update readme, test=document_fix (#33618)

README.md

-<p align="center">
+<p align="center">
 <img align="center" src="doc/imgs/logo.png", width=1600>
 <p>
     
@@ -50,10 +50,9 @@ Now our developers can acquire Tesla V100 online computing resources for free. I
      [Click here to learn more](https://github.com/PaddlePaddle/Fleet)
 
 
-- **Accelerated High-Performance Inference over Ubiquitous Deployments**
+- **High-Performance Inference Engines for Comprehensive Deployment Enviroments**
 
-    PaddlePaddle is not only compatible with other open-source frameworks for models training, but also works well on the ubiquitous developments, varying from platforms to devices. More specifically, PaddlePaddle accelerates the inference procedure with the fastest speed-up. Note that, a recent breakthrough of inference speed has been made by PaddlePaddle on Huawei's Kirin NPU, through the hardware/software co-optimization.
-     [Click here to learn more](https://github.com/PaddlePaddle/Paddle-Lite)
+   PaddlePaddle is not only compatible with models trained in 3rd party open-source frameworks , but also offers complete inference products for various production scenarios. Our inference product line includes [Paddle Inference](https://paddle-inference.readthedocs.io/en/latest/product_introduction/summary.html): Native inference library for high performance server and cloud inference; [Paddle Serving](https://github.com/PaddlePaddle/Serving): A service-oriented framework suitable for distributed and pipeline productions; [Paddle Lite](https://github.com/PaddlePaddle/Paddle-Lite): Ultra-Lightweight inference engine for mobile and IoT enviroments; [Paddle.js](https://www.paddlepaddle.org.cn/paddle/paddlejs): A frontend inference engine for browser and mini apps. Futhermore, by great amounts of optimization with leading hardwares in each scenarios, Paddle inference engines outperform most of the other mainstream frameworks.
      
      
 - **Industry-Oriented Models and Libraries with Open Source Repositories**

README_cn.md

-
+
 <p align="center">
 <img align="center" src="doc/imgs/logo.png", width=1600>
 <p>
@@ -47,10 +47,9 @@ PaddlePaddle用户可领取**免费Tesla V100在线算力资源**，训练模型
     [查看详情](https://github.com/PaddlePaddle/Fleet)
     
 
-- **多端多平台部署的高性能推理引擎**
+- **支持多端多平台的高性能推理部署工具**
 
-    飞桨不仅兼容其他开源框架训练的模型，还可以轻松地部署到不同架构的平台设备上。同时，飞桨的推理速度也是全面领先的。尤其经过了跟华为麒麟NPU的软硬一体优化，使得飞桨在NPU上的推理速度进一步突破。
-    [查看详情](https://github.com/PaddlePaddle/Paddle-Lite)
+    飞桨不仅广泛兼容第三方开源框架训练的模型部署，并且为不同的场景的生产环境提供了完备的推理引擎，包括适用于高性能服务器及云端推理的原生推理库 [Paddle Inference](https://paddle-inference.readthedocs.io/en/latest/product_introduction/summary.html)，面向分布式、流水线生产环境下自动上云、A/B测试等高阶功能的服务化推理框架 [Paddle Serving](https://github.com/PaddlePaddle/Serving)，针对于移动端、物联网场景的轻量化推理引擎 [Paddle Lite](https://github.com/PaddlePaddle/Paddle-Lite)，以及在浏览器、小程序等环境下使用的前端推理引擎 [Paddle.js](https://www.paddlepaddle.org.cn/paddle/paddlejs)。同时，透过与不同场景下的主流硬件高度适配优化及异构计算的支持, 飞桨的推理性能也领先绝大部分的主流实现。
 
 
 - **面向产业应用，开源开放覆盖多领域的工业级模型库。**