update README

README-cn.md

@@ -147,7 +147,6 @@ datasets:
 
 📂 **开放高质量数据**：与模型一同开源其背后的高质量训练数据，均属于 [UltraData](https://ultradata.openbmb.cn/) 数据体系：[UltraX](https://huggingface.co/datasets/openbmb/UltraX-Preview)，高质量网页预训练数据集；[UltraData-Code](https://huggingface.co/datasets/openbmb/UltraData-Code)，L0–L3 分级代码治理，推动代码能力显著跃升；[UltraData-SFT-Agent-2609](https://huggingface.co/datasets/openbmb/UltraData-SFT-Agent-2609)，50 万 Agent 训练样本，赋能端侧 Agent 综合能力提升；[UltraData-RL-2609](https://huggingface.co/datasets/openbmb/UltraData-RL-2609)，超 8 万条高质量 RL 训练样本，覆盖数学、代码、通用知识与长文本推理。
 
-
 ## 模型列表
 
 你可以按运行环境选择对应模型格式：
@@ -183,8 +182,6 @@ MiniCPM5-2B 具有以下特性：
 - **注意力头（GQA）**：16 个 Q heads / 2 个 KV heads
 - **上下文长度**：131,072
 
-
-
 ## 简介
 
 MiniCPM5-2B 是 MiniCPM5 系列的第二个模型，面向本地助手、coding agent、工具调用流程以及需要紧凑模型的推理场景。它在较小部署成本下提供原生长上下文能力。
@@ -195,7 +192,6 @@ MiniCPM5-2B 是 MiniCPM5 系列的第二个模型，面向本地助手、coding
 
 在这组对比中，MiniCPM5-2B 达到同尺寸开源模型 SOTA 水平（平均分 53.9 ），也超过了参与对比的全部更大规模模型（最高 51.1）。其优势主要体现在代码推理、数学推理、长文本、工具调用与多个智能体任务上。
 
-
 <div style="width:100%;max-width:1080px;margin:0 auto;padding:16px 0;background:#fff;
 font-family:-apple-system,BlinkMacSystemFont,'Segoe UI',Roboto,'PingFang SC',
 'Hiragino Sans GB','Microsoft YaHei',sans-serif;color:#171717">
@@ -258,26 +254,18 @@ MiniCPM5-2B 的训练过程是 **[UltraData 分级数据管理体系](https://ar
 
 **后训练阶段**分为 **SFT**、**RL** 与 **OPD** 三步。我们先使用 **400B tokens deep-thinking SFT** 建立深度思考和通用对话能力，相关 SFT 数据已同步开源为 [UltraData-SFT-2605](https://huggingface.co/datasets/openbmb/UltraData-SFT-2605)与[UltraData-SFT-Agent-2609](https://huggingface.co/datasets/openbmb/UltraData-SFT-Agent-2609)。随后针对数学、代码、Agent 和写作等方向训练专用 **RL teacher**（相关数据已同步开源为[UltraData-RL-2609](https://huggingface.co/datasets/openbmb/UltraData-RL-2609)），并通过 **On-Policy Distillation (OPD)** 将这些 teacher 的能力蒸馏回同一个发布模型。
 
-![MiniCPM5-2B 训练流程](https://raw.githubusercontent.com/OpenBMB/MiniCPM/minicpm5-2b/assets/minicpm5/minicpm5_2b_training_recipe.jpg)
-
 ![MiniCPM5-2B 训练流程](https://raw.githubusercontent.com/OpenBMB/MiniCPM/main/assets/minicpm5/minicpm5_2b_training_recipe.jpg)
 
 ### RL + OPD 带来了什么？
 
 **RL + OPD** 是 MiniCPM5-2B 后训练中的关键环节。**RL** 阶段，使用了 [JustRL II](https://app.notion.com/p/panhaoxuan/JustRL-II-Scaling-Small-LLMs-to-128K-Reasoning-with-a-Critic-3c77e972297c80adb8b5f4b05d267012#5f3eb56b29f048ebab5f71138f12e36f) 阐述的 critic-based 算法，大幅提升训练稳定性，并在多个领域取得了显著的收益。在下面列出的基准中，RL + OPD 在推理与通用能力上平均提升 **↑ 10.96 分**，Agent 能力平均提升 **↑ 6.96 分**。
 
-
 **OPD** 阶段对 16 个 RL 训练所得到的专家模型（含 5 个 agentic 专家模型）实现了能力合并。训练方式上，我们在 response 序列的每个位置分别对学生模型和教师模型 logits 计算全词表的反向 KL 散度作为优势估计值，替代原有的 verification-based advantage；训练数据上，我们的 OPD 直接复用各 RL teacher 训练时 prompt 作为蒸馏数据，无需额外构造语料。
 
-![MiniCPM5-2B RL + OPD 增益](https://raw.githubusercontent.com/OpenBMB/MiniCPM/minicpm5-2b/assets/minicpm5/minicpm5_2b_rl_opd_score_gains.png)
-
 ![MiniCPM5-2B RL + OPD 增益](https://raw.githubusercontent.com/OpenBMB/MiniCPM/main/assets/minicpm5/minicpm5_2b_rl_opd_score_gains.png)
 
-
 ## 快速上手
 
-
-
 ### vLLM
 
 ```bash
@@ -296,8 +284,6 @@ curl http://localhost:8000/v1/chat/completions \
   }'
 ```
 
-
-
 ### SGLang
 
 ```bash
@@ -328,8 +314,6 @@ python -m sglang.launch_server \
   --port 30000
 ```
 
-
-
 ### Transformers
 
 ```bash
@@ -369,15 +353,12 @@ python -m sglang.launch_server --model-path openbmb/MiniCPM5-2B --port 30000 \
     --tool-call-parser minicpm5      # 或：--tool-call-parser auto
 ```
 
-
-
 ## GitHub Cookbooks 与 Agent Skills
 
 MiniCPM5-2B 使用**标准** `LlamaForCausalLM` **架构**，主流推理引擎可直接加载，**无需自定义算子，也无模型代码 fork**。逐步部署和微调说明请参考下方 GitHub cookbooks；Agent Skills 作为 GitHub 资源提供给使用 Cursor / Claude Code 类 coding agent 的用户。
 
 ### 部署
 
-
 | 后端           | 模型格式 / 适用场景                                 | Cookbook                                                                                        | Agent Skill                                                                                                               |
 | ------------ | ------------------------------------------- | ----------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------- |
 | Transformers | BF16 / FP16，本地 Python 推理，GPU + CPU          | [transformers.md](https://github.com/OpenBMB/MiniCPM/blob/main/docs/deployment/transformers.md) | [minicpm5-deploy-transformers](https://github.com/OpenBMB/MiniCPM/blob/main/skills/minicpm5-deploy-transformers/SKILL.md) |
@@ -390,12 +371,8 @@ MiniCPM5-2B 使用**标准** `LlamaForCausalLM` **架构**，主流推理引擎
 | ArcLight     | GGUF 本地端侧 / CPU / 桌面 / 服务器                  | [arclight.md](https://github.com/OpenBMB/MiniCPM/blob/main/docs/deployment/arclight.md)         | [minicpm5-deploy-arclight](https://github.com/OpenBMB/MiniCPM/blob/main/skills/minicpm5-deploy-arclight/SKILL.md)         |
 | vLLM Ascend         | BF16 / FP16 OpenAI server                   | [vllm_ascend.md](https://github.com/OpenBMB/MiniCPM/blob/main/docs/deployment/vllm_ascend.md)                 | [minicpm5-deploy-vllm-ascend](https://github.com/OpenBMB/MiniCPM/blob/main/skills/minicpm5-deploy-vllm-ascend/SKILL.md)   |
 
-
-
-
 ### 微调
 
-
 | 框架            | 适用场景          | Cookbook                                                                                      | Agent Skill                                                                                                                   |
 | ------------- | ------------- | --------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------- |
 | TRL + PEFT    | LoRA / SFT 微调 | [trl.md](https://github.com/OpenBMB/MiniCPM/blob/main/docs/finetune/trl.md)                   | [minicpm5-finetune-trl](https://github.com/OpenBMB/MiniCPM/blob/main/skills/minicpm5-finetune-trl/SKILL.md)                   |
@@ -403,9 +380,6 @@ MiniCPM5-2B 使用**标准** `LlamaForCausalLM` **架构**，主流推理引擎
 | ms-swift      | 微调            | [ms_swift.md](https://github.com/OpenBMB/MiniCPM/blob/main/docs/finetune/ms_swift.md)         | [minicpm5-finetune-ms-swift](https://github.com/OpenBMB/MiniCPM/blob/main/skills/minicpm5-finetune-ms-swift/SKILL.md)         |
 | unsloth       | 微调            | [unsloth.md](https://github.com/OpenBMB/MiniCPM/blob/main/docs/finetune/unsloth.md)           | [minicpm5-finetune-unsloth](https://github.com/OpenBMB/MiniCPM/blob/main/skills/minicpm5-finetune-unsloth/SKILL.md)           |
 
-
-
-
 ### 其他支持的框架
 
 除上文列出的部署与微调框架外，MiniCPM5-2B 也支持通过 FlagOS 进行多芯片部署。
@@ -418,7 +392,6 @@ FlagOS 社区致力于打造面向多种 AI 芯片的统一、开源的系统软
 
 官网速递：[https://flagos.io](https://flagos.io/)
 
-
 <details>
 <summary>FlagOS 多 AI 芯片支持与使用方式</summary>
 
@@ -426,7 +399,6 @@ FlagOS 社区致力于打造面向多种 AI 芯片的统一、开源的系统软
 
 基于 FlagOS 极短时间内适配 MiniCPM5-2B 到 9 种不同的 AI 芯片，得益于众智 FlagOS 的多芯片统一 AI 系统软件栈的能力。目前，在 FlagOS 团队构建的面向多架构人工智能芯片的大模型自动迁移、适配与发布平台 FlagRelease 上，已发布 MiniCPM5-2B 的多芯片版本。细节如下：
 
-
 | Vendor    | ModelScope                                                                                                    | Huggingface                                                                                         |
 | --------- | ------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------- |
 | Nvidia    | [MiniCPM5-2B-nvidia-FlagOS](https://www.modelscope.cn/models/FlagRelease/MiniCPM5-2B-nvidia-FlagOS)       | [MiniCPM5-2B-nvidia-FlagOS](https://huggingface.co/FlagRelease/MiniCPM5-2B-nvidia-FlagOS)       |
@@ -439,28 +411,18 @@ FlagOS 社区致力于打造面向多种 AI 芯片的统一、开源的系统软
 | Ascend    | [MiniCPM5-2B-ascend-FlagOS](https://modelscope.cn/models/FlagRelease/MiniCPM5-2B-ascend-FlagOS)           | [MiniCPM5-2B-ascend-FlagOS](https://huggingface.co/FlagRelease/MiniCPM5-2B-ascend-FlagOS)       |
 | ARM-v9    | [MiniCPM5-2B-Armv9-FlagOS](https://modelscope.cn/models/FlagRelease/MiniCPM5-2B-Armv9-FlagOS)             | [MiniCPM5-2B-Armv9-FlagOS](https://huggingface.co/FlagRelease/MiniCPM5-2B-Armv9-FlagOS)         |
 
-
-
-
 #### FlagOS 使用方式
 
-
-
 ##### 使用 FlagOS 在 Nvidia 体验性能加速
 
-
-
 ###### From FlagRelease（**推荐**）
 
 FlagRelease是FlagOS团队构建的一套面向多架构人工智能芯片的大模型自动迁移、适配与发布平台，已发布MiniCPM5-2B的多芯片版本。FlagRelease 已内置相关软件包，无需用户安装。
 
 ###### FlagRelease 镜像关键版本信息
 
-
-
 ###### FlagRelease 使用速递
 
-
 | Vendor    | ModelScope                                                                                                    | Huggingface                                                                                         |
 | --------- | ------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------- |
 | Nvidia    | [MiniCPM5-2B-nvidia-FlagOS](https://www.modelscope.cn/models/FlagRelease/MiniCPM5-2B-nvidia-FlagOS)       | [MiniCPM5-2B-nvidia-FlagOS](https://huggingface.co/FlagRelease/MiniCPM5-2B-nvidia-FlagOS)       |
@@ -473,19 +435,12 @@ FlagRelease是FlagOS团队构建的一套面向多架构人工智能芯片的大
 | Ascend    | [MiniCPM5-2B-ascend-FlagOS](https://modelscope.cn/models/FlagRelease/MiniCPM5-2B-ascend-FlagOS)           | [MiniCPM5-2B-ascend-FlagOS](https://huggingface.co/FlagRelease/MiniCPM5-2B-ascend-FlagOS)       |
 | ARM-v9    | [MiniCPM5-2B-Armv9-FlagOS](https://modelscope.cn/models/FlagRelease/MiniCPM5-2B-Armv9-FlagOS)             | [MiniCPM5-2B-Armv9-FlagOS](https://huggingface.co/FlagRelease/MiniCPM5-2B-Armv9-FlagOS)         |
 
-
-
-
 ###### 从零开始
 
 - 依赖Python3.12, GLIBC_2.39, GLIBCXX_3.4.33, CXXABI_1.3.15 环境
 
-
-
 ###### Vllm 版本
 
-
-
 ###### 安装 FlagOS 算子库
 
 官方仓库：[https://github.com/flagos-ai/FlagGems](https://github.com/flagos-ai/FlagGems)
@@ -495,8 +450,6 @@ pip install flag-gems==4.2.1rc0
 pip install triton==3.5.1
 ```
 
-
-
 ###### 开启加速
 
 通过在vllm执行推理的源码中增加flagGems的导入即可开启flagGems加速
@@ -516,15 +469,12 @@ vllm serve ${model_path} \
 --gpu-memory-utilization 0.85
 ```
 
-
-
 ##### 使用 FlagOS 统一多芯片后端插件
 
 **[vllm-plugin-FL](https://github.com/flagos-ai/vllm-plugin-FL)** 是一个为 **vLLM** 推理/服务框架构建的插件，它基于 **FlagOS 的统一多芯片后端**开发，旨在扩展 vLLM 在多种硬件环境下的功能和性能表现。
 
 ###### vllm-plugin-FL 使用
 
-
 | 厂商  | 从零开始                                                                                                             | 从 FlagRelease 开始                                                                                     |                                                                                               |
 | --- | ---------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------- |
 | 英伟达 | [vllm-plugin-FL/MiniCPM5-2B](https://github.com/flagos-ai/vllm-plugin-FL/blob/main/examples/minicpm/README.md) | [MiniCPM5-2B-ModelScope](https://www.modelscope.cn/models/FlagRelease/MiniCPM5-2B-nvidia-FlagOS) | [MiniCPM5-2B-nvidia-FlagOS](https://huggingface.co/FlagRelease/MiniCPM5-2B-nvidia-FlagOS) |

README.md

@@ -155,8 +155,7 @@ We are releasing **MiniCPM5-2B**, the second model in the **MiniCPM5** series, f
   </svg>
 </div>
 
-📂 **Open High-Quality Data**: Alongside the model, we are releasing the high-quality training datasets behind it as part of the [UltraData](https://ultradata.openbmb.cn/) family: [UltraX](https://huggingface.co/datasets/openbmb/UltraX-Preview), a high-quality web pre-training dataset; [UltraData-Code](https://huggingface.co/datasets/openbmb/UltraData-Code), featuring L0–L3 tiered code data governance to drive a significant leap in coding capabilities; [UltraData-SFT-Agent-2609](https://huggingface.co/datasets/openbmb/UltraData-SFT-Agent-2609), comprising 500K agent training samples to enhance comprehensive on-device agent capabilities; and [UltraData-RL-2609](https://huggingface.co/datasets/openbmb/UltraData-RL-2609), with 80K+ high-quality RL training samples covering mathematics, code, general knowledge, and long-context reasoning.
-
+📂 **Open High-Quality Data**: Alongside the model, we are releasing the high-quality training datasets behind it as part of the [UltraData](https://ultradata.openbmb.cn/) family: [UltraX](https://huggingface.co/datasets/openbmb/UltraX-Preview), a high-quality web pre-training dataset; [UltraData-Code](https://huggingface.co/datasets/openbmb/UltraData-Code), featuring L0–L3 tiered code data management to drive a significant leap in coding capabilities; [UltraData-SFT-Agent-2609](https://huggingface.co/datasets/openbmb/UltraData-SFT-Agent-2609), comprising 500K agent training samples to enhance comprehensive on-device agent capabilities; and [UltraData-RL-2609](https://huggingface.co/datasets/openbmb/UltraData-RL-2609), with 80K+ high-quality RL training samples covering mathematics, code, general knowledge, and long-context reasoning.
 
 ## Model List
 
@@ -263,9 +262,7 @@ The training of MiniCPM5-2B is a full-stack practice of **[UltraData Tiered Data
 
 During **base training**, the model goes through stable training and decay training to build core language capability and training stability. It then enters **mid-training** to further strengthen target capabilities and adapt to the target data distribution. The training corpus is released alongside the model as [Ultra-FineWeb](https://huggingface.co/datasets/openbmb/Ultra-FineWeb), [Ultra-FineWeb-L3](https://huggingface.co/datasets/openbmb/Ultra-FineWeb-L3), [UltraX](https://huggingface.co/datasets/openbmb/UltraX-Preview), [UltraData-Code](https://huggingface.co/datasets/openbmb/UltraData-Code) and [UltraData-Math](https://huggingface.co/datasets/openbmb/UltraData-Math).
 
-During **post-training**, we proceed in three steps: **SFT**, **RL**, and **OPD**. We first use **400B tokens of deep-thinking SFT** to establish deep-thinking and general chat abilities; the SFT data is released as [UltraData-SFT-2605](https://huggingface.co/datasets/openbmb/UltraData-SFT-2605). We then train specialized **RL teachers** for math, code, agentic tasks, writing, and related domains(with the corresponding data also open-sourced as [UltraData-RL-2609](https://huggingface.co/datasets/openbmb/UltraData-RL-2609)), and use **On-Policy Distillation (OPD)** to distill these teachers back into one release model.
-
-![MiniCPM5-2B Training Recipe](https://raw.githubusercontent.com/OpenBMB/MiniCPM/minicpm5-2b/assets/minicpm5/minicpm5_2b_training_recipe.jpg)
+During **post-training**, we proceed in three steps: **SFT**, **RL**, and **OPD**. We first use **400B tokens of deep-thinking SFT** to establish deep-thinking and general chat abilities; the SFT data is released as [UltraData-SFT-2605](https://huggingface.co/datasets/openbmb/UltraData-SFT-2605) and the Agent SFT data is released as [UltraData-SFT-Agent-2609](https://huggingface.co/datasets/openbmb/UltraData-SFT-Agent-2609). We then train specialized **RL teachers** for math, code, agentic tasks, writing, and related domains (with the corresponding data also open-sourced as [UltraData-RL-2609](https://huggingface.co/datasets/openbmb/UltraData-RL-2609)), and use **On-Policy Distillation (OPD)** to distill these teachers back into one release model.
 
 ![MiniCPM5-2B Training Recipe](https://raw.githubusercontent.com/OpenBMB/MiniCPM/main/assets/minicpm5/minicpm5_2b_training_recipe.jpg)
 
@@ -275,8 +272,6 @@ During **post-training**, we proceed in three steps: **SFT**, **RL**, and **OPD*
 
 **OPD** merges the capabilities of 16 expert models produced by RL training, including 5 agentic expert models. At each response position, we compute the full-vocabulary reverse KL divergence between student and teacher logits as the advantage estimate, replacing the original verification-based advantage. OPD directly reuses the prompts used to train each RL teacher as distillation data, so no additional corpus construction is required.
 
-![MiniCPM5-2B RL + OPD Gains](https://raw.githubusercontent.com/OpenBMB/MiniCPM/minicpm5-2b/assets/minicpm5/minicpm5_2b_rl_opd_score_gains.png)
-
 ![MiniCPM5-2B RL + OPD Gains](https://raw.githubusercontent.com/OpenBMB/MiniCPM/main/assets/minicpm5/minicpm5_2b_rl_opd_score_gains.png)
 
 ## Quickstart