Fix URLs based on repo transfer and prep for DOI snapshot (#41)

Co-authored-by: Matthew Thompson <thompson.4509@osu.edu>

Author: egrace479

.github/workflows/validate-zenodo.yaml

@@ -0,0 +1,23 @@
+name: Check zenodo metadata
+
+on:
+    push:
+        paths:
+          - '.zenodo.json'
+          - '.github/workflows/validate-zenodo.yaml'
+
+jobs:
+  check-zenodo-metadata:
+
+    runs-on: ubuntu-latest
+
+    steps:
+      - uses: actions/checkout@v6
+      - uses: actions/setup-node@v6
+        with:
+          node-version: '24'
+      - name: Install dependencies
+        run: npm install zenodraft@0.14.1
+      - name: Check .zenodo.json file
+        run: |
+          npx zenodraft metadata validate .zenodo.json

.zenodo.json

@@ -0,0 +1,37 @@
+{
+    "creators": [
+      {
+        "name": "Stevens, Samuel",
+        "orcid": "https://orcid.org/0009-0000-9493-7766",
+        "affiliation": "The Ohio State University"
+      }
+    ],
+    "description": "saev is a package for training sparse autoencoders (SAEs) on vision transformers (ViTs) in PyTorch.",
+    "keywords": [
+      "imageomics",
+      "sparse autoencoders",
+      "interpretability",
+      "computer vision"
+    ],
+    "title": "saev: Sparse Autoencoders for Vision Transformers",
+    "version": "0.1.0",
+    "license": "MIT",
+    "publication_date": "2026-06-18",
+    "grants": [
+        {
+            "id": "021nxhr62::2118240"
+        }
+    ],
+    "related_identifiers": [
+        {
+        "identifier": "10.48550/arXiv.2502.06755",
+        "relation": "isSupplementTo",
+        "resource_type": "publication-preprint"
+        },
+        {
+        "identifier": "10.48550/arXiv.2511.17735",
+        "relation": "isSupplementTo",
+        "resource_type": "publication-preprint"
+        }
+    ]
+}

CITATION.cff

@@ -13,8 +13,8 @@ authors:
     email: samuel.robert.stevens@gmail.com
     orcid: 'https://orcid.org/0009-0000-9493-7766'
     affiliation: The Ohio State University
-repository-code: 'https://github.com/OSU-NLP-Group/saev'
-url: 'https://osu-nlp-group.github.io/saev/'
+repository-code: 'https://github.com/Imageomics/saev'
+url: 'https://imageomics.github.io/saev/'
 repository-artifact: 'https://pypi.org/project/saev/'
 abstract: >-
   saev is a package for training sparse autoencoders (SAEs)
@@ -23,7 +23,14 @@ keywords:
   - sparse autoencoders
   - interpretability
   - computer vision
-license: CC-BY-4.0
-commit: 6a34b6916fda7b04cc2d89749b3ad6425a8f39e6
-date-released: '2025-11-16'
-
+license: MIT
+date-released: '2026-06-18'
+identifiers:
+  - description: "The GitHub release URL of tag v0.1.0."
+    type: url
+    value: "https://github.com/Imageomics/saev/releases/tag/v0.1.0"
+  - description: "The GitHub URL of the commit tagged with v0.1.0."
+    type: url
+    value: "https://github.com/Imageomics/saev/tree/<commit-hash>" # Update on release
+version: 0.1.0
+#doi: <version-agnostic DOI from Zenodo>

LICENSE

@@ -1,5 +1,7 @@
 MIT License
 
+Copyright (c) 2024-2026 Samuel Stevens
+
 Copyright (c) 2023 Joseph Bloom
 
 Permission is hereby granted, free of charge, to any person obtaining a copy

README.md

@@ -2,16 +2,16 @@
 
 ![PyPI Downloads](https://static.pepy.tech/badge/saev)
 ![MIT License](https://img.shields.io/badge/License-MIT-efefef)
-![GitHub Repo stars](https://img.shields.io/github/stars/OSU-NLP-group/saev?style=flat&label=GitHub%20%E2%AD%90)
+![GitHub Repo stars](https://img.shields.io/github/stars/Imageomics/saev?style=flat&label=GitHub%20%E2%AD%90)
 
 Training sparse autoencoders (SAEs) on vision transformers (ViTs), implemented in PyTorch.
 
 ## Docs
 
-- [Docs](https://osu-nlp-group.github.io/saev/api)
-- [Colab Notebook for SAE Inference](https://colab.research.google.com/github/OSU-NLP-Group/saev/blob/main/examples/inference.ipynb)
-- [User guide](https://osu-nlp-group.github.io/saev/api/users/guide)
-- [API reference](https://osu-nlp-group.github.io/saev/api/api/saev/)
+- [Docs](https://imageomics.github.io/saev/api)
+- [Colab Notebook for SAE Inference](https://colab.research.google.com/github/Imageomics/saev/blob/main/examples/inference.ipynb)
+- [User guide](https://imageomics.github.io/saev/api/users/guide)
+- [API reference](https://imageomics.github.io/saev/api/api/saev/)
 
 ## Research
 
@@ -24,16 +24,6 @@ Trained SAE checkpoints are available at:
 
 - [Huggingface Models](https://huggingface.co/collections/osunlp/sae-v-67ab8c4fdf179d117db28195)
 
-If you want to cite the software, please cite it as:
-
-```bib
-@software{stevens2025saev,
-  author = {Stevens, Samuel},
-  license = {CC-BY-4.0},
-  month = apr,
-  title = {{saev}},
-  url = {https://github.com/OSU-NLP-Group/saev},
-  year = {2025}
-}
-```
+## Citation
 
+If you want to cite the software, please use the "Cite this repository" link on the GitHub page, which will provide you with the appropriate citation format.

contrib/birdsong/notebooks/001_explore.py

@@ -270,7 +270,8 @@ def _finalize_df(rows: list[dict[str, object]]):
         )
 
         df = (
-            df.unnest("config/sae", "config/train_data/metadata", separator="/")
+            df
+            .unnest("config/sae", "config/train_data/metadata", separator="/")
             .unnest("config/sae/activation", separator="/")
             .unnest(
                 "config/sae/activation/aux",

contrib/birdsong/notebooks/clips.py

@@ -92,7 +92,8 @@ def add_target(obs: pl.DataFrame, fields: list[str]) -> pl.DataFrame:
         obs = obs.with_columns([pl.col(field).fill_null("unknown") for field in fields])
 
         combos = (
-            obs.select(fields)
+            obs
+            .select(fields)
             .unique(maintain_order=True)  # first-seen ordering
             .with_columns(pl.arange(0, pl.len(), dtype=pl.Int32).alias("target"))
         )
@@ -101,7 +102,8 @@ def add_target(obs: pl.DataFrame, fields: list[str]) -> pl.DataFrame:
 
         target2fields = {
             target: tuple(rest)
-            for target, *rest in obs.unique(pl.col("target"))
+            for target, *rest in obs
+            .unique(pl.col("target"))
             .select("target", *fields)
             .iter_rows()
         }

contrib/freshwater_fish/scripts/make_gallery.py

@@ -74,9 +74,7 @@ def build_features(
     """Collect features that have pre-rendered images on disk."""
     features = []
     available = {
-        int(d.name)
-        for d in images_dpath.iterdir()
-        if d.is_dir() and d.name.isdigit()
+        int(d.name) for d in images_dpath.iterdir() if d.is_dir() and d.name.isdigit()
     }
 
     for row in var_df.iter_rows(named=True):
@@ -275,15 +273,49 @@ def build_features(
 
 def main():
     parser = argparse.ArgumentParser(description=__doc__)
-    parser.add_argument("--run", type=pathlib.Path, required=True, help="Run directory (e.g. /fs/ess/.../runs/um6hbn05)")
-    parser.add_argument("--shard", type=str, required=True, help="Shard ID (e.g. 8692dfa9)")
-    parser.add_argument("--dataset", type=pathlib.Path, default=None, help="Dataset root (segfolder with images/ dir, for label ordering)")
-    parser.add_argument("--split", type=str, default="validation", help="Dataset split name (default: validation)")
-    parser.add_argument("--hf-config", type=str, default="trait_segmentation", help="HuggingFace FishVista config name")
-    parser.add_argument("--hf-split", type=str, default="val", help="HuggingFace split name (default: val)")
-    parser.add_argument("--out", type=pathlib.Path, default=pathlib.Path("gallery.html"), help="Output HTML path")
+    parser.add_argument(
+        "--run",
+        type=pathlib.Path,
+        required=True,
+        help="Run directory (e.g. /fs/ess/.../runs/um6hbn05)",
+    )
+    parser.add_argument(
+        "--shard", type=str, required=True, help="Shard ID (e.g. 8692dfa9)"
+    )
+    parser.add_argument(
+        "--dataset",
+        type=pathlib.Path,
+        default=None,
+        help="Dataset root (segfolder with images/ dir, for label ordering)",
+    )
+    parser.add_argument(
+        "--split",
+        type=str,
+        default="validation",
+        help="Dataset split name (default: validation)",
+    )
+    parser.add_argument(
+        "--hf-config",
+        type=str,
+        default="trait_segmentation",
+        help="HuggingFace FishVista config name",
+    )
+    parser.add_argument(
+        "--hf-split",
+        type=str,
+        default="val",
+        help="HuggingFace split name (default: val)",
+    )
+    parser.add_argument(
+        "--out",
+        type=pathlib.Path,
+        default=pathlib.Path("gallery.html"),
+        help="Output HTML path",
+    )
     parser.add_argument("--quality", type=int, default=80, help="JPEG quality (0-100)")
-    parser.add_argument("--title", type=str, default="", help="Gallery subtitle/description")
+    parser.add_argument(
+        "--title", type=str, default="", help="Gallery subtitle/description"
+    )
     args = parser.parse_args()
 
     inference_dpath = args.run / "inference" / args.shard
@@ -313,7 +345,10 @@ def main():
     n_imgs = sum(len(f["images"]) for f in features)
     logger.info("Total images: %d", n_imgs)
 
-    title = args.title or f"SAE run {args.run.name}, shard {args.shard} | {len(features)} features, {n_imgs} images"
+    title = (
+        args.title
+        or f"SAE run {args.run.name}, shard {args.shard} | {len(features)} features, {n_imgs} images"
+    )
     html = HTML_TEMPLATE.replace("FEATURES_JSON", json.dumps(features))
     html = html.replace("TITLE_PLACEHOLDER", title)
     html = html.replace("RUN_ID_PLACEHOLDER", args.run.name)

contrib/interactive_interp/semseg/interactive.py

@@ -570,7 +570,8 @@ def make_upsampled_pred(
         logits_WHC: Float[Tensor, "width height classes"],
     ) -> Uint8[Tensor, "width height"]:
         return (
-            torch.nn.functional.interpolate(
+            torch.nn.functional
+            .interpolate(
                 logits_WHC.max(axis=-1).indices.view((1, 1, 16, 16)).float(),
                 scale_factor=14,
             )

contrib/interactive_interp/semseg/training.py

@@ -358,7 +358,8 @@ def __getitem__(self, i: int) -> dict[str, object]:
 
         pw, ph = self.patch_size_px
         patch_labels = (
-            einops.rearrange(pixel_labels, "(w pw) (h ph) -> w h (pw ph)", pw=pw, ph=ph)
+            einops
+            .rearrange(pixel_labels, "(w pw) (h ph) -> w h (pw ph)", pw=pw, ph=ph)
             .mode(axis=-1)
             .values
         )