Files gathered from <a target="_blank" rel="ugc" href="https://huggingface.co/bertbobson/Ideogram-4-INT8-ConvRot">https://huggingface.co/bertbobson/Ideogram-4-INT8-ConvRot</a> 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 <a target="_blank" rel="ugc" href="https://github.com/BobJohnson24/ComfyUI-INT8-Fast">https://github.com/BobJohnson24/ComfyUI-INT8-Fast</a>Speed is 1.78x faster(2.03s/it) than FP8(3.62s/it) on a 3090, without compile. 4.4-6.2s/it on my 3060 12gb~2x faster with torch compile.

Ideogram 4 INT8

Uploaded from <a target="_blank" rel="ugc" href="https://huggingface.co/bertbobson/Ideogram-4-INT8-ConvRot">https://huggingface.co/bertbobson/Ideogram-4-INT8-ConvRot</a>

Files with SHA256: 5e2ccce7bea32b6ab765d17d1c3f4b2eb96765cf137bb73cce23ceebb73e4f01