New! Added quantized versions!There is:Q5_1, 5-bit round-to-nearest quantization. Small but still ok(ish) quality.Q2_K, 2-bit quantization. Very small. Quality...not that good, but useable?<a target="_blank" rel="ugc" href="https://github.com/leejet/stable-diffusion.cpp/blob/master/docs/quantization_and_gguf.md">https://github.com/leejet/stable-diffusion.cpp/blob/master/docs/quantization_and_gguf.md</a><hr />This is a small test of the new functionality in <a target="_blank" rel="ugc" href="https://github.com/Runew0lf/RuinedFooocus">RuinedFooocus</a> that let you create .merge-files and use them as models. Instead of keeping dozens of models you can have a small file with the recipe for the merge and it will merge the model for you on-the-fly when it is needed.<pre><code>{
 "comment": "Top 5 models merge 2024-07-06",
 "base": {
 "name": "cyberrealisticXL_v21.safetensors",
 "weight": 2742
 },
 "models": [
 {
 "name": "halcyonSDXL_v17.safetensors",
 "weight": 1620
 },
 {
 "name": "corEpicaByZovya_v0Beta.safetensors",
 "weight": 483
 },
 {
 "name": "eldritchPhotography_v10.safetensors",
 "weight": 369
 },
 {
 "name": "enjoyXLSuperRealistic_v30ModifiedVersion.safetensors",
 "weight": 933
 }
 ],
 "normalize": 1.0,
 "cache": false
}</code></pre>Don't take this model too serious. It is simply a test-merge of the top 5 models, weighted according to the number of downloads they've had.

Top5

Q2_K, 2-bit quantization, SDXL model.

Files with SHA256: 07f38c12b0923db6c9c4a5ff51b33cad68bd2915af24e1b110b7a7c87d304013