Add model card for AutoSelection SAE

README.md

@@ -0,0 +1,44 @@
+---
+license: apache-2.0
+pipeline_tag: feature-extraction
+---
+
+# AutoSelection Sparse Autoencoder (SAE)
+
+This repository contains a Sparse Autoencoder (SAE) checkpoint used in the paper "[From Instance Selection to Fixed-Pool Data Recipe Search for Supervised Fine-Tuning](https://huggingface.co/papers/2605.12944)".
+
+## Model Description
+
+AutoSelection is a budgeted solver for fixed-pool data recipe search for Supervised Fine-Tuning (SFT). This SAE model is utilized by the AutoSelection framework to consume features during cold-start scoring and subset-state construction. These signals help the search controller discover executable data-curation recipes that construct high-quality training subsets.
+
+- **Repository:** [https://github.com/w253/AutoSelection](https://github.com/w253/AutoSelection)
+- **Paper:** [From Instance Selection to Fixed-Pool Data Recipe Search for Supervised Fine-Tuning](https://huggingface.co/papers/2605.12944)
+
+## Usage
+
+As described in the project's [GitHub README](https://github.com/w253/AutoSelection), you can download the SAE checkpoint using the `huggingface-cli`:
+
+```bash
+# Example for downloading to a local directory
+huggingface-cli download <REPLACE_WITH_REPO_ID> \
+  --local-dir models/sae/checkpoint
+
+# Point the SAE_PATH to the specific layer directory
+export SAE_PATH=models/sae/checkpoint/layer.27
+```
+
+The AutoSelection engine expects the `SAE_PATH` environment variable to point to the directory containing the SAE artifact (e.g., `layers.27`).
+
+## Citation
+
+```bibtex
+@misc{wu2026instanceselectionfixedpooldata,
+      title={From Instance Selection to Fixed-Pool Data Recipe Search for Supervised Fine-Tuning},
+      author={Haodong Wu and Jiahao Zhang and Lijie Hu and Yongqi Zhang},
+      year={2026},
+      eprint={2605.12944},
+      archivePrefix={arXiv},
+      primaryClass={cs.LG},
+      url={https://arxiv.org/abs/2605.12944},
+}
+```