Refine OFoldX model card provenance

README.md

@@ -49,6 +49,7 @@ Converted RFdiffusion checkpoint for backbone generation, motif scaffolding, and
 
 - **Upstream Project**: RFdiffusion
 - **Source Checkpoint**: `Complex_base_ckpt.pt`
+- **Source Release**: [https://github.com/RosettaCommons/RFdiffusion/blob/main/scripts/download_models.sh](https://github.com/RosettaCommons/RFdiffusion/blob/main/scripts/download_models.sh)
 - **Primary Paper**: [De novo design of protein structure and function with RFdiffusion](https://doi.org/10.1038/s41586-023-06415-8)
 - **Upstream License**: BSD for upstream RFdiffusion code and referenced model weights
 
@@ -74,6 +75,7 @@ Converted RFdiffusion checkpoint for backbone generation, motif scaffolding, and
 - **Hub repository**: [oteam/rfdiffusion-complex-base](https://huggingface.co/oteam/rfdiffusion-complex-base)
 - **Upstream paper**: [De novo design of protein structure and function with RFdiffusion](https://doi.org/10.1038/s41586-023-06415-8)
 - **Upstream repository**: [RFdiffusion](https://github.com/RosettaCommons/RFdiffusion)
+- **Source checkpoint release**: [https://github.com/RosettaCommons/RFdiffusion/blob/main/scripts/download_models.sh](https://github.com/RosettaCommons/RFdiffusion/blob/main/scripts/download_models.sh)
 - **Code**: [`ofoldx/models/rfdiffusion/model.py`](https://github.com/OTeam-AI4S/OFoldX/tree/main/ofoldx/models/rfdiffusion/model.py)
 - **Project repository**: [https://github.com/OTeam-AI4S/OFoldX](https://github.com/OTeam-AI4S/OFoldX)
 - **Issues**: [https://github.com/OTeam-AI4S/OFoldX/issues](https://github.com/OTeam-AI4S/OFoldX/issues)
@@ -113,11 +115,11 @@ metadata for loading it.
 
 ### Training Data
 
-RFdiffusion training and use are described in the upstream Nature paper and repository. OFoldX does not redistribute the training set.
+RFdiffusion fine-tunes a RoseTTAFold-style structure network as a denoising diffusion model over PDB protein structures. OFoldX does not redistribute the training set.
 
 ### Training Procedure
 
-OFoldX converts released RFdiffusion checkpoints into `model.safetensors`; it does not run RFdiffusion training.
+Upstream RFdiffusion noises residue frames with Gaussian C-alpha translation noise and rotational Brownian motion, trains denoising to true frames with self-conditioning, and uses checkpoint-specific inference configs. OFoldX converts released RFdiffusion checkpoints into `model.safetensors`; it does not run RFdiffusion training.
 
 ## Evaluation