feat: add pyproject.toml for pip install and fix unit tests

docs_roll/docs/Getting Started/Debugging Guide/debug_guide.md

@@ -64,7 +64,7 @@ conda activate python310_torch260_em
 2. Install dependencies:
 ```bash
 pip3 install torch torchvision torchaudio py-cpuinfo
-pip install -r requirements_em_local_debug.txt
+pip install -e ".[local-debug]"
 ```
 
 3. Run the test script:

docs_roll/docs/Getting Started/Installation/installation.md

@@ -19,8 +19,10 @@ vLLM >= 0.7.3
 # Clone the repository and install
 git clone https://github.com/alibaba/ROLL.git
 cd ROLL
-pip install -r requirements.txt # Or follow your specific installation steps
-# For development, consider: pip install -e .
+# Install with specific PyTorch version and inference engine, e.g.:
+pip install -e ".[torch260-vllm]"  # PyTorch 2.6.0 + vLLM
+# Other available extras: torch260-sglang, torch280-vllm, torch280-sglang, torch260-diffsynth, local-debug, gem
+# For basic install: pip install -e .
 ```
 
 For AMD users, please ensure you meet the following prerequisites:
@@ -33,7 +35,7 @@ vLLM >= 0.8.4
 # Clone the repository and install
 git clone https://github.com/alibaba/ROLL.git
 cd ROLL
-pip install -r requirements.txt # Or follow your specific installation steps
+pip install -e ".[torch260-vllm]"  # Or choose other extras based on your environment
 ```
 We highly suggest to use pre-built Docker images from [Image Addresses](https://alibaba.github.io/ROLL/docs/QuickStart/image_address) instead of installation from Custom Environment for ROCm users.
 

docs_roll/docs/Getting Started/Quick Start/aliyun_serverless_devpod_quick_start.md

@@ -35,7 +35,7 @@ git clone https://github.com/alibaba/ROLL.git
 
 # 2. Install project dependencies
 cd ROLL
-pip install -r requirements_torch260_vllm.txt -i https://mirrors.aliyun.com/pypi/simple/  
+pip install -e ".[torch260-vllm]" -i https://mirrors.aliyun.com/pypi/simple/  
 ```
 
 ### Run a Pipeline Example

docs_roll/docs/Getting Started/Quick Start/multi_nodes_quick_start.md

@@ -30,9 +30,9 @@ nvidia-smi
 # 4. Clone the project code
 git clone https://github.com/alibaba/ROLL.git
 
-# 5. Install project dependencies (choose the requirements file corresponding to your image)
+# 5. Install project dependencies (choose the extras corresponding to your environment)
 cd ROLL
-pip install -r requirements_torch260_vllm.txt -i https://mirrors.aliyun.com/pypi/simple/
+pip install -e ".[torch260-vllm]" -i https://mirrors.aliyun.com/pypi/simple/
 ```
 
 ## Pipeline Execution

docs_roll/docs/Getting Started/Quick Start/single_node_quick_start.md

@@ -30,9 +30,9 @@ nvidia-smi
 # 4. Clone the project code
 git clone https://github.com/alibaba/ROLL.git
 
-# 5. Install project dependencies (choose the requirements file corresponding to your image)
+# 5. Install project dependencies (choose the extras corresponding to your environment)
 cd ROLL
-pip install -r requirements_torch260_vllm.txt -i https://mirrors.aliyun.com/pypi/simple/
+pip install -e ".[torch260-vllm]" -i https://mirrors.aliyun.com/pypi/simple/
 ```
 
 ## Pipeline Execution

docs_roll/docs/User Guides/Algorithms/Reward_FL.md

@@ -79,7 +79,7 @@ The following parameters related to Wan2_2 are as follows:
 
 ## Setup environments
 ```
-pip install -r requirements_torch260_diffsynth.txt
+pip install -e ".[torch260-diffsynth]"
 ```
 
 ## Reference Example

docs_roll/docs/User Guides/Hardware Support/ascend_usage.md

@@ -73,7 +73,7 @@ pip install vllm-ascend==0.13.0
 ```
 git clone https://github.com/alibaba/ROLL.git
 cd ROLL
-pip install -r requirements_common.txt
+pip install -e .
 pip install deepspeed==0.16.4
 cd ..
 ```

docs_roll/docs/User Guides/Pipeline/agent_pipeline_start.md

@@ -52,7 +52,7 @@ Before you begin, ensure you have the following:
 2. **Python Dependencies** – Install all necessary Python dependencies, typically via the requirements file:
 
    ```bash
-   pip install -r requirements.txt   # Or a specific file like requirements_torch260.txt
+   pip install -e .   # Or with specific extras like pip install -e ".[torch260-vllm]"
    ```
 
    Ensure any specific dependencies for your chosen agentic environments or models are also met.

docs_roll/docs/User Guides/Pipeline/agentic_pipeline_start.md

@@ -218,7 +218,7 @@ bash examples/qwen2.5-0.5B-agentic/run_agentic_pipeline_frozen_lake.sh
 * Ensure all necessary dependencies are installed, it's recommended to start from [image launch](../../Getting%20Started/Installation/installation.md):
 
   ```bash
-  pip install -r requirements.txt
+  pip install -e .
   ```
 
 * Confirm all model paths in the configuration are accessible.

docs_roll/docs/User Guides/Pipeline/distill_pipeline_start.md

@@ -205,7 +205,7 @@ bash examples/qwen2.5-7B-distill_megatron/run_distill_pipeline.sh
 * Ensure all necessary dependencies are installed:
 
   ```bash
-  pip install -r requirements.txt
+  pip install -e .
   ```
 
 * Verify that all model paths in the configuration are accessible.

docs_roll/docs/User Guides/Pipeline/dpo_pipeline_start.md

@@ -200,7 +200,7 @@ bash bash examples/qwen2.5-3B-dpo_megatron/run_dpo_pipeline.sh
 * Ensure all necessary dependencies are installed:
 
   ```bash
-  pip install -r requirements.txt
+  pip install -e .
   ```
 
 * Verify that all model paths in the configuration are accessible.

docs_roll/docs/User Guides/Pipeline/on_policy_distill_pipeline_start.md

@@ -349,7 +349,7 @@ bash examples/qwen3-8B-onpolicy-distill-megatron/run_onpolicy_distill_pipeline.s
 * Ensure all necessary dependencies are installed:
 
   ```bash
-  pip install -r requirements.txt
+  pip install -e .
   ```
 
 * Verify that all model paths in the configuration are accessible.

docs_roll/docs/User Guides/Pipeline/rlvr_pipeline_start.md

@@ -215,11 +215,11 @@ Example data format (Code domain):
   "source": "codeforeces",
   "difficulty": "0",
   "prompt": "You are an expert Python programmer. You will be given a question (problem specification) and will generate a correct Python program that matches the specification and passes all tests. \n\n### Question: Write a function that takes an array of distinct integers and returns all possible permutations (in any order). Each permutation should be represented as an array of integers. The function should handle arrays of different lengths efficiently.\n\n### Format: You will use the following starter code to write the solution to the problem and enclose your code within delimiters.\n```python\ndef permute(nums):\n```\n\n### Answer: (use the provided format with backticks)",
-  "messages": "[{\"role\": \"user\", \"content\": \"You are an expert Python programmer. You will be given a question (problem specification) and will generate a correct Python program that matches the specification and passes all tests. \\n\\n### Question: Write a function that takes an array of distinct integers and returns all possible permutations (in any order). Each permutation should be represented as an array of integers. The function should handle arrays of different lengths efficiently.\\n\\n### Format: You will use the following starter code to write the solution to the problem and enclose your code within delimiters.\\n```python\\ndef permute(nums):\\n```\\n\\n### Answer: (use the provided format with backticks)\"}]",
-  "ground_truth": "[\"def permute(nums):\\n    \\\"\\\"\\\"\\n    Given an array of distinct integers, return all possible permutations.\\n    Each permutation is an array of integers.\\n    \\\"\\\"\\\"\\n    def backtrack(start, end):\\n        if start == end:\\n            permutations.append(nums[:])\\n        for i in range(start, end):\\n            nums[start], nums[i] = nums[i], nums[start]\\n            backtrack(start + 1, end)\\n            nums[start], nums[i] = nums[i], nums[start]\\n\\n    permutations = []\\n    backtrack(0, len(nums))\\n    return permutations\"]",
+  "messages": "[{\"role\": \"user\", \"content\": \"You are an expert Python programmer. You will be given a question (problem specification) and will generate a correct Python program that matches the specification and passes all tests. \n\n### Question: Write a function that takes an array of distinct integers and returns all possible permutations (in any order). Each permutation should be represented as an array of integers. The function should handle arrays of different lengths efficiently.\n\n### Format: You will use the following starter code to write the solution to the problem and enclose your code within delimiters.\n```python\ndef permute(nums):\n```\n\n### Answer: (use the provided format with backticks)\"}]",
+  "ground_truth": "[\"def permute(nums):\n    \\\"\\\"\\\"\n    Given an array of distinct integers, return all possible permutations.\n    Each permutation is an array of integers.\n    \\\"\\\"\\\"\n    def backtrack(start, end):\n        if start == end:\n            permutations.append(nums[:])\n        for i in range(start, end):\n            nums[start], nums[i] = nums[i], nums[start]\n            backtrack(start + 1, end)\n            nums[start], nums[i] = nums[i], nums[start]\n\n    permutations = []\n    backtrack(0, len(nums))\n    return permutations\"]",
   "case_type": "pytest",
   "test_case_function": " ",
-  "test_cases": "[{\"assert_code\": \"\\n\\n\\ndef test_permute_single_element():\\n    assert permute([1]) == [[1]]\\n\\ndef test_permute_two_elements():\\n    result = permute([1, 2])\\n    expected = [[1, 2], [2, 1]]\\n    assert sorted(result) == sorted(expected)\\n\\ndef test_permute_three_elements():\\n    result = permute([1, 2, 3])\\n    expected = [[1, 2, 3], [1, 3, 2], [2, 1, 3], [2, 3, 1], [3, 1, 2], [3, 2, 1]]\\n    assert sorted(result) == sorted(expected)\\n\\ndef test_permute_four_elements():\\n    result = permute([1, 2, 3, 4])\\n    expected = [\\n        [1, 2, 3, 4], [1, 2, 4, 3], [1, 3, 2, 4], [1, 3, 4, 2], [1, 4, 2, 3], [1, 4, 3, 2],\\n        [2, 1, 3, 4], [2, 1, 4, 3], [2, 3, 1, 4], [2, 3, 4, 1], [2, 4, 1, 3], [2, 4, 3, 1],\\n        [3, 1, 2, 4], [3, 1, 4, 2], [3, 2, 1, 4], [3, 2, 4, 1], [3, 4, 1, 2], [3, 4, 2, 1],\\n        [4, 1, 2, 3], [4, 1, 3, 2], [4, 2, 1, 3], [4, 2, 3, 1], [4, 3, 1, 2], [4, 3, 2, 1]\\n    ]\\n    assert sorted(result) == sorted(expected)\"}]",
+  "test_cases": "[{\"assert_code\": \"\n\n\ndef test_permute_single_element():\n    assert permute([1]) == [[1]]\n\ndef test_permute_two_elements():\n    result = permute([1, 2])\n    expected = [[1, 2], [2, 1]]\n    assert sorted(result) == sorted(expected)\n\ndef test_permute_three_elements():\n    result = permute([1, 2, 3])\n    expected = [[1, 2, 3], [1, 3, 2], [2, 1, 3], [2, 3, 1], [3, 1, 2], [3, 2, 1]]\n    assert sorted(result) == sorted(expected)\n\ndef test_permute_four_elements():\n    result = permute([1, 2, 3, 4])\n    expected = [\n        [1, 2, 3, 4], [1, 2, 4, 3], [1, 3, 2, 4], [1, 3, 4, 2], [1, 4, 2, 3], [1, 4, 3, 2],\n        [2, 1, 3, 4], [2, 1, 4, 3], [2, 3, 1, 4], [2, 3, 4, 1], [2, 4, 1, 3], [2, 4, 3, 1],\n        [3, 1, 2, 4], [3, 1, 4, 2], [3, 2, 1, 4], [3, 2, 4, 1], [3, 4, 1, 2], [3, 4, 2, 1],\n        [4, 1, 2, 3], [4, 1, 3, 2], [4, 2, 1, 3], [4, 2, 3, 1], [4, 3, 1, 2], [4, 3, 2, 1]\n    ]\n    assert sorted(result) == sorted(expected)\"}]",
   "tag": "KodCode"
 }
 ```
@@ -314,7 +314,7 @@ bash examples/qwen2.5-7B-rlvr_megatron/run_rlvr_pipeline.sh
 * Ensure all necessary dependencies are installed:
 
   ```bash
-  pip install -r requirements.txt
+  pip install -e .
   ```
 
 * Verify that all model paths in the configuration are accessible.

docs_roll/docs/User Guides/Pipeline/sft_pipeline_start.md

@@ -228,7 +228,7 @@ Key items to check:
 ### Step 2: Prepare Environment and Dependencies
 
 ```bash
-pip install -r requirements.txt
+pip install -e .
 ```
 
 Also ensure:

docs_roll/docs/User Guides/Pipeline/vl_rlvr_pipeline_start.md

@@ -300,7 +300,7 @@ bash examples/qwen2.5-vl-7B-rlvr/run_rlvr_pipeline.sh
 * Ensure all necessary dependencies are installed. NOTE: VLLM is the only supported inference engine for VLM pipeline currently, thus use the corresponding requirement files:
 
   ```bash
-  pip install -r requirements_torch260_vllm.txt
+  pip install -e ".[torch260-vllm]"
   ```
 
 * Verify that all model paths in the configuration are accessible.

docs_roll/i18n/zh-Hans/docusaurus-plugin-content-docs/current/Getting Started/Debugging Guide/debug_guide.md

@@ -62,7 +62,7 @@ conda activate python310_torch260_em
 2. 安装依赖：
 ```bash
 pip3 install torch torchvision torchaudio py-cpuinfo
-pip install -r requirements_em_local_debug.txt
+pip install -e ".[local-debug]"
 ```
 
 3. 运行测试脚本：

docs_roll/i18n/zh-Hans/docusaurus-plugin-content-docs/current/Getting Started/Installation/installation.md

@@ -19,6 +19,8 @@ vLLM >= 0.7.3
 # 克隆仓库并安装
 git clone https://github.com/alibaba/ROLL.git
 cd ROLL
-pip install -r requirements.txt # 或按照您的特定安装步骤
-# 对于开发，可以考虑使用：pip install -e .
+# 安装指定 PyTorch 版本和推理引擎，例如：
+pip install -e ".[torch260-vllm]"  # PyTorch 2.6.0 + vLLM
+# 其他可选项：torch260-sglang, torch280-vllm, torch280-sglang, torch260-diffsynth, local-debug, gem
+# 基础安装：pip install -e .
 ```

docs_roll/i18n/zh-Hans/docusaurus-plugin-content-docs/current/Getting Started/Quick Start/aliyun_serverless_devpod_quick_start.md

@@ -35,7 +35,7 @@ git clone https://github.com/alibaba/ROLL.git
 
 # 2. 安装项目依赖
 cd ROLL
-pip install -r requirements_torch260_vllm.txt -i https://mirrors.aliyun.com/pypi/simple/
+pip install -e ".[torch260-vllm]" -i https://mirrors.aliyun.com/pypi/simple/
 ```
 
 ### 运行 pipeline 示例

docs_roll/i18n/zh-Hans/docusaurus-plugin-content-docs/current/Getting Started/Quick Start/multi_nodes_quick_start.md

@@ -30,9 +30,9 @@ nvidia-smi
 # 4. 克隆项目代码
 git clone https://github.com/alibaba/ROLL.git
 
-# 5. 安装项目依赖（选择对应镜像的requirements文件）
+# 5. 安装项目依赖（选择对应环境的可选项）
 cd ROLL
-pip install -r requirements_torch260_vllm.txt -i https://mirrors.aliyun.com/pypi/simple/
+pip install -e ".[torch260-vllm]" -i https://mirrors.aliyun.com/pypi/simple/
 ```
 
 ## pipeline运行

docs_roll/i18n/zh-Hans/docusaurus-plugin-content-docs/current/Getting Started/Quick Start/single_node_quick_start.md

@@ -30,9 +30,9 @@ nvidia-smi
 # 4. 克隆项目代码
 git clone https://github.com/alibaba/ROLL.git
 
-# 5. 安装项目依赖（选择对应镜像的requirements文件）
+# 5. 安装项目依赖（选择对应环境的可选项）
 cd ROLL
-pip install -r requirements_torch260_vllm.txt -i https://mirrors.aliyun.com/pypi/simple/
+pip install -e ".[torch260-vllm]" -i https://mirrors.aliyun.com/pypi/simple/
 ```
 
 ## pipeline运行

docs_roll/i18n/zh-Hans/docusaurus-plugin-content-docs/current/User Guides/Algorithms/Reward_FL.md

@@ -77,7 +77,7 @@ Wan2_2 相关参数如下：
 
 ## 环境配置
 ```
-pip install -r requirements_torch260_diffsynth.txt
+pip install -e ".[torch260-diffsynth]"
 ```
 
 ## 参考示例

docs_roll/i18n/zh-Hans/docusaurus-plugin-content-docs/current/User Guides/Hardware Support/ascend_usage.md

@@ -78,7 +78,7 @@ pip install vllm-ascend==0.11.0rc1
 ```
 git clone https://github.com/alibaba/ROLL.git
 cd ROLL
-pip install -r requirements_common.txt
+pip install -e .
 pip install deepspeed==0.16.4
 cd ..
 ```

docs_roll/i18n/zh-Hans/docusaurus-plugin-content-docs/current/User Guides/Pipeline/agentic_pipeline_start.md

@@ -214,7 +214,7 @@ bash examples/qwen2.5-0.5B-agentic/run_agentic_pipeline_frozen_lake.sh
 * 确保已安装所有必要依赖，建议从[镜像启动](../../Getting%20Started/Installation/installation.md)：
 
   ```bash
-  pip install -r requirements.txt
+  pip install -e .
   ```
 
 * 确认配置中所有模型路径均可访问。

docs_roll/i18n/zh-Hans/docusaurus-plugin-content-docs/current/User Guides/Pipeline/distill_pipeline_start.md

@@ -191,7 +191,7 @@ bash examples/qwen2.5-7B-distill_megatron/run_distill_pipeline.sh
 * 确保已安装所有必要依赖：
 
   ```bash
-  pip install -r requirements.txt
+  pip install -e .
   ```
 
 * 确认配置中所有模型路径均可访问。

docs_roll/i18n/zh-Hans/docusaurus-plugin-content-docs/current/User Guides/Pipeline/dpo_pipeline_start.md

@@ -186,7 +186,7 @@ bash examples/qwen2.5-3B-dpo_megatron/run_dpo_pipeline.sh
 * 确保已安装所有必要依赖：
 
   ```bash
-  pip install -r requirements.txt
+  pip install -e .
   ```
 
 * 确认配置中所有模型路径均可访问。

docs_roll/i18n/zh-Hans/docusaurus-plugin-content-docs/current/User Guides/Pipeline/on_policy_distill_pipeline_start.md

@@ -358,7 +358,7 @@ bash examples/qwen3-8B-onpolicy-distill-megatron/run_onpolicy_distill_pipeline.s
 * 确保安装了所有必要的依赖：
 
   ```bash
-  pip install -r requirements.txt
+  pip install -e .
   ```
 
 * 验证配置中的所有模型路径是否可访问。

docs_roll/i18n/zh-Hans/docusaurus-plugin-content-docs/current/User Guides/Pipeline/rlvr_pipeline_start.md

@@ -214,11 +214,11 @@ rlvr流水线使用JSON格式的数据文件。不同领域需要特定字段：
   "source": "codeforeces",
   "difficulty": "0",
   "prompt": "你是一位专业的Python程序员。你将收到一个问题（问题描述）并生成一个正确的Python程序，该程序符合描述并能通过所有测试。\\n\\n### 问题：编写一个函数，接收一个不同整数的数组并返回所有可能的排列（任意顺序）。每个排列应表示为一个整数数组。该函数应能高效处理不同长度的数组。\\n\\n### 格式：你将使用以下起始代码编写问题的解决方案，并将代码包含在分隔符内。\\n```python\\ndef permute(nums):\\n```\\n\\n### 答案：（使用提供的格式和反引号）",
-  "messages": "[{\"role\": \"user\", \"content\": \"你是一位专业的Python程序员。你将收到一个问题（问题描述）并生成一个正确的Python程序，该程序符合描述并能通过所有测试。 \\n\\n### 问题：编写一个函数，接收一个不同整数的数组并返回所有可能的排列（任意顺序）。每个排列应表示为一个整数数组。该函数应能高效处理不同长度的数组。\\n\\n### 格式：你将使用以下起始代码编写问题的解决方案，并将代码包含在分隔符内。\\n```python\\ndef permute(nums):\\n```\\n\\n### 答案：（使用提供的格式和反引号）\"}]",
-  "ground_truth": "[\"def permute(nums):\\n    \\\"\\\"\\\"\\n    给定一个不同整数的数组，返回所有可能的排列。\\n    每个排列是一个整数数组。\\n    \\\"\\\"\\\"\\n    def backtrack(start, end):\\n        if start == end:\\n            permutations.append(nums[:])\\n        for i in range(start, end):\\n            nums[start], nums[i] = nums[i], nums[start]\\n            backtrack(start + 1, end)\\n            nums[start], nums[i] = nums[i], nums[start]\\n\\n    permutations = []\\n    backtrack(0, len(nums))\\n    return permutations\"]",
+  "messages": "[{\"role\": \"user\", \"content\": \"你是一位专业的Python程序员。你将收到一个问题（问题描述）并生成一个正确的Python程序，该程序符合描述并能通过所有测试。 \n\n### 问题：编写一个函数，接收一个不同整数的数组并返回所有可能的排列（任意顺序）。每个排列应表示为一个整数数组。该函数应能高效处理不同长度的数组。\n\n### 格式：你将使用以下起始代码编写问题的解决方案，并将代码包含在分隔符内。\n```python\ndef permute(nums):\n```\n\n### 答案：（使用提供的格式和反引号）\"}]",
+  "ground_truth": "[\"def permute(nums):\n    \\\"\\\"\\\"\n    给定一个不同整数的数组，返回所有可能的排列。\n    每个排列是一个整数数组。\n    \\\"\\\"\\\"\n    def backtrack(start, end):\n        if start == end:\n            permutations.append(nums[:])\n        for i in range(start, end):\n            nums[start], nums[i] = nums[i], nums[start]\n            backtrack(start + 1, end)\n            nums[start], nums[i] = nums[i], nums[start]\n\n    permutations = []\n    backtrack(0, len(nums))\n    return permutations\"]",
   "case_type": "pytest",
   "test_case_function": " ",
-  "test_cases": "[{\"assert_code\": \"\\n\\n\\ndef test_permute_single_element():\\n    assert permute([1]) == [[1]]\\n\\ndef test_permute_two_elements():\\n    result = permute([1, 2])\\n    expected = [[1, 2], [2, 1]]\\n    assert sorted(result) == sorted(expected)\\n\\ndef test_permute_three_elements():\\n    result = permute([1, 2, 3])\\n    expected = [[1, 2, 3], [1, 3, 2], [2, 1, 3], [2, 3, 1], [3, 1, 2], [3, 2, 1]]\\n    assert sorted(result) == sorted(expected)\\n\\ndef test_permute_four_elements():\\n    result = permute([1, 2, 3, 4])\\n    expected = [\\n        [1, 2, 3, 4], [1, 2, 4, 3], [1, 3, 2, 4], [1, 3, 4, 2], [1, 4, 2, 3], [1, 4, 3, 2],\\n        [2, 1, 3, 4], [2, 1, 4, 3], [2, 3, 1, 4], [2, 3, 4, 1], [2, 4, 1, 3], [2, 4, 3, 1],\\n        [3, 1, 2, 4], [3, 1, 4, 2], [3, 2, 1, 4], [3, 2, 4, 1], [3, 4, 1, 2], [3, 4, 2, 1],\\n        [4, 1, 2, 3], [4, 1, 3, 2], [4, 2, 1, 3], [4, 2, 3, 1], [4, 3, 1, 2], [4, 3, 2, 1]\\n    ]\\n    assert sorted(result) == sorted(expected)\"}]",
+  "test_cases": "[{\"assert_code\": \"\n\n\ndef test_permute_single_element():\n    assert permute([1]) == [[1]]\n\ndef test_permute_two_elements():\n    result = permute([1, 2])\n    expected = [[1, 2], [2, 1]]\n    assert sorted(result) == sorted(expected)\n\ndef test_permute_three_elements():\n    result = permute([1, 2, 3])\n    expected = [[1, 2, 3], [1, 3, 2], [2, 1, 3], [2, 3, 1], [3, 1, 2], [3, 2, 1]]\n    assert sorted(result) == sorted(expected)\n\ndef test_permute_four_elements():\n    result = permute([1, 2, 3, 4])\n    expected = [\n        [1, 2, 3, 4], [1, 2, 4, 3], [1, 3, 2, 4], [1, 3, 4, 2], [1, 4, 2, 3], [1, 4, 3, 2],\n        [2, 1, 3, 4], [2, 1, 4, 3], [2, 3, 1, 4], [2, 3, 4, 1], [2, 4, 1, 3], [2, 4, 3, 1],\n        [3, 1, 2, 4], [3, 1, 4, 2], [3, 2, 1, 4], [3, 2, 4, 1], [3, 4, 1, 2], [3, 4, 2, 1],\n        [4, 1, 2, 3], [4, 1, 3, 2], [4, 2, 1, 3], [4, 2, 3, 1], [4, 3, 1, 2], [4, 3, 2, 1]\n    ]\n    assert sorted(result) == sorted(expected)\"}]",
   "tag": "KodCode"
 }
 ```
@@ -313,7 +313,7 @@ bash examples/qwen2.5-7B-rlvr_megatron/run_rlvr_pipeline.sh
 * 确保安装了所有必要的依赖：
 
   ```bash
-  pip install -r requirements.txt
+  pip install -e .
   ```
 
 * 验证配置中的所有模型路径是否可访问。

docs_roll/i18n/zh-Hans/docusaurus-plugin-content-docs/current/User Guides/Pipeline/sft_pipeline_start.md

@@ -229,7 +229,7 @@ bash examples/qwen2.5-7B-sft_megatron/run_sft_pipeline.sh
 ### 步骤2：准备环境和依赖
 
 ```bash
-pip install -r requirements.txt
+pip install -e .
 ```
 
 并确保：