After last update inference time has multiplied by factor of 8

Nobody has claimed this yet.

Hello, I updated to latest as well as did --xformers-reinstall and --torch-reinstall due to the console log telling me I can update. When I did a 5 second inference time for a 512x512 @ 20 samples image on a evga gtx 1080 sc multiplied by 8 up to 40 seconds per inference. I popped my settings.json and ui-settings.json and tried again with the same results. Some of my stable diffusion models are now reporting an error, including the included one. This error happens about after half the samples:
```
ONNXRuntimeError] : 2 : INVALID_ARGUMENT : Got invalid dimensions for input: images for the following indices
index: 1 Got: 4 Expected: 3
Please fix either the inputs/outputs or the model.
```

I use the following flags:
```
--api --listen --xformers --always-gpu
```

Start by reproducing a 512x512 image with 20 samples using --api --listen --xformers --always-gpu, comparing the reported five-second and 40-second inference times. Check settings.json, ui-settings.json, and the included model around the ONNXRuntime invalid-dimensions error. Done means the performance regression and the index-1 dimension error are both explained and corrected.

Written by the indexing model from the issue text.