diff --git a/tutorials/source_en/beginner/images/introduction3.png b/tutorials/source_en/beginner/images/introduction3.png new file mode 100644 index 0000000000000000000000000000000000000000..9075349cb1172c59ef879439f2e694a801ae532f Binary files /dev/null and b/tutorials/source_en/beginner/images/introduction3.png differ diff --git a/tutorials/source_en/beginner/introduction.md b/tutorials/source_en/beginner/introduction.md index 15d35403b024bf8412eca1daa00e103ad74c26ac..14f59e204cf92994c7c23ab87ab3811b6bfb1372 100644 --- a/tutorials/source_en/beginner/introduction.md +++ b/tutorials/source_en/beginner/introduction.md @@ -20,6 +20,7 @@ The following figure shows the overall MindSpore architecture: - **Extend**: The MindSpore expansion package supports new domain scenarios, such as GNN, deep probabilistic programming, and reinforcement learning. More developers are expected to contribute and build the library. - **Science**: MindScience is a scientific computing kit for various industries based on the converged MindSpore framework. It contains the industry-leading datasets, basic network structures, high-precision pre-trained models, and pre-and post-processing tools that accelerate application development of the scientific computing ([More Information](https://mindspore.cn/mindscience/docs/en/master/index.html)). - **Expression**: Python-based frontend expression and programming interfaces. In the future, Huawei plans to continue to provide interconnection with third-party front-end systems, such as C/C++ and Huawei-developed programming language front-end (currently in the pre-research phase), to introduce more third-party ecosystems. +- **Third-party front-end**: support for third-party multilingual front-end expression, the future plans to successively provide C/C++, Huawei self-developed programming language front-end - Cangjie (currently still in the pre-research stage) and other third-party front-end docking work, and the introduction of more third-party ecology. - **Data**: provides functions such as efficient data processing, common dataset loading and programming interfaces, and allows users to flexibly define processing registration and pipeline parallel optimization. - **Compiler**: The core compiler of the layer, which implements three major functions based on the unified device-cloud MindIR, including hardware-independent optimization (type derivation, automatic differentiation, and expression simplification), hardware-related optimization (automatic parallelism, memory optimization, graph kernel fusion, and pipeline execution), and optimization related to deployment and inference (quantification and pruning). - **Runtime**: MindSpore runtime system, including the runtime system on the cloud host, runtime system on the device, and lightweight runtime system of the IoT platform. @@ -73,13 +74,13 @@ After the neural network model is trained, you can export the model or load the ### API Level Structure -To support network building, entire graph execution, subgraph execution, and single-operator execution, MindSpore provides users with three levels of APIs. In descending order, these are High-Level Python API, Medium-Level Python API, and Low-Level Python API. +MindSpore provides users with three different levels of APIs to support AI application (algorithm/model) development, from high to low: High-Level Python API, Medium-Level Python API and Low-Level Python API. The High-Level API provides better encapsulation, the Low-Level API provides better flexibility, and the Mid-Level API combines flexibility and encapsulation to meet the needs of developers in different fields and levels. ![MindSpore API](https://mindspore-website.obs.cn-north-4.myhuaweicloud.com/website-images/r1.9/tutorials/source_en/beginner/images/introduction3.png) - High-Level Python API - High-level APIs are at the first layer. Based on the medium-level API, these advanced APIs include training and inference management, mixed precision training, and debugging and optimization, enabling users to control the execution process of the entire network and implement training, inference, and optimization of the neural network. For example, by utilizing the Model API, users can specify the neural network model to be trained as well as related training settings, train the neural network model, and debug the neural network performance through the Profiler API. + High-level APIs are at the first layer. Based on the medium-level API, these advanced APIs include training and inference management, mixed precision training, and debugging and optimization, enabling users to control the execution process of the entire network and implement training, inference, and optimization of the neural network. For example, by utilizing the Model API, users can specify the neural network model to be trained as well as related training settings, train the neural network model. - Medium-Level Python API diff --git a/tutorials/source_zh_cn/beginner/images/introduction2.png b/tutorials/source_zh_cn/beginner/images/introduction2.png index c8bd3c1d6970e1bab4f17c712f641b6ce30de354..2427a7215cc31c86621262c51f83946563c0b8cd 100644 Binary files a/tutorials/source_zh_cn/beginner/images/introduction2.png and b/tutorials/source_zh_cn/beginner/images/introduction2.png differ diff --git a/tutorials/source_zh_cn/beginner/images/introduction3.png b/tutorials/source_zh_cn/beginner/images/introduction3.png index 26ee41747d16bfb9c47a6bb699deeb425d244290..9075349cb1172c59ef879439f2e694a801ae532f 100644 Binary files a/tutorials/source_zh_cn/beginner/images/introduction3.png and b/tutorials/source_zh_cn/beginner/images/introduction3.png differ diff --git a/tutorials/source_zh_cn/beginner/introduction.ipynb b/tutorials/source_zh_cn/beginner/introduction.ipynb index ee4bb7b82f21e2bcd9943485190ae93d80d86cf6..eea384cb4fa8f814eef00c0409080331b02a4cd4 100644 --- a/tutorials/source_zh_cn/beginner/introduction.ipynb +++ b/tutorials/source_zh_cn/beginner/introduction.ipynb @@ -34,7 +34,8 @@ "- **ModelZoo（模型库）**：ModelZoo提供可用的深度学习算法网络，也欢迎更多开发者贡献新的网络([ModelZoo地址](https://gitee.com/mindspore/models))。\n", "- **Extend（扩展库）**：昇思MindSpore的领域扩展库，支持拓展新领域场景，如GNN/深度概率编程/强化学习等，期待更多开发者来一起贡献和构建。\n", "- **Science（科学计算）**：MindScience是基于昇思MindSpore融合架构打造的科学计算行业套件，包含了业界领先的数据集、基础模型、预置高精度模型和前后处理工具，加速了科学行业应用开发（[了解更多](https://mindspore.cn/mindscience/docs/zh-CN/master/index.html)）。\n", - "- **Expression（全场景统一API）**：基于Python的前端表达与编程接口。同时未来计划陆续提供C/C++、华为自研编程语言前端-仓颉（目前还处于预研阶段）等第三方前端的对接工作，引入更多的第三方生态。\n", + "- **Expression（全场景统一API）**：基于Python的前端表达与编程接口，支持两个融合（函数/OOP编程范式融合、AI+数值计算表达融合）以及两个统一（动静表达统一、单机分布式表达统一）。\n", + "- **第三方前端**：支持第三方多语言前端表达，未来计划陆续提供C/C++、华为自研编程语言前端-仓颉（目前还处于预研阶段）等第三方前端的对接工作，引入更多的第三方生态。\n", "- **Data（数据处理层）**：提供高效的数据处理、常用数据集加载等功能和编程接口，支持用户灵活地定义处理注册和pipeline并行优化。\n", "- **Compiler（AI编译器）**：图层的核心编译器，主要基于端云统一的MindIR实现三大功能，包括硬件无关的优化（类型推导、自动微分、表达式化简等）、硬件相关优化（自动并行、内存优化、图算融合、流水线执行等）、部署推理相关的优化（量化、剪枝等）。\n", "- **Runtime（全场景运行时）**：昇思MindSpore的运行时系统，包含云侧主机侧运行时系统、端侧以及更小IoT的轻量化运行时系统。\n", @@ -90,13 +91,13 @@ "\n", "### 层次结构\n", "\n", - "昇思MindSpore向用户提供了3个不同层次的API，支撑用户进行网络构建、整图执行、子图执行以及单算子执行，从高到低分别为High-Level Python API、Medium-Level Python API以及Low-Level Python API。\n", + "昇思MindSpore向用户提供了3个不同层次的API，支撑用户进行AI应用（算法/模型）开发，从高到低分别为High-Level Python API、Medium-Level Python API以及Low-Level Python API。高阶API提供了更好的封装性，低阶API提供更好的灵活性，中阶API兼顾灵活及封装，满足不同领域和层次的开发者需求\n", "\n", "![MindSpore API](https://mindspore-website.obs.cn-north-4.myhuaweicloud.com/website-images/r1.9/tutorials/source_zh_cn/beginner/images/introduction3.png)\n", "\n", "- High-Level Python API\n", "\n", - " 第一层为高阶API，其在中阶API的基础上又提供了训练推理的管理、混合精度训练、调试调优等高级接口，方便用户控制整网的执行流程和实现神经网络的训练推理及调优。例如用户使用Model接口，指定要训练的神经网络模型和相关的训练设置，对神经网络模型进行训练，通过Profiler接口调试神经网络性能。\n", + " 第一层为高阶API，其在中阶API的基础上又提供了训练推理的管理、混合精度训练、调试调优等高级接口，方便用户控制整网的执行流程和实现神经网络的训练推理及调优。例如用户使用Model接口，指定要训练的神经网络模型和相关的训练设置，对神经网络模型进行训练。\n", "\n", "- Medium-Level Python API\n", "\n",