File size: 888 Bytes
fbf0718 086eec4 83d79aa ed67a8d f0dc99b 19e96f2 3dd8642 5faeb5e a5ce63e ed67a8d baf30ab | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 | ---
license: other
license_name: ideogram-4-non-commercial
license_link: https://huggingface.co/ideogram-ai/ideogram-4-fp8/blob/main/LICENSE.md
pipeline_tag: text-to-image
tags:
- text-to-image
- image-generation
- diffusion
- flow-matching
- dit
- ideogram
---
This is an unavoidable double quantization due to the release state of Ideogram4.
The FP8 weights were cast to FP32 with the FP8 scales, then downcast to BF16 before being converted to INT8.
For use in ComfyUI with https://github.com/BobJohnson24/ComfyUI-INT8-Fast
Speed is 1.78x faster(2.03s/it) than FP8(3.62s/it) on my 3090, without compile.
<s>~2x faster with torch compile.</s>
After further inspection, it appears there may be quality issues with torch compiling this model.
Quick comparison:
<img src="Comparison.jpg" width="1000" height="500">
<img src="Comparison2.jpg" width="1000" height="500"> |