Stage-A SFT checkpoint for Pref-Restore — released so the community can skip the most expensive stage and jump straight to PhaseB preference-RL training.

📄 Paper: Bridging Information Asymmetry: A Hierarchical Framework for Blind Face Restoration with Reduced Uncertainty (TPAMI 2026) 🧑‍💻 Code: https://github.com/zjYao36/Pref_Restore

PhaseA (hierarchical SFT) is by far the most compute-heavy step of the full Pref-Restore pipeline. In our experiments, restoration quality keeps improving as PhaseA training continues, but with diminishing marginal returns — most of the easy gains land early; later iterations cost a lot of GPU-hours for a small numerical bump.

To make PhaseB training accessible without re-running our entire SFT, we publish this PhaseA snapshot. It is tuned to lean toward restoration fidelity and image realism, at the cost of slightly weaker aesthetic quality — exactly the trade-off you want as a base model that PhaseB's preference-RL will then push toward perceptual preference.

Then in DiffusionNFT/config/pref_restore_gt.py set:

and launch PhaseB as documented in the repo README.

You can also use it directly for inference with the inference_batch_noPrompt_fixLQ_vae.py script in the code repo — pass this folder as --model_path.

blip3oQwenForCausalLMVAE — a BLIP-3o-NEXT-style multimodal AR backbone (~3B) coupled with a SANA 1.5 diffusion decoder and a VAE-encoded HQ branch (see config.json).

The checkpoint contains only the inference weights (model-0000{1,2,3}-of-00003.safetensors + config + tokenizer); the DeepSpeed optimizer state has been stripped.

Apache-2.0. If you use this checkpoint, please cite the paper: