Trying to work out training settings with Illustrious models; Overcooked it, the first time around. Was testing out Illustrious training, and went max on parameters with the first one. The v1.1 has more 'conventional' parameters.
V2.0 onwards are Flux models trained on an entirely different dataset and use 'c0mix2' trigger keyword.
Description
This version has been trained on more 'conventional' (still slightly higher learning rates). Here are details for those interested.
"engine": "kohya",
"unetLR": 0.001,
"clipSkip": 2,
"loraType": "lora",
"keepTokens": 1,
"networkDim": 32,
"numRepeats": 4,
"resolution": 1024,
"lrScheduler": "cosine",
"minSnrGamma": 5,
"noiseOffset": 0.1,
"targetSteps": 1867,
"enableBucket": true,
"networkAlpha": 16,
"optimizerType": "Prodigy",
"textEncoderLR": 0.0002,
"maxTrainEpochs": 10,
"shuffleCaption": true,
"trainBatchSize": 3,
"flipAugmentation": false,
"lrSchedulerNumCycles": 3
FAQ
Details
Files
Available On (9 platforms)
Same model published on other platforms. May have additional downloads or version variants.