Update README.md

README.md

@@ -4,6 +4,107 @@ base_model:
 - Qwen/Qwen2.5-7B-Instruct
 pipeline_tag: question-answering
 tags:
-- agent
+- LLM Agent
+- Knowledge Graph
+- Question Answering
+- Reasoning
 ---
-GraphWalker-7B.
+
+# GraphWalker-7B
+
+[**📄 Paper (arXiv:2603.28533)**](https://arxiv.org/abs/2603.28533) | [**💻 GitHub**](https://github.com/XuShuwenn/GraphWalker) | [**🤗 Model**](https://huggingface.co/xushuwen23/GraphWalker-7B)
+
+**GraphWalker-7B** is a specialized large language model fine-tuned from [Qwen2.5-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-7B-Instruct) for **Agentic Knowledge Graph Question Answering (KGQA)**. GraphWalker treats multi-hop KGQA as a sequential graph-walking process, learning to navigate knowledge graphs via synthetic trajectory curriculum — achieving strong generalization with a single, compact 7B model.
+
+---
+
+## 🌟 Overview
+
+Multi-hop KGQA requires reasoning over complex, interconnected entities across a knowledge graph. Existing approaches either rely on rigid retrieval pipelines or expensive multi-module LLM orchestration. **GraphWalker** addresses this with two core ideas:
+
+1. **Agentic Graph Walking:** The model acts as an agent that iteratively traverses a KG, selecting which edges to follow at each step based on the question and accumulated context — effectively decomposing multi-hop questions into a series of local, grounded decisions.
+
+2. **Synthetic Trajectory Curriculum (STC):** Instead of relying on expensive human-annotated reasoning chains, GraphWalker is trained on *synthetically generated* graph-walking trajectories. The curriculum is structured to progressively increase trajectory complexity, enabling the model to internalize robust multi-hop strategies.
+
+Together, these designs allow GraphWalker-7B to outperform much larger models and complex multi-agent systems on standard KGQA benchmarks, while remaining efficient at inference time.
+
+---
+
+## 🔑 Key Features
+
+- **Agentic KG Navigation:** Frames KGQA as an iterative, step-by-step graph traversal rather than a single-shot retrieval-and-generate task.
+- **Synthetic Trajectory Curriculum:** Trains on automatically constructed walking trajectories with progressively increasing difficulty, eliminating the need for costly human annotation.
+- **Lazy Greedy Search with Frontier Expansion:** An efficient beam-search-style graph traversal algorithm that maximizes information gain while keeping context size tractable.
+- **Single-Model Efficiency:** Achieves competitive performance with a single 7B model, without multi-agent overhead.
+- **Broad KG Compatibility:** Designed to generalize across standard KGQA benchmarks (e.g., WebQSP, CWQ, GrailQA).
+
+---
+
+## 🛠️ Usage
+
+### 1. Environment Setup
+
+```bash
+pip install vllm transformers
+```
+
+### 2. Download the Model
+
+```bash
+# Via huggingface-cli
+huggingface-cli download <your-org>/GraphWalker-7B --local-dir ./GraphWalker-7B
+```
+
+### 3. Inference with vLLM (Recommended)
+
+**Start the vLLM server:**
+
+```bash
+vllm serve "./GraphWalker-7B" \
+    --host 0.0.0.0 --port 22240 \
+    --served-model-name graphwalker-7b \
+    --gpu-memory-utilization 0.9 \
+    --dtype auto \
+    --chat-template "./GraphWalker-7B/chat_template.jinja"
+```
+
+
+
+---
+
+## 📈 Evaluation Results
+
+| Method | Backbone | CWQ EM | CWQ F1 | WebQSP EM | WebQSP F1 |
+|:---|:---|:---:|:---:|:---:|:---:|
+| **GraphWalker** | | | | | |
+| †Vanilla Agent | Qwen2.5-7B-Instruct | 40.7 | 33.2 | 68.4 | 66.1 |
+| †Vanilla Agent | GPT-4o-mini | 63.4 | 60.3 | 79.6 | 70.6 |
+| †Vanilla Agent | DeepSeek-V3.2 | 69.8 | 63.5 | 76.7 | 71.8 |
+| GraphWalker-7B-SFT | Qwen2.5-7B-Instruct | 68.3 | 63.2 | 82.0 | 79.1 |
+| GraphWalker-3B-SFT-RL | Qwen2.5-3B-Instruct | 70.9 | 65.2 | 83.5 | 81.7 |
+| GraphWalker-8B-SFT-RL | LLaMA3.1-8B-Instruct | <u>78.5</u> | 69.6 | <u>88.2</u> | <u>84.5</u> |
+| **GraphWalker-7B-SFT-RL** | **Qwen2.5-7B-Instruct** | **79.6** | **74.2** | **91.5** | **88.6** |
+
+---
+
+## 📝 Citation
+
+If you use GraphWalker-7B or find this work helpful, please cite:
+
+```bibtex
+@misc{xu2026graphwalkeragenticknowledgegraph,
+      title={GraphWalker: Agentic Knowledge Graph Question Answering via Synthetic Trajectory Curriculum}, 
+      author={Shuwen Xu and Yao Xu and Jiaxiang Liu and Chenhao Yuan and Wenshuo Peng and Jun Zhao and Kang Liu},
+      year={2026},
+      eprint={2603.28533},
+      archivePrefix={arXiv},
+      primaryClass={cs.CL},
+      url={https://arxiv.org/abs/2603.28533}, 
+}
+```
+
+---
+
+## 📄 License
+
+This model is released under the [Apache 2.0 License](https://www.apache.org/licenses/LICENSE-2.0), consistent with the base model [Qwen2.5-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-7B-Instruct).