Upload 5 files

generator-348160.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d1b95b1de4cd562319e4757992b77012c13f663b421cdc94b88cccbd1d837d1f
+size 6251923744

sotediffusion-wr3_3b-step00166912.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9abdd2c046e867ed03d71d13daee65ebac8f7aa502748598e629fbf03bdbab2d
+size 14356558016

sotediffusion-wr3_3b-step00166912_text_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e018a6ec07a008b05e40e4bfc4eb316bf91341753572f968316f5f57f9345da1
+size 2778702232

stage_b.yaml

@@ -0,0 +1,28 @@
+experiment_id: sotediffusion-wr3_3b-stage_b
+model_version: 3B
+dtype: float32
+use_fsdp: False
+
+batch_size: 8
+grad_accum_steps: 4
+updates: 1024000
+backup_every: 2048
+save_every: 1024
+warmup_updates: 128
+
+lr: 8.0e-6
+optimizer_type: AdamW
+adaptive_loss_weight: False
+stochastic_rounding: False
+
+image_size: 1024
+multi_aspect_ratio: [1/1, 1/2, 1/3, 2/3, 3/4, 1/5, 2/5, 3/5, 4/5, 1/6, 5/6, 9/16]
+shift: 4
+
+checkpoint_path:  /home/ubuntu/out/
+output_path: /home/ubuntu/out/
+webdataset_path: file:/home/ubuntu/newest_best.tar
+
+effnet_checkpoint_path: /home/ubuntu/models/wuerstchen3/effnet_encoder.safetensors
+stage_a_checkpoint_path: /home/ubuntu/models/wuerstchen3/stage_a.safetensors
+generator_checkpoint_path: /home/ubuntu/models/wuerstchen3/generator-020480.safetensors

train-sc.sh

@@ -0,0 +1,52 @@
+#!/bin/sh
+
+OUT_PATH="/home/ubuntu/out"
+DATASET_PATH="/home/ubuntu/dataset"
+PROMPT_PATH="/home/ubuntu/sotediffusion-prompt.txt"
+WUER_PATH="/home/ubuntu/models/wuerstchen3"
+
+CUDA_VISIBLE_DEVICES="1,2,3,4,5,6,7" accelerate launch --mixed_precision no --multi_gpu /home/ubuntu/sd-scripts/stable_cascade_train_stage_c.py \
+--mixed_precision no \
+--save_precision float \
+--sdpa \
+--gradient_checkpointing \
+--train_text_encoder \
+--resolution "1024,1024" \
+--train_batch_size 12 \
+--learning_rate 2e-6 \
+--learning_rate_te1 1e-7 \
+--lr_scheduler constant_with_warmup \
+--lr_warmup_steps 100 \
+--optimizer_type adamw8bit \
+--token_warmup_min 1 \
+--token_warmup_step 0 \
+--caption_separator ", " \
+--caption_dropout_rate 0.1 \
+--caption_tag_dropout_rate 0 \
+--caption_dropout_every_n_epochs 0 \
+--dataset_repeats 1 \
+--save_state \
+--save_every_n_steps 1024 \
+--sample_every_n_steps 1024 \
+--max_token_length 225 \
+--max_grad_norm 0 \
+--max_train_epochs 10 \
+--caption_extension ".txt" \
+--max_data_loader_n_workers 32 \
+--persistent_data_loader_workers \
+--enable_bucket \
+--min_bucket_reso 256 \
+--max_bucket_reso 4096 \
+--bucket_reso_steps 128 \
+--bucket_no_upscale \
+--log_with wandb \
+--output_name sotediffusion-wr3_3b \
+--train_data_dir $DATASET_PATH \
+--in_json $DATASET_PATH.json \
+--output_dir $OUT_PATH \
+--logging_dir $OUT_PATH/logs \
+--stage_c_checkpoint_path $WUER_PATH/sotediffusion-wr3_3b-base.safetensors \
+--text_model_checkpoint_path $WUER_PATH/sotediffusion-wr3_text_model.safetensors \
+--effnet_checkpoint_path $WUER_PATH/effnet_encoder.safetensors \
+--previewer_checkpoint_path $WUER_PATH/previewer.safetensors \
+--sample_prompts $PROMPT_PATH