Text-to-Image Generation on MJHQ
1.0438IRTMPDiff
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| TMPDiffPrecision/Quantization Scheme=Ours, Model=Sana, Resolution=1024x1024, Guidance scale=4.5, Number of samples=10242026.03 | 1.0438 | 27.163 | 55.99 | — | — | |
| w4a4r32Precision/Quantization Scheme=w4a4r32, Model=Sana, Resolution=1024x1024, Guidance scale=4.5, Number of samples=10242026.03 | 1.0424 | 27.172 | 55.947 | — | — | |
| FP16 BaselineFormat=FP16, Precision (W-A)=16-162026.04 | 0.944 | — | 16.6 | — | — | |
| bfloat16Precision/Quantization Scheme=bfloat16, Model=Sana, Resolution=1024x1024, Guidance scale=4.5, Number of samples=10242026.03 | 0.9239 | 26.971 | 57.196 | — | — | |
| LoRaQFormat=MXINT4, Precision (W-A)=4-42026.04 | 0.901 | — | 15.4 | 0.339 | 16.8 | |
| SVDQuantFormat=MXINT4, Precision (W-A)=4-42026.04 | 0.738 | — | 18.9 | 0.424 | 15.9 |