<h2 id="why-t4joihywy">Why?</h2>This LoRA model aims to solve the "strawberry" problem of ML image generation models."Oh, your billion dollar algorithm can not solve the full wine glass problem!"Not anymore.<h2 id="open-source-is-for-everyone-i2wmle6yx">Open Source is For Everyone</h2>Open source is not bound by huge corporate workflows and processes. It took me 32 minutes to train this model after manually captioning 20 images.Training Details:Epochs: 12Steps: 1920Optimizer: --optimizer_type=adopt.ADOPTLR: 8e-5TE LR: 4e-5Scheduler: constant_with_warmup 2% (important to warm the refrigerated wine)Rank: 128/64Debiased Est. Loss: TrueNo flipping or caption shuffling because flipping would not work well with clocks (duh).<h3 id="-kuwqwt79p"></h3><h3 id="problems-encountered-during-preparation-and-training:-azanri29r">Problems Encountered During Preparation &amp; Training:</h3><ul><li>It's slightly harder to find full wine glass pictures and clocks showing different time (duh). Thanks reddit for the clocks and instagram for full wine glasses. Some people are animals.</li><li>Turns out I forgot how to read analog clocks. Some images might have been captioned wrong.</li><li>The wine and clock biases are harder to fully correct than I expected.</li><li>OOD generations still underperform. More examples and steps might be needed.</li></ul>

Breaking New Grounds - Full Glass of Wine + Diverse Clocks

<ul><li>Trained with bf16 mixed precision and saved as float to increase accuracy.</li></ul>

Files with SHA256: 1a9d132a4ca55a3c6338eb6db99b1df39887a008f253db0c3f0c446d50711298