Closed rickstaa closed 1 month ago
What does this pull request do? Explain your changes. (required)
Updates the formula used to calculate latency score for T2I request so that num_inference_steps is included in the measurement.
num_inference_steps
Specific updates (required)
took.Seconds() / float64(outPixels) / (float64(numImages) * float64(numInferenceSteps))
How did you test each of these updates (required)
Tested with high-end range values for num_inference_steps and num_images_per_prompt
num_images_per_prompt
Results: 25 inference 0.00000006567261756897 - 1 image 0.00000006524371048584 - 25 images 100 inference 0.00000004808495975494 - 1 image 0.00000004837596923332 - 25 images
Does this pull request close any open issues?
LIV-362
Checklist:
make
./test.sh
What does this pull request do? Explain your changes. (required)
Updates the formula used to calculate latency score for T2I request so that
num_inference_steps
is included in the measurement.Specific updates (required)
took.Seconds() / float64(outPixels) / (float64(numImages) * float64(numInferenceSteps))
How did you test each of these updates (required)
Tested with high-end range values for
num_inference_steps
andnum_images_per_prompt
Does this pull request close any open issues?
LIV-362
Checklist:
make
runs successfully./test.sh
pass