Sure, I've experimented with a few PyTorch models, loading them via ONNX. In the new 0.4 release will be availalbe a few models converted this way. Please feel free to share any details of your experiment in issues here https://github.com/JetBrains/KotlinDL/issues