Add model card and metadata

README.md

@@ -1,3 +1,31 @@
----
-license: mit
----
+---
+license: mit
+library_name: transformers
+pipeline_tag: text-generation
+---
+
+# SmartSearch: Process Reward-Guided Query Refinement for Search Agents
+
+This repository contains the model for **SmartSearch**, introduced in the paper [SmartSearch: Process Reward-Guided Query Refinement for Search Agents](https://huggingface.co/papers/2601.04888).
+
+SmartSearch is a framework designed to enhance the reasoning capabilities of search agents by optimizing intermediate search queries. It introduces two key mechanisms:
+1. **Process rewards**: Providing fine-grained supervision for query quality through Dual-Level Credit Assessment.
+2. **Query refinement**: Selectively refining low-quality search queries to improve overall search efficiency and retrieval results.
+
+The model is trained using a three-stage curriculum learning framework encompassing imitation, alignment, and generalization.
+
+## Resources
+
+- **Paper:** [SmartSearch: Process Reward-Guided Query Refinement for Search Agents](https://huggingface.co/papers/2601.04888)
+- **Repository:** [GitHub - MYVAE/SmartSearch](https://github.com/MYVAE/SmartSearch)
+
+## Citation
+
+```bibtex
+@article{smartsearch2026,
+  title={SmartSearch: Process Reward-Guided Query Refinement for Search Agents},
+  author={Tongyu Wen and Guanting Dong and Zhicheng Dou},
+  journal={arXiv preprint arXiv:2601.04888},
+  year={2026}
+}
+```