This project explores and demonstrates how to deploy stable diffusion model with TensorFlow Serving.

• multiple TensorFlow Serving

  • three TensorFlow Servings for text encoder, diffuser, and decoder
  • a central back-end server (or a client) to call each TensorFlow Servings sequentially
  • enable/disable GPU/TPU capability for each TensorFlow Serving
  • enable dynamic batching capability

• one signature architecture (TBD)

  • a signature to expose an endpoint, and make sure text encoder, diffuser, decoder models are called sequentially internally
  • enable GPU/TPU capability
  • enable dynamic batching capability