We are excited to share the Neural Magic demo. Now you can see what we've been working on for the past few months.
This is the demo we showed at NeurIPS in Vancouver last week. We showed it 363 times in 4 days!
The video shows how it's possible to achieve GPU-class performance on commodity CPUs for deep learning by using the Neural Magic inference engine. Our inference engine allows you to increase performance, save cost and energy, all without sacrificing accuracy or changing your model. In this video, we focus on ResNet50 and MobileNetV2 models, but are capable to run others like Google's EfficientNet B0.
If you want to run Neural Magic on your systems, check out our GitHub repos.