[MNN:Sync] Sync Internal 3.1.4.

Author: wangzhaode

LICENSE.txt

@@ -1,18 +1,12 @@
 ![MNN](doc/banner.png)
----
-[![License](https://img.shields.io/github/license/alibaba/MNN)](LICENSE.txt)
-[![Documentation](https://img.shields.io/badge/Documentation-Read-green)](https://mnn-docs.readthedocs.io/en/latest/)
-[![中文版本](https://img.shields.io/badge/Language-%E7%AE%80%E4%BD%93%E4%B8%AD%E6%96%87-green)](README_CN.md)
-[![日本語バージョン](https://img.shields.io/badge/Language-%E6%97%A5%E6%9C%AC%E8%AA%9E-green)](README_JP.md)
-[![MNN Homepage](https://img.shields.io/badge/Homepage-Visit-green)](http://www.mnn.zone)
+
+[中文版本](README_CN.md)
+
+[日本語バージョン](README_JP.md)
+
+[MNN Homepage](http://www.mnn.zone)
 
 ## News 🔥
-- [2025/04/30] android app support qwen3 and dark mode [MnnLlmChat](./apps/Android/MnnLlmChat/README.md#version-040).
-<p align="center">
-  <img width="20%" alt="Icon"  src="https://meta.alicdn.com/data/mnn/qwen_3.gif" style="margin: 0 10px;">
-</p>
-<details>
-<summary> History News </summary>
 - [2025/02/18] iOS multimodal LLM App is released [MNN LLM iOS](./apps/iOS/MNNLLMChat/README.md).
 <p align="center">
   <img width="20%" alt="Icon"  src="./apps/iOS/MNNLLMChat/assets/introduction.gif" style="margin: 0 10px;">
@@ -30,7 +24,6 @@
   <img width="20%" alt="Icon" src="./apps/Android/MnnLlmChat/assets/image_sound.jpg" style="margin: 0 10px;">
   <img width="20%" alt="Icon" src="./apps/Android/MnnLlmChat/assets/image_image.jpg" style="margin: 0 10px;">
 </p>
-</details>
 
 ## Intro
 MNN is a highly efficient and lightweight deep learning framework. It supports inference and training of deep learning models and has industry-leading performance for inference and training on-device. At present, MNN has been integrated into more than 30 apps of Alibaba Inc, such as Taobao, Tmall, Youku, DingTalk, Xianyu, etc., covering more than 70 usage scenarios such as live broadcast, short video capture, search recommendation, product searching by image, interactive marketing, equity distribution, security risk control. In addition, MNN is also used on embedded devices, such as IoT.

README.md

@@ -1,11 +1,10 @@
 ![MNN](doc/banner.png)
----
-[![License](https://img.shields.io/github/license/alibaba/MNN)](LICENSE.txt)
-[![Documentation](https://img.shields.io/badge/Documentation-Read-green)](https://mnn-docs.readthedocs.io/en/latest/)
-[![English Version](https://img.shields.io/badge/Language-English-green)](README.md)
-[![日本語バージョン](https://img.shields.io/badge/Language-%E6%97%A5%E6%9C%AC%E8%AA%9E-green)](README_JP.md)
-[![MNN Homepage](https://img.shields.io/badge/Homepage-Visit-green)](http://www.mnn.zone)
 
+[English Version](README.md)
+
+[日本語バージョン](README_JP.md)
+
+[MNN Homepage](http://www.mnn.zone)
 
 [MNN](https://github.com/alibaba/MNN)是一个轻量级的深度神经网络引擎，支持深度学习的推理与训练。适用于服务器/个人电脑/手机/嵌入式各类设备。目前，MNN已经在阿里巴巴的手机淘宝、手机天猫、优酷等30多个App中使用，覆盖直播、短视频、搜索推荐、商品图像搜索、互动营销、权益发放、安全风控等场景。
 
@@ -41,15 +40,15 @@ MNN文档：
 
 ## 整体特点
 
-### 轻量性
+### 轻量性 
 
-- 主体功能（模型推理CPU+GPU）无任何依赖，代码精简，可以方便地部署到移动设备和各种嵌入式设备中。
+- 主体功能（模型推理CPU+GPU）无任何依赖，代码精简，可以方便地部署到移动设备和各种嵌入式设备中。 
    - iOS平台：功能全开的MNN静态库 armv7+arm64大小12MB左右，链接生成可执行文件增加大小2M左右。可裁剪主体功能后静态库大小6.1M ，链接生成可执行文件增加大小 600 KB。
    - Android平台：主体功能 armv7a - c++_shared 动态库大小800KB左右。
 - 支持采用 Mini 编辑选项进一步降低包大小，大约能在上述库体积基础上进一步降低 25% 左右。
 - 支持模型FP16/Int8压缩与量化，可减少模型50% - 75% 的体积
 
-### 通用性
+### 通用性 
 
 - 支持 Tensorflow、Caffe、ONNX、Torchscripts 等主流模型文件格式，支持CNN / RNN / GAN / Transformer 等主流网络结构。
 - 支持多输入多输出，支持任意维度的输入输出，支持动态输入（输入大小可变），支持带控制流的模型
@@ -110,7 +109,7 @@ MNN适配的硬件架构与精度详见下表：
 ## 社区交流与反馈
 钉钉群组：
 
-- 钉钉群1:23329087
+- 钉钉群1:23329087 
 - 钉钉群2:23350225
 - 钉钉群3:扫描二维码加入
 

README_CN.md

@@ -1,18 +1,18 @@
 ![MNN](doc/banner.png)
----
-[![License](https://img.shields.io/github/license/alibaba/MNN)](LICENSE.txt)
-[![Documentation](https://img.shields.io/badge/Documentation-Read-green)](https://mnn-docs.readthedocs.io/en/latest/)
-[![English Version](https://img.shields.io/badge/Language-English-green)](README.md)
-[![中文版本](https://img.shields.io/badge/Language-%E7%AE%80%E4%BD%93%E4%B8%AD%E6%96%87-green)](README_CN.md)
-[![MNN ホームページ](https://img.shields.io/badge/Homepage-Visit-green)](http://www.mnn.zone)
+
+[English Version](README.md)
+
+[中文版本](README_CN.md)
+
+[MNN ホームページ](http://www.mnn.zone)
 
 ## ニュース 🔥
 - [2025/01/23] フルマルチモーダル LLM Android アプリをリリースしました: [MNN-LLM-Android](./project/android/apps/MnnLlmApp/README.md)。テキストからテキスト、画像からテキスト、音声からテキスト、テキストから画像生成を含みます。
 <p align="center">
-  <img width="20%" alt="Icon"  src="./apps/Android/MnnLlmChat/assets/image_home.jpg" style="margin: 0 10px;">
-  <img width="20%" alt="Icon" src="./apps/Android/MnnLlmChat/assets/image_diffusion.jpg" style="margin: 0 10px;">
-  <img width="20%" alt="Icon" src="./apps/Android/MnnLlmChat/assets/image_sound.jpg" style="margin: 0 10px;">
-  <img width="20%" alt="Icon" src="./apps/Android/MnnLlmChat/assets/image_image.jpg" style="margin: 0 10px;">
+  <img width="20%" alt="Icon"  src="./project/android/apps/MnnLlmApp/assets/image_home.jpg" style="margin: 0 10px;">
+  <img width="20%" alt="Icon" src="./project/android/apps/MnnLlmApp/assets/image_diffusion.jpg" style="margin: 0 10px;">
+  <img width="20%" alt="Icon" src="./project/android/apps/MnnLlmApp/assets/image_sound.jpg" style="margin: 0 10px;">
+  <img width="20%" alt="Icon" src="./project/android/apps/MnnLlmApp/assets/image_image.jpg" style="margin: 0 10px;">
 </p>
 
 ## 紹介

README_JP.md

@@ -36,12 +36,11 @@ static void _makeResource(Backend* backend, std::shared_ptr<CPUConvolution::Reso
     auto alphaPtr = resource->mDequantize.mScaleBias->host<float>();
     auto biasPtr = reinterpret_cast<float*>(reinterpret_cast<uint8_t*>(alphaPtr) + ocUp4 * core->bytes);
     ::memset(alphaPtr, 0, 2 * ocUp4 * core->bytes);
-    auto wZero = resourceInt8->mWeightQuantZero->host<int32_t>(); // has packed to outputUp4
     auto wScale = resourceInt8->mOriginScale->host<float>();
     int h = ocUp4;
     for (int i=0; i< h; ++i) {
         alphaPtr[i] = wScale[i];
-        biasPtr[i] = (-1.f) * wZero[i] * wScale[i];
+        biasPtr[i] = wScale[i + ocUp4];
     }
 }
 
@@ -185,8 +184,8 @@ ErrorCode GemmInt8Executor::onExecute(const std::vector<Tensor *> &inputs, const
 
     quanParam.useInt8 = 0; // Save result as float data type.
     quanParam.biasFloat = reinterpret_cast<float*>(mQuantBias.data());
-    quanParam.weightQuanBias = mKernelSum.data();
-    quanParam.extraScale = nullptr;
+    quanParam.weightKernelSum = mKernelSum.data();
+    quanParam.inputScale = nullptr;
     float dequantScale = mMutableResource.mResource->mInputScale;
 
     SumByAxisParams sumParams;