notebooks/stable-diffusion-torchdynamo-backend/README.md

Image Generation with Stable Diffusion using OpenVINO TorchDynamo backend

This notebook demonstrates how to use a Stable Diffusion model for image generation with OpenVINO TorchDynamo backend. The torch.compile feature enables you to use OpenVINO for PyTorch-native applications. It speeds up PyTorch code by JIT-compiling it into optimized kernels.

By default, Torch code runs in eager-mode, but with the use of torch.compile it goes through the following steps:

1. Graph acquisition - the model is rewritten as blocks of subgraphs that are either:
   • compiled by TorchDynamo and “flattened”,
   • falling back to the eager-mode, due to unsupported Python constructs (like control-flow code).
2. Graph lowering - all PyTorch operations are decomposed into their constituent kernels specific to the chosen backend.
3. Graph compilation - the kernels call their corresponding low-level device-specific operations.

Notebook Contents

This notebook demonstrates how to run stable diffusion using OpenVINO TorchDynamo backend.

Notebook contains the following steps:

1. Create PyTorch models pipeline using Diffusers library.
2. Import OpenVINO backend using torch.compile.
3. Run Stable Diffusion pipeline with OpenVINO TorchDynamo backend.

Installation Instructions

This is a self-contained example that relies solely on its own code.
We recommend running the notebook in a virtual environment. You only need a Jupyter server to start. For details, please refer to Installation Guide.