Add verified flow/AE pair and pinned public source

LICENSE

@@ -0,0 +1,28 @@
+BSD 3-Clause License
+
+Copyright (c) 2026, Institute for Protein Design, University of Washington
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+* Redistributions of source code must retain the above copyright notice, this
+  list of conditions and the following disclaimer.
+
+* Redistributions in binary form must reproduce the above copyright notice,
+  this list of conditions and the following disclaimer in the documentation
+  and/or other materials provided with the distribution.
+
+* Neither the name of the copyright holder nor the names of its
+  contributors may be used to endorse or promote products derived from
+  this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
+FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

README.md

@@ -0,0 +1,26 @@
+---
+license: bsd-3-clause
+tags:
+- biology
+- protein-design
+---
+# RFD4-Proteina
+
+Private mirror of the default released flow and autoencoder checkpoint pair.
+The native `flow_checkpoints/` and `ae_checkpoints/` layout matches `DATA_PATH` in RFD4.
+Both checkpoints must be used together. The `.sha256` sidecars record verified release checksums.
+
+The `source/` archive is an exact snapshot of RosettaCommons/RFD4-Proteina main at
+`66e4394d38a04a0b901a76507cd5528af2d7f845`. It has the same file tree as the private
+release branch checked on October 5, 2026. See `provenance.json` for identifiers.
+
+## Installation status
+
+This source revision still references the removed AtomWorks `release/atomworks-3-0`
+branch. AtomWorks 3.0.0 is available on PyPI; public RFD4 PR #5 contains the dependency
+update. The source archive is preserved unmodified for provenance and is not claimed
+to be a clean-install-ready release. Update the pinned source after release validation.
+
+These files are native RFD4 checkpoints, not Transformers weights or converted
+`from_pretrained` bundles. Use the RFD4 CLI and the documented compatible runtime.
+Code and third-party components retain their included license notices.

ae_checkpoints/2026_09_04_release_ae_z16_400k-ema.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b5ab42dda2702bf29f5d0b15e652cf1a88d2d8254ca81a4e7ec4aac55cf4147a
+size 3075639017

ae_checkpoints/2026_09_04_release_ae_z16_400k-ema.ckpt.sha256

@@ -0,0 +1 @@
+b5ab42dda2702bf29f5d0b15e652cf1a88d2d8254ca81a4e7ec4aac55cf4147a  2026_09_04_release_ae_z16_400k-ema.ckpt

flow_checkpoints/chk_epoch_00000169_step_000000340000-ema.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d95c187520650824be6da9cd348d3d1a09127533047b8a0ed7ffa189f4123747
+size 4200426717

flow_checkpoints/chk_epoch_00000169_step_000000340000-ema.ckpt.sha256

@@ -0,0 +1 @@
+d95c187520650824be6da9cd348d3d1a09127533047b8a0ed7ffa189f4123747  chk_epoch_00000169_step_000000340000-ema.ckpt

inference.yaml

@@ -0,0 +1,62 @@
+# Default configs for inference through generate.py
+# The defaults are geared more towards ease-of-use by designers, so running with
+# the existing analysis scripts will necessitate overrides
+
+defaults:
+  - _self_
+  - inference_experiment: inference_on_user_inputs # Default for design
+  - recycle_strategy: thinking_medium
+
+# NOTE: The experiment configs can override values specified in this config
+
+run_name_: inference
+verbose: false # Set true for routine warnings and detailed diagnostic output.
+
+out_dir: null  # If null, defaults to ./runs/<run_name>. For production, set to an external
+               # path (e.g. /shared/runs/experiment_name). SLURM scripts set this
+               # automatically. All outputs (samples, metrics) go under this directory.
+
+# Current SOTA; generation and regression tooling resolve the immutable checkpoint from here.
+# Regenerate the inference baselines when advancing it; use CLI overrides for one-off checkpoints.
+ckpt_name: chk_epoch_00000169_step_000000340000-ema.ckpt
+ckpt_path: ${oc.env:DATA_PATH}/flow_checkpoints
+
+# Only applies when a pre-oracle design filter is active (i.e. its tag gate matched).
+# Determines whether to delete structure files for designs that fail this filter
+delete_failing_pre_oracle: True
+
+autoencoder_ckpt_path: ${oc.env:DATA_PATH}/ae_checkpoints/2026_09_04_release_ae_z16_400k-ema.ckpt
+
+num_replicates: 1 # Number of times to replicate each example
+
+# Token counts to pad batches to for torch.compile. Multiples of 16; null disables.
+compile_lengths: [128, 160, 192, 224, 256, 288, 320, 352, 384, 448, 512, 640, 768, 896, 1024]
+
+precision: bf16-mixed
+
+batch_size: 1
+subset_to_examples: null # If specified, only run inference on these example indices
+prevalidate_inputs: false # Whether to sanity-check each unique input once prior to generation (fail fast on bad inputs).
+seed: null # Will set a random seed if null
+torch_deterministic: false  # Locks cuDNN/cuBLAS to deterministic algorithms; significantly reduces throughput
+skip_existing: true # If true, skip examples whose output structure files already exist in the output directory
+save_chain_of_interest: false # If true, write per-design chain-of-interest CIFs (only needed for foldseek clustering).
+compress_outputs: false # Compress persistent structure outputs, including refolds and CIF trajectories.
+metrics_tags: [] # If specified, determines which metrics are computed during inference
+
+num_workers: 4  # Number of dataloader workers
+
+hardware:
+  accelerator: gpu
+  ngpus_per_node_: 1
+  nnodes_: 1
+
+# Cluster job settings (used by launch_infer_from_local.sh)
+ncpus_: 24  # Number of CPUs per task
+gen_njobs: 1  # Number of parallel generation jobs per config
+eval_njobs: 1  # Number of parallel evaluation jobs per config
+job_id: null  # Set by launch_generate.sh; used to disambiguate per-job output files
+
+hydra:
+  run:
+    dir: ./hydra_outputs

provenance.json

@@ -0,0 +1,10 @@
+{
+  "public_repository": "https://github.com/RosettaCommons/RFD4-Proteina",
+  "public_commit": "66e4394d38a04a0b901a76507cd5528af2d7f845",
+  "public_tree": "04af6de14c3c6eafe6481825ea2bc3c7649d6fbb",
+  "source_archive_sha256": "d7043bfbb6ab970ab88049801da10a36755aeea9dfd8e4f3e8757d948be1e144",
+  "checkpoints": {
+    "flow_checkpoints/chk_epoch_00000169_step_000000340000-ema.ckpt": "d95c187520650824be6da9cd348d3d1a09127533047b8a0ed7ffa189f4123747",
+    "ae_checkpoints/2026_09_04_release_ae_z16_400k-ema.ckpt": "b5ab42dda2702bf29f5d0b15e652cf1a88d2d8254ca81a4e7ec4aac55cf4147a"
+  }
+}

source/RFD4-Proteina-66e4394d.tar.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d7043bfbb6ab970ab88049801da10a36755aeea9dfd8e4f3e8757d948be1e144
+size 36057565