Interesting note from the repo [1]:
> A Note on Model Release
> Due to organizational policy constraints, we are unable to release the full production-trained model weights. To support the research community, we plan to release a foundation checkpoint with a small number of layers removed from both the LLM backbone and the diffusion head (flow head). The remaining layers and all other components (vision encoder, projections, embeddings, etc.) are fully preserved. With a short fine-tuning pass on your own data, the removed layers can be quickly re-learned and the model restored to full quality.
[1] https://github.com/facebookresearch/tuna-2#a-note-on-model-r...
