Quantization of Krea 2 Turbo <span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px">primarily meant for low VRAM/RAM usage, expect degradation in quality compared to INT8 Convrot only quants or BF16, saves ~64.3% in storage relative to BF16.<hr />Layer count<ul><li>139 layers in convrot_w4a4_mse (it's just convrot_w4a4 but with additional logic for selecting better scales, reduces error quite a lot numerically but I am unsure about visually)</li><li>78 layers in int8_convrot</li><li>1 layer in convrot_w4a4</li></ul>Once again, used a simple T2I workflow, no second stage or upscaling done as I'm too lazy for that.Tested on:<ul><li>NVIDIA GTX 1660 Super 6 GB</li><li>32 GB System RAM</li></ul>Test settings:<ul><li>8 steps</li><li>CFG 1</li><li>Euler / Simple</li><li>696 × 1048 resolution (0.7 on resolution selector)</li><li>Approximately 9 s/it on a GTX 1660 Super</li></ul>Quantized using my <a rel="ugc" href="https://github.com/bakapotatolord/potatoforge-quantization">toolkit </a>

Files with SHA256: d127dfebe19d2e92354c0b1819aa3ea1c96695ab1db213a485be3112eece0842