Bring AI experiences to all your Windows Devices
summary
DirectML provides a single DirectX API that runs the same models and code across Intel, AMD, NVIDIA and older integrated GPUs and soon NPUs, with builds available as a NuGet or PyPI package and as part of the Windows SDK. It demonstrates four-bit AWQ quantization that cuts model size ~4x by quantizing 99% of weights while keeping the top 1% precise, runs Phi-3 Mini and Stable Diffusion Turbo locally, and plugs into ONNX Runtime, a new ORT generate API, and PyTorch DirectML for browser and native apps.
AI-written summary · about
chapters (15)
- 0:00Session overview
- 1:31DirectML and AI on Windows
- 3:38How DirectML works
- 7:12Scaling AI on Windows
- 10:27Optimizing models locally
- 14:46DirectML with PyTorch
- 18:18GenAI with PyTorch and DirectML
- 20:26AI toolkit for VS Code
- 23:15AI on the web with WebNN
- 26:10WebNN demos in Edge
- 27:56Khan Academy’s AI tutor
- 30:28WebNN with ORT Web
- 33:41WebNN developer preview
- 35:11Recap and call to action
- 36:17Q&A
related talks
39mQualcomm Technologies and the power of Windows AI Q&A
45mDeliver AI-powered experiences across cloud and edge with Windows
37mHow to Optimize, Validate and Deploy ML Models On Device
36mApplied AI on the edge
07mPyTorch at Microsoft
01h 00mAI for Every Developer
10mDeploying machine learning in wasm to CPU, GPU and NPU with the new wa...
17mRunning Intelligence at the Edge: Web AI in Action