Diffusers
Safetensors
English
RFD4ProteinaPipeline
biology
protein-design
biomolecular-design
structure-generation
flow-matching
annotated-cif
Instructions to use nvidia/RFD4-Proteina-Diffusers with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use nvidia/RFD4-Proteina-Diffusers with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("nvidia/RFD4-Proteina-Diffusers", dtype=torch.bfloat16, device_map="cuda") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
Commit History
2 days ago
-
Cards: point links at rfd4-proteina.org docs and the new wordmark
Verified signature92fc3ae -
Model card: support is merged into Diffusers, install from git until the next release
Verified signature7e6724e -
Install Diffusers from git until the release carries RFD4-Proteina
Verified signatureRFD4-Proteina was merged into Diffusers upstream as huggingface/diffusers#15088 on 2026-10-20, so the pipeline no longer comes from a staging branch. It is not on PyPI yet, which the next Diffusers release fixes, so the quickstart installs from git and shows the plain pip line commented out beside it. Checked rather than assumed, in a clean 3.12 environment: the git install imports RFD4ProteinaPipeline, and nvidia/RFD4-Proteina-Diffusers loads through it with all four components and no newly-initialized warnings. accelerate is listed because the load fails without it: "`low_cpu_mem_usage` cannot be False when `keep_in_fp32_modules` is True".6e0e4bb -
Point links at the public RosettaCommons/RFD4-Proteina repository, pin the 2026-10-19 revision, update install paths
Verified signature8f8b7ad -
Model card: explicit pipeline arguments, link the open Diffusers pull request
Verified signature155e470 -
Drop transformer options that are constant for the release
Verified signaturec0b2e85 -
Point the examples section at a reproduction path that exists
Verified signatureThe section told readers to run examples/run_examples.py from the prep repository. That script reads a flat <name>.cif from a local --inputs-dir, not this repository layout of examples/<name>/input.cif, and the prep repository ships seven of these nine inputs: the two antibody ones are excluded there. So the instruction could not reproduce the table it sat under. It now points at this repository own inputs through the Quickstart above, which already downloads examples/protein_binder/input.cif, and names run_hub_examples.py for the seven that prep does package, since that runner pins the model, prep and Diffusers revisions. The Quickstart required list also gains coords_nm. It is a named pipeline argument that partial diffusion and inverse folding read; leaving it in conditioning_features raises "partial and inverse generation require clean coords_nm". A full run ignores it, so the snippet behaves identically.4aa8963 -
Model card: examples are generated with these weights
Verified signature898fa20 -
Regenerate the example designs with the 2026-10-19 weights
Verified signature9ef3f6a -
Model card: 2026-10-19 release, Diffusers branch
Verified signature98ee190 -
Model card: two schedulers, recycle_depths in model_index.json
Verified signature28926d9 -
Split the scheduler per modality, move recycle_depths to the pipeline config, rename feature-embedding keys
Verified signature5db5138 -
Model card: pin the preparation package to a revision
Verified signature8c7b00d
3 days ago
-
Model card: parameter count, fine-tuning note
Verified signaturef13341a -
Model card: new pipeline call signature, 180 trunk passes
Verified signaturec6de904 -
Update to the 2026-10-19 release (LoRA merged); pipeline call signature with explicit inputs
Verified signaturecbd93f5
4 days ago
-
Flatten the transformer state dict: no model. prefix, sequence_head
Verified signatureb21bec6 -
Model card: the autoencoder component is registered as vae
Verified signature7694a9e -
Rename the autoencoder component to vae
Verified signature428e1dc -
Update to the 2026-10-15 unified release (147-class, latent 32)
Verified signatureConverted from nvidia/RFD4-Proteina at 976e0c90 (weights_sha256 5f00cdcd...) with converter revision 0291146a. The sampler recipe is read from the RFD4 repository's own configs/generation/inference_base.yaml rather than a transcribed default, so the scheduler records tsr_matched_noise, t_lim_ode 0.95 and tsr_sigma 12.0. Requires the unreleased rfd4-unified-checkpoint branch and the matching prep package; the existing example outputs are from the previous release and will not reproduce.36e146b
7 days ago
-
Publication layout: model card, provenance, examples, assets
Verified signatureaa612cc -
Publication layout: model card, provenance, examples, assets
Verified signature3b14c84 -
Publication layout: model card, provenance, examples, assets
Verified signature866eab7 -
Publication layout: model card, provenance, examples, assets
Verified signature00bff37 -
Publication layout: model card, provenance, examples, assets
Verified signaturea1a523e -
initial commit
Verified signature2a058c4