Upload 9 files
Browse files- raw/generator-348160.safetensors +3 -0
- raw/sotediffusion-wr3_3b-step00166912.safetensors +3 -0
- raw/sotediffusion-wr3_3b-step00166912_text_model.safetensors +3 -0
- raw/stage_b.yaml +28 -0
- raw/train-sc.sh +52 -0
- sotediffusion-v2-stage_b.safetensors +3 -0
- sotediffusion-v2-stage_c.safetensors +3 -0
- sotediffusion-v2-stage_c_text_encoder.safetensors +3 -0
- stage_a_ft_hq.safetensors +3 -0
raw/generator-348160.safetensors
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:d1b95b1de4cd562319e4757992b77012c13f663b421cdc94b88cccbd1d837d1f
|
3 |
+
size 6251923744
|
raw/sotediffusion-wr3_3b-step00166912.safetensors
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:9abdd2c046e867ed03d71d13daee65ebac8f7aa502748598e629fbf03bdbab2d
|
3 |
+
size 14356558016
|
raw/sotediffusion-wr3_3b-step00166912_text_model.safetensors
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:e018a6ec07a008b05e40e4bfc4eb316bf91341753572f968316f5f57f9345da1
|
3 |
+
size 2778702232
|
raw/stage_b.yaml
ADDED
@@ -0,0 +1,28 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
experiment_id: sotediffusion-wr3_3b-stage_b
|
2 |
+
model_version: 3B
|
3 |
+
dtype: float32
|
4 |
+
use_fsdp: False
|
5 |
+
|
6 |
+
batch_size: 8
|
7 |
+
grad_accum_steps: 4
|
8 |
+
updates: 1024000
|
9 |
+
backup_every: 2048
|
10 |
+
save_every: 1024
|
11 |
+
warmup_updates: 128
|
12 |
+
|
13 |
+
lr: 8.0e-6
|
14 |
+
optimizer_type: AdamW
|
15 |
+
adaptive_loss_weight: False
|
16 |
+
stochastic_rounding: False
|
17 |
+
|
18 |
+
image_size: 1024
|
19 |
+
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]
|
20 |
+
shift: 4
|
21 |
+
|
22 |
+
checkpoint_path: /home/ubuntu/out/
|
23 |
+
output_path: /home/ubuntu/out/
|
24 |
+
webdataset_path: file:/home/ubuntu/newest_best.tar
|
25 |
+
|
26 |
+
effnet_checkpoint_path: /home/ubuntu/models/wuerstchen3/effnet_encoder.safetensors
|
27 |
+
stage_a_checkpoint_path: /home/ubuntu/models/wuerstchen3/stage_a.safetensors
|
28 |
+
generator_checkpoint_path: /home/ubuntu/models/wuerstchen3/generator-020480.safetensors
|
raw/train-sc.sh
ADDED
@@ -0,0 +1,52 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#!/bin/sh
|
2 |
+
|
3 |
+
OUT_PATH="/home/ubuntu/out"
|
4 |
+
DATASET_PATH="/home/ubuntu/dataset"
|
5 |
+
PROMPT_PATH="/home/ubuntu/sotediffusion-prompt.txt"
|
6 |
+
WUER_PATH="/home/ubuntu/models/wuerstchen3"
|
7 |
+
|
8 |
+
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 \
|
9 |
+
--mixed_precision no \
|
10 |
+
--save_precision float \
|
11 |
+
--sdpa \
|
12 |
+
--gradient_checkpointing \
|
13 |
+
--train_text_encoder \
|
14 |
+
--resolution "1024,1024" \
|
15 |
+
--train_batch_size 12 \
|
16 |
+
--learning_rate 2e-6 \
|
17 |
+
--learning_rate_te1 1e-7 \
|
18 |
+
--lr_scheduler constant_with_warmup \
|
19 |
+
--lr_warmup_steps 100 \
|
20 |
+
--optimizer_type adamw8bit \
|
21 |
+
--token_warmup_min 1 \
|
22 |
+
--token_warmup_step 0 \
|
23 |
+
--caption_separator ", " \
|
24 |
+
--caption_dropout_rate 0.1 \
|
25 |
+
--caption_tag_dropout_rate 0 \
|
26 |
+
--caption_dropout_every_n_epochs 0 \
|
27 |
+
--dataset_repeats 1 \
|
28 |
+
--save_state \
|
29 |
+
--save_every_n_steps 1024 \
|
30 |
+
--sample_every_n_steps 1024 \
|
31 |
+
--max_token_length 225 \
|
32 |
+
--max_grad_norm 0 \
|
33 |
+
--max_train_epochs 10 \
|
34 |
+
--caption_extension ".txt" \
|
35 |
+
--max_data_loader_n_workers 32 \
|
36 |
+
--persistent_data_loader_workers \
|
37 |
+
--enable_bucket \
|
38 |
+
--min_bucket_reso 256 \
|
39 |
+
--max_bucket_reso 4096 \
|
40 |
+
--bucket_reso_steps 128 \
|
41 |
+
--bucket_no_upscale \
|
42 |
+
--log_with wandb \
|
43 |
+
--output_name sotediffusion-wr3_3b \
|
44 |
+
--train_data_dir $DATASET_PATH \
|
45 |
+
--in_json $DATASET_PATH.json \
|
46 |
+
--output_dir $OUT_PATH \
|
47 |
+
--logging_dir $OUT_PATH/logs \
|
48 |
+
--stage_c_checkpoint_path $WUER_PATH/sotediffusion-wr3_3b-base.safetensors \
|
49 |
+
--text_model_checkpoint_path $WUER_PATH/sotediffusion-wr3_text_model.safetensors \
|
50 |
+
--effnet_checkpoint_path $WUER_PATH/effnet_encoder.safetensors \
|
51 |
+
--previewer_checkpoint_path $WUER_PATH/previewer.safetensors \
|
52 |
+
--sample_prompts $PROMPT_PATH
|
sotediffusion-v2-stage_b.safetensors
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:02139c3eaff76b043c59ec7485a93ebee23ee96d4e38ff2b9ca063e20808c2b1
|
3 |
+
size 3126040976
|
sotediffusion-v2-stage_c.safetensors
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:b4f582137f9f2562f8f378140cf01ded3aa81aedbde13e2b4bf9367e9fba23e5
|
3 |
+
size 7178349464
|
sotediffusion-v2-stage_c_text_encoder.safetensors
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:c6d44dcdb17269476725c5cfdcbfb220a9048050594385f31a479430d727bcc0
|
3 |
+
size 1389382176
|
stage_a_ft_hq.safetensors
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:ecf8591d8f6d06ec27d299f28331e6f5832f29092ca3e0c1a58a7f04eb08bedd
|
3 |
+
size 73639536
|