Align with the current version on GitHub

Author: helloml0326

README.md

@@ -16,7 +16,7 @@
 [![PyPI](https://img.shields.io/badge/pypi-v0.2.0-blue?logo=pypi)](https://pypi.org/project/py-openjudge/)
 [![Documentation](https://img.shields.io/badge/docs-online-blue?logo=readthedocs&logoColor=white)](https://modelscope.github.io/OpenJudge/)
 
-[📖 Documentation](https://modelscope.github.io/OpenJudge/) | [🤝 Contributing](./docs/community/contributing.md) | [中文](./README_zh.md)
+[📖 Documentation](https://modelscope.github.io/OpenJudge/) | [🤝 Contributing](https://modelscope.github.io/OpenJudge/community/contributing/) | [中文](./README_zh.md)
 
 </div>
 
@@ -89,15 +89,15 @@ Access **50+ production-ready graders** featuring a comprehensive taxonomy, rigo
 </tr>
 </table>
 
-- 🌐 **Multi-Scenario Coverage:** Extensive support for diverse domains including Agent, text, code, math, and multimodal tasks. 👉  [Explore Supported Scenarios](./docs/built_in_graders/overview.md)
-- 🔄 **Holistic Agent Evaluation:** Beyond final outcomes, we assess the entire lifecycle—including trajectories, Memory, Reflection, and Tool Use. 👉  [Agent Lifecycle Evaluation](./docs/built_in_graders/agent_graders.md)
+- 🌐 **Multi-Scenario Coverage:** Extensive support for diverse domains including Agent, text, code, math, and multimodal tasks. 👉  [Explore Supported Scenarios](https://modelscope.github.io/OpenJudge/built_in_graders/overview/)
+- 🔄 **Holistic Agent Evaluation:** Beyond final outcomes, we assess the entire lifecycle—including trajectories, Memory, Reflection, and Tool Use. 👉  [Agent Lifecycle Evaluation](https://modelscope.github.io/OpenJudge/built_in_graders/agent_graders/)
 - ✅ **Quality Assurance:** Every grader comes with benchmark datasets and pytest integration for validation. 👉  [View Benchmark Datasets](https://huggingface.co/datasets/agentscope-ai/OpenJudge)
 
 
 ### 🛠️ Flexible Grader Building Methods
 Choose the build method that fits your requirements:
-* **Customization:** Easily extend or modify pre-defined graders to fit your specific needs.  👉 [Custom Grader Development Guide](./docs/building_graders/create_custom_graders.md)
-* **Data-Driven Rubrics:** Have a few examples but no clear rules? Use our tools to automatically generate white-box evaluation criteria (Rubrics) based on your data.👉 [Automatic Rubric Generation Tutorial](./docs/building_graders/generate_graders_from_data.md)
+* **Customization:** Easily extend or modify pre-defined graders to fit your specific needs.  👉 [Custom Grader Development Guide](https://modelscope.github.io/OpenJudge/building_graders/create_custom_graders/)
+* **Data-Driven Rubrics:** Have a few examples but no clear rules? Use our tools to automatically generate white-box evaluation criteria (Rubrics) based on your data.👉 [Automatic Rubric Generation Tutorial](https://modelscope.github.io/OpenJudge/building_graders/generate_graders_from_data/)
 * **Training Judge Models ( Coming Soon🚀):** For high-scale and specialized scenarios, we are developing the capability to train dedicated Judge models. Support for SFT, Bradley-Terry models, and Reinforcement Learning workflows is on the way to help you build high-performance, domain-specific graders.
 
 
@@ -122,7 +122,7 @@ We're actively building seamless connectors for mainstream observability platfor
 pip install py-openjudge
 ```
 
-> 💡 More installation methods can be found in the [Quickstart Guide](./docs/get_started/quickstart.md).
+> 💡 More installation methods can be found in the [Quickstart Guide](https://modelscope.github.io/OpenJudge/get_started/quickstart/#installation).
 
 ---
 
@@ -156,7 +156,7 @@ if __name__ == "__main__":
     asyncio.run(main())
 ```
 
-> 📚 Complete Quickstart can be found in the [Quickstart Guide](./docs/get_started/quickstart.md).
+> 📚 Complete Quickstart can be found in the [Quickstart Guide](https://modelscope.github.io/OpenJudge/get_started/quickstart/).
 
 ---
 
@@ -184,7 +184,7 @@ We love your input! We want to make contributing to OpenJudge as easy and transp
 > **📝 Improving Docs** — Clearer explanations or better examples are always welcome  
 > **💡 Proposing Features** — Have ideas for new integrations? Let's discuss!
 
-📖 See full [Contributing Guidelines](./docs/community/contributing.md) for coding standards and PR process.
+📖 See full [Contributing Guidelines](https://modelscope.github.io/OpenJudge/community/contributing/) for coding standards and PR process.
 
 ---
 
@@ -204,7 +204,7 @@ pip install rm-gallery
 
 We preserved the source code of **v0.1.7 (the latest v0.1.x release)** in the [`v0.1.7-legacy` branch](https://github.com/modelscope/OpenJudge/tree/v0.1.7-legacy).
 
-- **Migrate to v0.2.0 (recommended)**: follow the **[Installation](#-installation)** section above, then walk through **[Quickstart](#-quickstart)** (or the full [Quickstart Guide](./docs/get_started/quickstart.md)) to update your imports / usage.
+- **Migrate to v0.2.0 (recommended)**: follow the **[Installation](#-installation)** section above, then walk through **[Quickstart](#-quickstart)** (or the full [Quickstart Guide](https://modelscope.github.io/OpenJudge/get_started/quickstart/)) to update your imports / usage.
 
 If you run into migration issues, please [open an issue](https://github.com/modelscope/OpenJudge/issues) with your minimal repro and current version.
 

README_zh.md

@@ -16,7 +16,7 @@
 [![PyPI](https://img.shields.io/badge/pypi-v0.2.0-blue?logo=pypi)](https://pypi.org/project/py-openjudge/)
 [![Documentation](https://img.shields.io/badge/docs-online-blue?logo=readthedocs&logoColor=white)](https://modelscope.github.io/OpenJudge/)
 
-[📖 文档](https://modelscope.github.io/OpenJudge/) | [🤝 贡献指南](./docs/community/contributing.md) | [English](./README.md)
+[📖 文档](https://modelscope.github.io/OpenJudge/) | [🤝 贡献指南](https://modelscope.github.io/OpenJudge/community/contributing/) | [English](./README.md)
 
 </div>
 
@@ -89,15 +89,15 @@ OpenJudge 将评估指标和奖励信号统一为标准化的 **Grader** 接口
 </tr>
 </table>
 
-- 🌐 **多场景覆盖：** 广泛支持包括智能体、文本、代码、数学和多模态任务在内的多种领域。→ [探索支持的场景](./docs/built_in_graders/overview.md)
-- 🔄 **全面的智能体评估：** 不仅评估最终结果，我们还评估整个生命周期——包括轨迹、记忆、反思和工具使用。→ [智能体生命周期评估](./docs/built_in_graders/agent_graders.md)
+- 🌐 **多场景覆盖：** 广泛支持包括智能体、文本、代码、数学和多模态任务在内的多种领域。→ [探索支持的场景](https://modelscope.github.io/OpenJudge/built_in_graders/overview/)
+- 🔄 **全面的智能体评估：** 不仅评估最终结果，我们还评估整个生命周期——包括轨迹、记忆、反思和工具使用。→ [智能体生命周期评估](https://modelscope.github.io/OpenJudge/built_in_graders/agent_graders/)
 - ✅ **质量保证：** 每个评分器都配有基准数据集和 pytest 集成用于验证。→ [查看基准数据集](https://huggingface.co/datasets/agentscope-ai/OpenJudge)
 
 
 ### 🛠️ 灵活的评分器构建方法
 选择适合您需求的构建方法：
-* **自定义：** 轻松扩展或修改预定义的评分器以满足您的特定需求。👉 [自定义评分器开发指南](./docs/building_graders/create_custom_graders.md)
-* **数据驱动的评分标准：** 有一些示例但没有明确规则？使用我们的工具根据您的数据自动生成白盒评估标准（Rubrics）。👉 [自动评分标准生成教程](./docs/building_graders/generate_graders_from_data.md)
+* **自定义：** 轻松扩展或修改预定义的评分器以满足您的特定需求。👉 [自定义评分器开发指南](https://modelscope.github.io/OpenJudge/building_graders/create_custom_graders/)
+* **数据驱动的评分标准：** 有一些示例但没有明确规则？使用我们的工具根据您的数据自动生成白盒评估标准（Rubrics）。👉 [自动评分标准生成教程](https://modelscope.github.io/OpenJudge/building_graders/generate_graders_from_data/)
 * **训练评判模型（即将推出🚀）：** 对于大规模和专业化场景，我们正在开发训练专用评判模型的能力。SFT、Bradley-Terry 模型和强化学习工作流的支持即将推出，帮助您构建高性能、领域特定的评分器。
 
 
@@ -122,7 +122,7 @@ OpenJudge 将评估指标和奖励信号统一为标准化的 **Grader** 接口
 pip install py-openjudge
 ```
 
-> 💡 更多安装方法可在 [快速开始指南](./docs/get_started/quickstart.md) 中找到。
+> 💡 更多安装方法可在 [快速开始指南](https://modelscope.github.io/OpenJudge/get_started/quickstart/) 中找到。
 
 ---
 
@@ -156,7 +156,7 @@ if __name__ == "__main__":
     asyncio.run(main())
 ```
 
-> 📚 完整的快速开始内容可在 [快速开始指南](./docs/get_started/quickstart.md) 中找到。
+> 📚 完整的快速开始内容可在 [快速开始指南](https://modelscope.github.io/OpenJudge/get_started/quickstart/) 中找到。
 
 ---
 
@@ -184,7 +184,7 @@ if __name__ == "__main__":
 > **📝 改进文档** — 更清晰的解释或更好的示例总是受欢迎的  
 > **💡 提议新功能** — 有新集成的想法？让我们讨论！
 
-📖 查看完整的 [贡献指南](./docs/community/contributing.md) 了解编码标准和 PR 流程。
+📖 查看完整的 [贡献指南](https://modelscope.github.io/OpenJudge/community/contributing/) 了解编码标准和 PR 流程。
 
 ---
 
@@ -202,7 +202,7 @@ pip install rm-gallery
 
 我们在 [`v0.1.7-legacy` 分支](https://github.com/modelscope/OpenJudge/tree/v0.1.7-legacy) 中保留了 **v0.1.7（最新的 v0.1.x 版本）** 的源代码。
 
-- **迁移到 v0.2.0（推荐）**：按照上方的 **[安装](#-安装)** 章节操作，然后浏览 **[快速开始](#-快速开始)**（或完整的 [快速开始指南](./docs/get_started/quickstart.md)）来更新您的导入/用法。
+- **迁移到 v0.2.0（推荐）**：按照上方的 **[安装](#-安装)** 章节操作，然后浏览 **[快速开始](#-快速开始)**（或完整的 [快速开始指南](https://modelscope.github.io/OpenJudge/get_started/quickstart/)）来更新您的导入/用法。
 
 如果您遇到迁移问题，请 [提交 issue](https://github.com/modelscope/OpenJudge/issues) 并附上您的最小复现代码和当前版本。
 

docs/get_started/core_concepts.md

@@ -50,7 +50,7 @@ Flexible [data mapping](#data-mapping) ensures your existing data formats can be
 
 Based on your evaluation purpose, you can choose the appropriate assessment approach:
 
-- **Code-Based graders**: For objective, quantitative evaluations using the [FunctionGrader](../../openjudge/graders/function_grader.py), these graders use predefined functions or algorithms to compute scores. These graders are deterministic and fast, ideal for metrics like exact match, format validation, or simple checks. This approach is best suited for quantitative analysis where you need consistent, reproducible results based on mathematical or logical operations.
+- **Code-Based graders**: For objective, quantitative evaluations using the [FunctionGrader](https://github.com/modelscope/OpenJudge/blob/main/openjudge/graders/function_grader.py), these graders use predefined functions or algorithms to compute scores. These graders are deterministic and fast, ideal for metrics like exact match, format validation, or simple checks. This approach is best suited for quantitative analysis where you need consistent, reproducible results based on mathematical or logical operations.
 
 ```python
 # A simple function grader that checks if response contains reference answer
@@ -60,7 +60,7 @@ def contains_reference(response, reference):
 contains_grader = FunctionGrader(contains_reference)
 ```
 
-- **LLM-Based graders**: For subjective, qualitative evaluations using the [LLMGrader](../../openjudge/graders/llm_grader.py), these graders leverage large language models to perform sophisticated evaluations. These graders can assess complex qualities like helpfulness, safety, or coherence by using natural language understanding. This approach is best suited for qualitative analysis where nuanced understanding is required, such as evaluating the helpfulness of responses, detecting subtle hallucinations, or assessing the quality of creative content.
+- **LLM-Based graders**: For subjective, qualitative evaluations using the [LLMGrader](https://github.com/modelscope/OpenJudge/blob/main/openjudge/graders/llm_grader.py), these graders leverage large language models to perform sophisticated evaluations. These graders can assess complex qualities like helpfulness, safety, or coherence by using natural language understanding. This approach is best suited for qualitative analysis where nuanced understanding is required, such as evaluating the helpfulness of responses, detecting subtle hallucinations, or assessing the quality of creative content.
 
 ```python
 # An LLM grader that evaluates helpfulness of responses
@@ -71,9 +71,9 @@ helpfulness_grader = HelpfulnessGrader(model=OpenAIChatModel("qwen3-32b"))
 
 Based on the evaluation purpose, you can choose the appropriate evaluation mode. Each mode produces specific output types:
 
-- **Pointwise evaluation**: Assesses individual samples independently, generating a score for each input-output pair. This approach is suitable for quantitative assessment of individual responses, producing a [GraderScore](../../openjudge/graders/schema.py) for each sample.
+- **Pointwise evaluation**: Assesses individual samples independently, generating a score for each input-output pair. This approach is suitable for quantitative assessment of individual responses, producing a [GraderScore](https://github.com/modelscope/OpenJudge/blob/main/openjudge/graders/schema.py) for each sample.
 
-- **Listwise evaluation**: Ranks multiple samples relative to each other, comparing several responses to the same query. This approach generates relative rankings rather than absolute scores, producing a [GraderRank](../../openjudge/graders/schema.py) that indicates the relative quality of responses.
+- **Listwise evaluation**: Ranks multiple samples relative to each other, comparing several responses to the same query. This approach generates relative rankings rather than absolute scores, producing a [GraderRank](https://github.com/modelscope/OpenJudge/blob/main/openjudge/graders/schema.py) that indicates the relative quality of responses.
 
 ### Understanding Results
 
@@ -121,7 +121,7 @@ For detailed instructions on creating custom graders, please refer to the [Build
 
 ## How Runners Use Graders
 
-The [GradingRunner](../../openjudge/runner/grading_runner.py) is the central execution engine of OpenJudge that orchestrates the evaluation process across multiple graders. It acts as the conductor of an orchestra, coordinating all the different graders to create a harmonious evaluation process. The GradingRunner is specifically designed to serve Graders by providing the infrastructure they need to operate efficiently and effectively. It serves Graders by providing execution orchestration, data mapping services, concurrency management, result aggregation, and resource optimization to execute graders concurrently and maximize throughput.
+The [GradingRunner](https://github.com/modelscope/OpenJudge/blob/main/openjudge/runner/grading_runner.py) is the central execution engine of OpenJudge that orchestrates the evaluation process across multiple graders. It acts as the conductor of an orchestra, coordinating all the different graders to create a harmonious evaluation process. The GradingRunner is specifically designed to serve Graders by providing the infrastructure they need to operate efficiently and effectively. It serves Graders by providing execution orchestration, data mapping services, concurrency management, result aggregation, and resource optimization to execute graders concurrently and maximize throughput.
 
 To better understand how Runners use Graders, let's look at a complete configuration example:
 
@@ -188,7 +188,7 @@ Types of mappers include:
 
 #### Aggregation Configuration
 
-After running multiple graders, you might want to combine their results into a single score. The [aggregator submodule](../../openjudge/runner/aggregator/) provides components that take multiple grader results and combine them into a unified result:
+After running multiple graders, you might want to combine their results into a single score. The [aggregator submodule](https://github.com/modelscope/OpenJudge/tree/main/openjudge/runner/aggregator/) provides components that take multiple grader results and combine them into a unified result:
 
 - **WeightedSumAggregator**: Combining results using weighted averages. In our example, we assign 60% weight to helpfulness and 40% to relevance: `WeightedSumAggregator(weights={"helpfulness": 0.6, "relevance": 0.4})`
 - **MaxAggregator**: Taking the maximum score among all graders
@@ -208,11 +208,11 @@ Concurrency control enables efficient processing of large datasets while maintai
 
 ## How to Analyze Grader
 
-After running evaluations with the **GradingRunner**, you can use the [analyzer module](../../openjudge/analyzer/) to process the results and gain deeper insights. Analyzers are optional components that help you understand your evaluation results better and assess the quality of your graders.
+After running evaluations with the **GradingRunner**, you can use the [analyzer module](https://github.com/modelscope/OpenJudge/tree/main/openjudge/analyzer/) to process the results and gain deeper insights. Analyzers are optional components that help you understand your evaluation results better and assess the quality of your graders.
 
 Types of analyzers include:
-- **Statistical analyzers**: Compute statistics on evaluation results (e.g., [DistributionAnalyzer](../../openjudge/analyzer/statistical/distribution_analyzer.py)) to understand score distributions and identify potential issues with grader consistency
-- **Validation analyzers**: Compare evaluation results with reference labels (e.g., [AccuracyAnalyzer](../../openjudge/analyzer/validation/accuracy_analyzer.py), [F1ScoreAnalyzer](../../openjudge/analyzer/validation/f1_score_analyzer.py)) to measure how well your graders correlate with known ground truth
+- **Statistical analyzers**: Compute statistics on evaluation results (e.g., [DistributionAnalyzer](https://github.com/modelscope/OpenJudge/blob/main/openjudge/analyzer/statistical/distribution_analyzer.py)) to understand score distributions and identify potential issues with grader consistency
+- **Validation analyzers**: Compare evaluation results with reference labels (e.g., [AccuracyAnalyzer](https://github.com/modelscope/OpenJudge/blob/main/openjudge/analyzer/validation/accuracy_analyzer.py), [F1ScoreAnalyzer](https://github.com/modelscope/OpenJudge/blob/main/openjudge/analyzer/validation/f1_score_analyzer.py)) to measure how well your graders correlate with known ground truth
 
 These analyzers help you:
 - Evaluate the effectiveness of your graders by comparing their outputs to known standards

docs/requirements.txt

@@ -0,0 +1,12 @@
+# MkDocs and theme
+mkdocs==1.6.1
+mkdocs-shadcn==0.9.5
+
+# Plugins
+mkdocstrings==0.30.1
+mkdocstrings-python==1.18.2
+
+# Markdown extensions (pymdownx is included in pymdown-extensions)
+pymdown-extensions==10.16.1
+
+