Lightning Talk: Accelerating PyTorch Performance with OpenVINO
summary
OpenVINO loads PyTorch models either by convert_model -> OpenV IR (inline graph optimizations) or as a PyTorch 2.0 torch.compile backend (TorchDynamo backend with graph partitioning and fallbacks). Supported inputs are torch.nn and TorchScript modules/functions; deployment targets include Intel CPUs, integrated and discrete GPUs, VPUs and upcoming NPU. Installable via pip install openvino and enabled with two lines (import openvino.torch and set backend in torch.compile). Demoed by running Automatic1111 Stable Diffusion on an Arc 770 GPU at about 2 seconds per image and batched stylization of four outputs.
AI-written summary · about
chapters (6)
- 0:00OpenVINO and PyTorch intro
- 1:26OpenVINO toolkit features
- 2:52PyTorch model conversion
- 5:20PyTorch compile backend
- 8:10Stable Diffusion on Intel Arc
- 10:21Q&A
related talks
23mOptimize your network inference time with OpenVINO
42mAI Everywhere– accelerate your development from PC to cloud
20mComputer Vision at the Edge with OpenVINO
14mLightning Talk: Full-Stack PyTorch Robotics VLA: From Data To...
29mEfficient AI with Python: running AI models on CPUs
21mThe New Vision of Compiler-Accelerated
47mGet to Know PyTorch - A Cutting Edge Open Source AI Framework for Deep Learning
29mKeynote