Add PACE-Bench dataset card

README.md

@@ -1,3 +1,145 @@
 ---
 license: mit
+language:
+  - en
+pretty_name: PACE-Bench
+tags:
+  - benchmark
+  - physics
+  - code-generation
+  - agents
+  - self-evolving-agents
+  - simulation
 ---
+
+# PACE-Bench
+
+**PACE-Bench: Benchmarking Physics Adaptation via Code Evolution in Dynamic Environments** evaluates whether an agent can adapt an executable physical design after its environment changes.
+
+## Overview
+
+PACE-Bench contains **144 source-to-target adaptation pairs across six physics domains**. Each pair preserves the same task goal and interface:
+
+1. A code-driven design succeeds in a source environment.
+2. The same design fails in a mutated target environment.
+3. The agent receives diagnostic sandbox feedback and iteratively revises the design.
+4. The adapted design must succeed under the target physics.
+
+| Property | Count |
+| --- | ---: |
+| Physics domains | 6 |
+| Base tasks | 36 |
+| Environments per task | 5 |
+| Evaluation environments | 180 |
+| Source-to-target pairs | 144 |
+
+The six domains cover **statics, kinematics, dynamics, granular/fluid interaction, control, and exotic physics**.
+
+## Repository contents
+
+```text
+tasks/
+├── Category1_Statics_Equilibrium/
+├── Category2_Kinematics_Linkages/
+├── Category3_Dynamics_Energy/
+├── Category4_Granular_FluidInteraction/
+├── Category5_Cybernetics_Control/
+├── Category6_ExoticPhysics/
+└── primitives_api.json
+```
+
+Each domain contains six tasks. Every task directory includes:
+
+| File | Role |
+| --- | --- |
+| `agent.py` | Source and four target reference solutions |
+| `environment.py` | Box2D world, primitives, and mutable physics |
+| `evaluator.py` | Success criteria, scores, constraints, and metrics |
+| `feedback.py` | Diagnostic feedback derived from measured metrics |
+| `prompt.py` | Task description, exposed values, and primitive API |
+| `renderer.py` | Evaluation-neutral visualization |
+| `stages.py` | Four target mutations and visibility-aware prompt updates |
+
+These files are **executable benchmark definitions**, not a conventional row-based dataset. The Hugging Face Dataset Viewer is therefore not the main interface.
+
+## Download
+
+Download the complete snapshot:
+
+```python
+from huggingface_hub import snapshot_download
+
+dataset_path = snapshot_download(
+    repo_id="YuhaoZhan/PACE-Bench",
+    repo_type="dataset",
+)
+print(dataset_path)
+```
+
+Download only the executable tasks:
+
+```python
+from huggingface_hub import snapshot_download
+
+tasks_path = snapshot_download(
+    repo_id="YuhaoZhan/PACE-Bench",
+    repo_type="dataset",
+    allow_patterns=["tasks/**"],
+)
+```
+
+You can also clone the dataset repository:
+
+```bash
+git clone https://huggingface.co/datasets/YuhaoZhan/PACE-Bench
+```
+
+## Running the benchmark
+
+The task definitions depend on the PACE-Bench runtime, evaluator, and shared prompt infrastructure. Install and run the full benchmark from the official code repository:
+
+```bash
+git clone https://github.com/thunlp/PACE-Bench.git
+cd PACE-Bench
+conda create -n pace-bench python=3.10 -y
+conda activate pace-bench
+python -m pip install -r requirements.txt
+
+pace-bench list --task S_01
+pace-bench validate --task S_01
+```
+
+See the [PACE-Bench GitHub repository](https://github.com/thunlp/PACE-Bench) for evaluation commands, supported providers and methods, result reporting, and coding-agent evaluation.
+
+## Intended use
+
+PACE-Bench is intended for:
+
+- evaluating adaptation after controlled physical environment changes;
+- studying feedback-driven code evolution and self-evolving agents;
+- comparing context-, memory-, search-, and parameter-based methods;
+- analyzing physical reasoning, redesign, exploration, and convergence failures.
+
+## Scope and limitations
+
+- The current release focuses on **2D rigid-body systems in Box2D**.
+- It does not cover 3D or deformable physics, full fluids, perception, navigation, or multi-agent coordination.
+- Prompts and diagnostic feedback are currently English-only.
+- Generated solutions should be evaluated on a dedicated host without unrelated credentials.
+
+## License
+
+PACE-Bench is released under the **MIT License**.
+
+## Citation
+
+The paper citation will be added when the public preprint is available. Until then, please cite the project repository:
+
+```bibtex
+@misc{zhan2026pacebench,
+  title        = {PACE-Bench: Benchmarking Physics Adaptation via Code Evolution in Dynamic Environments},
+  author       = {Zhan, Yuhao and He, Bingxiang and Tang, Zecong and Xiao, Chaojun},
+  year         = {2026},
+  howpublished = {\url{https://github.com/thunlp/PACE-Bench}}
+}
+```