Update README.md

README.md

@@ -1,19 +1,31 @@
 ---
-license: apache-2.0
-license_link: https://huggingface.co/GAIR/OpenSWE-32B/blob/main/LICENSE
+license: other
+license_name: qwen
+license_link: https://huggingface.co/GAIR/OpenSWE-32B/LICENSE
 language:
 - en
 pipeline_tag: text-generation
+library_name: transformers
+base_model:
+- Qwen/Qwen2.5-32B
+tags:
+- software
+- environment
+- agent
+- code
 ---
 
-
 # OpenSWE: Efficient SWE Environment Synthesis at Scale
 
-<p align="center">
-  <img src="asset/arxiv-logo.svg" alt="arXiv" width="16" height="16"> <a href="https://arxiv.org/abs/" target="_blank">Paper</a> &nbsp; | &nbsp;
-  <a href="https://github.com/GAIR-NLP/OpenSWE" target="_blank">Code</a> &nbsp; | &nbsp;
-  <a href="https://github.com/GAIR-NLP/OpenSWE" target="_blank">Environments & Scripts</a>
-</p>
+<div align="center">
+
+[![Paper](https://img.shields.io/badge/Paper-PDF-1f6feb.svg)](https://github.com/GAIR-NLP/OpenSWE/blob/main/asset/paper.pdf)
+[![arXiv](https://img.shields.io/badge/arXiv-2601.18418-b31b1b.svg)](https://arxiv.org/pdf/2603.13023)
+[![GitHub](https://img.shields.io/badge/GitHub-Repository-green)](https://github.com/GAIR-NLP/OpenSWE)
+[![Hugging Face](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Dataset-blue)](https://huggingface.co/datasets/GAIR/OpenSWE)
+[![Hugging Face](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Model-blue)](https://huggingface.co/GAIR/OpenSWE-72B)
+
+</div>
 
 <p align="center"> <img src="asset/teaser.png" style="width: 93%;" id="title-icon"> </p>
 </div>
@@ -75,52 +87,6 @@ This repository contains the official implementation of the OpenSWE pipeline—a
 
 Training on OpenSWE alone yields large improvements over SWE-rebench across all model sizes and scaffolds; combining with SWE-rebench further improves 72B (e.g., 68.0% SWE-Agent). Data scaling analysis shows log-linear improvement with no saturation (see paper for curves). General capability evaluation shows gains on code (e.g., HumanEval +29), math (e.g., MATH-500 +12.2 for 72B), and science benchmarks without degrading factual recall.
 
-## Quick Start
-
-### 1. Data schema
-
-Collect your dataset in the following schema:
-
-| Field | Type | Description |
-|-------|------|-------------|
-| `instance_id` | `str` | Unique identifier for the sample. |
-| `repo` | `str` | Full GitHub repo name (e.g., `psf/requests`). |
-| `base_commit` | `str` | SHA of the commit immediately before the PR's first change. |
-| `end_commit` | `str` | SHA of the final commit in the PR. |
-| `problem_statement` | `str` | Issue description or problem to solve. |
-| `patch` | `str` | Diff of changes to functional (non-test) code. |
-| `test_patch` | `str` | Diff of changes to the test suite. |
-| `language` | `str` | Primary programming language of the repo. |
-
-### 2. (Recommended) Prepare system
-
-- Download all git repositories into a _repocache_ directory.
-- Build base Docker images with `scripts/prepare_baseimg.py`.
-
-### 3. Apply patches for SWE-bench evaluation
-
-Before running evaluation, apply:
-
-- **swe-agent.patch** — for [SWE-agent/SWE-agent](https://github.com/SWE-agent/SWE-agent): adds `skip_fetch` and OpenSWE instance fields.
-- **swe-bench-fork.patch** — for [SWE-rebench/SWE-bench-fork](https://github.com/SWE-rebench/SWE-bench-fork): adds `eval_script` support and `OPENSWE_EXIT_CODE` grading.
-
-Replace `/path/to/openswe` with your OpenSWE repo root. On conflicts use `git apply --reject` and resolve `.rej` files. Apply each patch once per repo.
-
-### 4. Configure and run
-
-Edit `examples/run.sh` (set `OPENSWE_ROOT`, `DATA_PATH`, `OUTPUT_DIR`, `SETUP_DIR`, `RESULT_DIR`, `DATA_PATH`, API keys, and `DOCKER_REPOSITORY`), then:
-
-```bash
-bash examples/run.sh
-```
-
-For multi-machine building, see [Parallel Task Execution System](./scripts/parallel).
-
-## Troubleshooting
-
-- **Dataset missing**: Ensure your dataset JSONL exists at the path set in `DATA_PATH`; check schema matches the table above.
-- **Patch conflicts**: Resolve `.rej` files after `git apply --reject` for swe-agent and swe-bench-fork.
-
 ## Acknowledgement
 
 OpenSWE is inspired by [SWE-Rebench](https://arxiv.org/abs/2505.20411) and [SWE-Factory](https://arxiv.org/abs/2506.10954). We thank these teams for their open-source contributions. 
@@ -134,10 +100,13 @@ This project is licensed under AGPL-3.0. See [LICENSE](./LICENSE) for details.
 If you find OpenSWE useful, please cite:
 
 ```bibtex
-@article{openswe2026,
-  title={daVinci-Env: Open SWE Environment Synthesis at Scale},
-  author={Dayuan Fu and Shenyu Wu and Yunze Wu and Zerui Peng and Yaxing Huang and Jie Sun and Ji Zeng and Mohan Jiang and Lin Zhang and Yukun Li and Jiarui Hu and Liming Liu and Jinlong Hou and Pengfei Liu},
-  journal={arXiv preprint},
-  year={2026}
+@misc{fu2026davincienvopensweenvironment,
+      title={daVinci-Env: Open SWE Environment Synthesis at Scale}, 
+      author={Dayuan Fu and Shenyu Wu and Yunze Wu and Zerui Peng and Yaxing Huang and Jie Sun and Ji Zeng and Mohan Jiang and Lin Zhang and Yukun Li and Jiarui Hu and Liming Liu and Jinlong Hou and Pengfei Liu},
+      year={2026},
+      eprint={2603.13023},
+      archivePrefix={arXiv},
+      primaryClass={cs.SE},
+      url={https://arxiv.org/abs/2603.13023}, 
 }
 ```