  reimplementation of the DiffusionGemma forward pass against MLX, not a derivative of the
  transformers implementation, but it is written to match it numerically.
- `transformers` (Apache-2.0) is used only to load the tokenizer.