Spaces:
Running
Running
README: speed line
Browse files
README.md
CHANGED
|
@@ -31,7 +31,7 @@ and the TAESD decoder all run on your own GPU through WebGPU (onnxruntime-web).
|
|
| 31 |
|
| 32 |
## Speed and accuracy
|
| 33 |
|
| 34 |
-
- About 6–
|
| 35 |
setup per visit.
|
| 36 |
- With identical initial noise, the browser output matches the PyTorch reference pipeline within 5/255
|
| 37 |
per pixel (run the self-test with `?parity=1`).
|
|
|
|
| 31 |
|
| 32 |
## Speed and accuracy
|
| 33 |
|
| 34 |
+
- About 6–7 s per 50-step 256×256 image on an RTX 4060, after roughly 10 s of one-time model
|
| 35 |
setup per visit.
|
| 36 |
- With identical initial noise, the browser output matches the PyTorch reference pipeline within 5/255
|
| 37 |
per pixel (run the self-test with `?parity=1`).
|