Update README.md

README.md

@@ -1,5 +1,39 @@
----
-license: other
-license_name: fair-ai-public-license-1.0-sd
-license_link: LICENSE
----
+---
+license: other
+license_name: fair-ai-public-license-1.0-sd
+license_link: LICENSE
+base_model:
+- CabalResearch/NoobAI-Flux2VAE-RectifiedFlow
+pipeline_tag: text-to-image
+library_name: diffusers
+---
+
+# Small experimental finetune on top of NoobAI Flux2VAE RF v0.1
+- Original [NoobAI-Flux2VAE-RectifiedFlow](https://huggingface.co/CabalResearch/NoobAI-Flux2VAE-RectifiedFlow) by Cabal Research
+
+# Purpose:
+- Add knowledge of some artists and game characters
+- Feeling whimiscal
+
+# Training Details:
+- Trained at resolution of **1280**
+## First stage:
+...
+## Second stage:
+```
+accelerate launch /content/sd-scripts/sdxl_train.py --pretrained_model_name_or_path "/content/drive/MyDrive/RFNai/Flux2V1/RFNAIFLux2V-000001.safetensors" --vae "/content/drive/MyDrive/diffusion_pytorch_model.safetensors" --no_half_vae --train_data_dir "/content/zipOutput/" --max_train_epochs 10 --output_name "RFNAIFLux2V3" --output_dir "/content/drive/MyDrive/RFNai/Flux2V3/" --save_precision "bf16" --save_model_as "safetensors" --save_every_n_epochs 1 --save_last_n_epochs 10 --train_batch_size 8 --console_log_simple --noise_offset 0.05 --optimizer_type "AdamW8bit" --lr_scheduler "cosine" --loss_type "l2" --learning_rate 0.000035 --caption_dropout_rate 0.01 --caption_tag_dropout_rate 0.15 --max_grad_norm 1.0 --caption_extension ".txt" --max_data_loader_n_workers 4 --persistent_data_loader_workers --mixed_precision "bf16" --full_bf16 --gradient_accumulation_steps 2 --seed 43 --max_token_length 225 --resolution 1280 --optimizer_args weight_decay=0.001 eps=1e-8 betas=0.9,0.999 --enable_bucket --min_bucket_reso 256 --max_bucket_reso 2048 --bucket_reso_steps 32 --gradient_checkpointing --color_aug --random_crop --shuffle_caption --ip_noise_gamma 0.1 --flow_model --flow_use_ot --flow_timestep_distribution uniform --flow_uniform_static_ratio 2.5 --vae_type flux2 --latent_channels 32 --vae_custom_scale 0.6043 --vae_custom_shift 0.0760 --use_zero_cond=True --xformers --train_text_encoder --learning_rate_te1 0.00002 --learning_rate_te2 0.00002
+```
+  
+# Training Config:
+...
+
+# Training Code:
+[Bluvoll's Fork of sd-scripts](https://github.com/bluvoll/sd-scripts) 
+
+# Thanks:
+- [Anzhc](https://huggingface.co/Anzhc)
+- [Bluvoll](https://huggingface.co/Bluvoll)
+- [Cabal Research](https://huggingface.co/CabalResearch)
+
+
+