# AI 开发者工作流程文档 Qualcomm AI Runtime SDK 促进在 Qualcomm 硬件上进行高性能机器学习推理。它支持 TensorFlow、PyTorch、ONNX 和 LiteRT 等框架。它旨在 Qualcomm 硬件上快速高效地运行已经训练过的网络。 ## 概述概述 https://docs.qualcomm.com/doc/80-70022-15BY/topic/ai-ml-developer-workflow.html ## 编译和优化 AI 模型涵盖模型转换、量化和精度微调。 ![aihubicon](data:image/png;base64,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) 使用 AI Hub 优化 AI 模型使用 AI Hub 可在不到 5 分钟的时间内自带模型或尝试预先优化的模型。 https://docs.qualcomm.com/doc/80-70022-15BY/topic/ai-hub.html ![tfliteicon](data:image/png;base64,UklGRn4DAABXRUJQVlA4THIDAAAvTAATAAkGbSM5UnL9bvgTvi8UIvo/AXwv1qdzUmsFTjNDheSqApiBb/6rSZ5eF0BC+P9dH4gB6X8ag0WOtn+K21gQZmbGkqFkZs4xcoQcwUdIqVKlS5cuVbrc0qVKM86MZ3Y3TF/syf8xaYyRFA7HkWyryq16SeBuyx8A7kHYt0D+x5cs3TUE9xcAAbizJABuERDcNnIkVTVqvZpQHXzhDX3Ax7ds/8J4aiJaF7ylIuXLK6a1VUtF1OqFYESB9keTWGqlYUyPjDQgAD1IGy1JtaSVpxf7EDUzMEJEW6QkxXyzVKcXQQ/JQgZp7bWtkxK1IhtRcpgBEYIWAFnEt18SiSJb6jGmTlzUyIJUKoOcegIPlVpFkd8ECYLpCwIIizB2B0xPlh99SAaGEjFiVYqMx9ojWk+2Uv4IhMGUYMLGNr+pjsfVpKU3eN7KXMUOi9qG8iNqxSGapOYMGdRmHWyxiABMU6+2X1yb/Cg/fmf++k5YYBEhC4YZ/enBesHiJOZ/RJQ+BE0rmIABG27XZ9SLI6VMwRK1rEVLqyjyR3TviBHbFgH5ncFmsSRZl3bIozBkak0qehhsKIY5BRy9qRdbLNprDHNzUwGDBisMQlCwdrujRlu1wFoDltiDoY1rMeFj7Msf2GHur8+3PIRg2DdHGQTY673fArD199HY88NR1iNU1Ils9bpeBq8osPM3kCHna9fXW73qDgmuIbCw16lNwt7fDWSl5va8pvSoOBeOsdclbXusMUCO/N/pfHw4Jnncy5hRYe1v5a45Dsa2ej1uxT9vaedmj4H59osOqC41OJXr0rGYdnBOWJrZYrS1hZ4bnd/XGqzDvxhrlnLEQvnyreitGeBW57kWsaSDv5UHqsektRp15MmoUzhjMfzaGUfNLMHhIw6M1tHjnO7YUCrEFs7+jrfE4G/lLvOXVGAhRXgoEsO2fxXTgAXur6ueEAIpwf58bXyrF9KCgdH78/trDynxN2TI5Ojjw3jy/jbl5P+DsbmUYDUzpJ4aDJAgtiukwN/DlFRheONgkVRh3ka40VGrVKRivBCEkNxoNNQAMeGgTmOpJqmx17tWgRheot1fPSclh85LYyCAYWz1JqgAf4z1GI1zvbVSQ8I5NtMitV30f2XP10hBnpJKhWhc/gytZpxUZA4ul7YybXv89RHvSQk=) 使用 LiteRT 优化 AI 模型将您的 LiteRT AI 模型移植到 Qualcomm 平台。 https://docs.qualcomm.com/doc/80-70022-15BY/topic/tflite.html 使用 Qualcomm AI Runtime SDK 优化 AI 模型使用一体化、可轻松自定义的 Qualcomm AI Runtime (QAIRT) SDK 移植您的模型 https://docs.qualcomm.com/doc/80-70022-15BY/topic/qairt.html ## 运行 AI/ML 示例应用运行默认示例应用来运行 AI/ML 模型，或者使用 Qualcomm 评估套件或 [Qualcomm Device Cloud (QDC)](https://qdc.qualcomm.com) 自定义示例应用。使用 EVK 运行一个示例应用。使用默认示例应用和 EVK 对图像进行分类。 https://docs.qualcomm.com/doc/80-70022-15BY/topic/sample-app-evk.html 使用 QDC 运行示例应用使用默认示例应用和 Qualcomm Device Cloud (QDC) 对图像进行分类。 https://docs.qualcomm.com/doc/80-70022-15BY/topic/sample-app-qdc.html ## 开发您自己的应用程序介绍如何在 Qualcomm 平台上实现自己的应用程序。这是一个高级应用篇，适用于经验丰富的开发者。使用 Qualcomm Visual Studio Code Extension 开发您自己的 AI/ML 应用程序 https://docs.qualcomm.com/doc/80-70022-15BY/topic/develop-your-own-application-ide.html 使用 Qualcomm IM SDK 开发您自己的 AI/ML 应用程序 https://docs.qualcomm.com/doc/80-70022-15BY/topic/develop-your-own-application-im-sdk.html ## 使用 AI Hub 模型和标签与 GStreamer API 说明如何在 Qualcomm 评估套件上如何将 AI Hub 模型和标签与 gstreamer 命令结合使用。对图像进行分类 https://docs.qualcomm.com/doc/80-70022-15BY/topic/image-classification.html 检测对象 https://docs.qualcomm.com/doc/80-70022-15BY/topic/object-detection.html 对视频帧应用语义分割 https://docs.qualcomm.com/doc/80-70022-15BY/topic/semantic-segmentation.html 使用超分辨率对图像进行放大 https://docs.qualcomm.com/doc/80-70022-15BY/topic/super-resolution.html ## 故障排除和常见问题解答进一步支持获取常见问题的答案并了解如何获取更多帮助。 https://docs.qualcomm.com/doc/80-70022-15BY/topic/troubleshooting.html ## 模型移植最佳实践使用 Neural Processing Engine SDK 导出 YoloV8 模型 https://docs.qualcomm.com/doc/80-70022-15BY/topic/appx-export-yolov8.html 运行演示 https://docs.qualcomm.com/doc/80-70022-15BY/topic/appx-export-yolov8.html#appx-run-the-demo Last Published: Nov 05, 2025 [Next Topic 概述](https://docs.qualcomm.com/bundle/publicresource/80-70022-15BY/topics/ai-ml-developer-workflow.md)