Skip to content

Latest commit

 

History

History
43 lines (32 loc) · 1.77 KB

README.md

File metadata and controls

43 lines (32 loc) · 1.77 KB

Stable Diffusion Inference

Description

This document has instructions for running inference based on a text prompt using the KerasCV implementation of stability.ai's text-to-image model, Stable Diffusion. Stable Diffusion is a powerful, open-source text-to-image generation model.

Run the model

Run on Linux

  1. Install the Optimized TensorFlow:

  2. Install model dependencies under requirements.txt.

# cd to your model zoo directory
cd models
export PRECISION=<set the precision to "fp32" or "bfloat32" or "bfloat16" or "fp16"">
export OUTPUT_DIR=<path to the directory where i) log files will be saved and ii) (only in case of accuracy run) original & generated images will be written>
# For a custom batch size, set env var `BATCH_SIZE` or it will run with a default value.
export BATCH_SIZE=<customized batch size value>

Inference

  1. inference_throughput_multi_instance.sh Runs multi instance batch inference using 1 socket per instance with the specified precision (fp32, bfloat32, bfloat16, or fp16) with 200 steps. Waits for all instances to complete, then prints a summarized throughput value.
./models_v2/tensorflow/stable_diffusion/inference/cpu/inference_throughput_multi_instance.sh

Accuracy

  1. accuracy.sh
# Additionally, set DATASET_DIR for accuracy runs. The original and the images generated by the model will be saved under OUTPUT_DIR.
export DATASET_DIR=<path to the dataset directory>
./models_v2/tensorflow/stable_diffusion/inference/cpu/accuracy.sh