Spaces:
Running
on
Zero
Running
on
Zero
Upload 12 files
Browse files- configs/inference/controlnet_c_3b_canny.yaml +14 -0
- configs/inference/controlnet_c_3b_identity.yaml +17 -0
- configs/inference/controlnet_c_3b_inpainting.yaml +15 -0
- configs/inference/controlnet_c_3b_sr.yaml +15 -0
- configs/inference/lora_c_3b.yaml +15 -0
- configs/inference/stage_b_1b.yaml +13 -0
- configs/inference/stage_b_3b.yaml +13 -0
- configs/inference/stage_c_1b.yaml +7 -0
- configs/inference/stage_c_3b.yaml +7 -0
- configs/training/cfg_control_lr.yaml +48 -0
- configs/training/lora_personalization.yaml +38 -0
- configs/training/t2i.yaml +29 -0
configs/inference/controlnet_c_3b_canny.yaml
ADDED
|
@@ -0,0 +1,14 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# GLOBAL STUFF
|
| 2 |
+
model_version: 3.6B
|
| 3 |
+
dtype: bfloat16
|
| 4 |
+
|
| 5 |
+
# ControlNet specific
|
| 6 |
+
controlnet_blocks: [0, 4, 8, 12, 51, 55, 59, 63]
|
| 7 |
+
controlnet_filter: CannyFilter
|
| 8 |
+
controlnet_filter_params:
|
| 9 |
+
resize: 224
|
| 10 |
+
|
| 11 |
+
effnet_checkpoint_path: models/effnet_encoder.safetensors
|
| 12 |
+
previewer_checkpoint_path: models/previewer.safetensors
|
| 13 |
+
generator_checkpoint_path: models/stage_c_bf16.safetensors
|
| 14 |
+
controlnet_checkpoint_path: models/canny.safetensors
|
configs/inference/controlnet_c_3b_identity.yaml
ADDED
|
@@ -0,0 +1,17 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# GLOBAL STUFF
|
| 2 |
+
model_version: 3.6B
|
| 3 |
+
dtype: bfloat16
|
| 4 |
+
|
| 5 |
+
# ControlNet specific
|
| 6 |
+
controlnet_bottleneck_mode: 'simple'
|
| 7 |
+
controlnet_blocks: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63]
|
| 8 |
+
controlnet_filter: IdentityFilter
|
| 9 |
+
controlnet_filter_params:
|
| 10 |
+
max_faces: 4
|
| 11 |
+
p_drop: 0.00
|
| 12 |
+
p_full: 0.0
|
| 13 |
+
|
| 14 |
+
effnet_checkpoint_path: models/effnet_encoder.safetensors
|
| 15 |
+
previewer_checkpoint_path: models/previewer.safetensors
|
| 16 |
+
generator_checkpoint_path: models/stage_c_bf16.safetensors
|
| 17 |
+
controlnet_checkpoint_path:
|
configs/inference/controlnet_c_3b_inpainting.yaml
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# GLOBAL STUFF
|
| 2 |
+
model_version: 3.6B
|
| 3 |
+
dtype: bfloat16
|
| 4 |
+
|
| 5 |
+
# ControlNet specific
|
| 6 |
+
controlnet_blocks: [0, 4, 8, 12, 51, 55, 59, 63]
|
| 7 |
+
controlnet_filter: InpaintFilter
|
| 8 |
+
controlnet_filter_params:
|
| 9 |
+
thresold: [0.04, 0.4]
|
| 10 |
+
p_outpaint: 0.4
|
| 11 |
+
|
| 12 |
+
effnet_checkpoint_path: models/effnet_encoder.safetensors
|
| 13 |
+
previewer_checkpoint_path: models/previewer.safetensors
|
| 14 |
+
generator_checkpoint_path: models/stage_c_bf16.safetensors
|
| 15 |
+
controlnet_checkpoint_path: models/inpainting.safetensors
|
configs/inference/controlnet_c_3b_sr.yaml
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# GLOBAL STUFF
|
| 2 |
+
model_version: 3.6B
|
| 3 |
+
dtype: bfloat16
|
| 4 |
+
|
| 5 |
+
# ControlNet specific
|
| 6 |
+
controlnet_bottleneck_mode: 'large'
|
| 7 |
+
controlnet_blocks: [0, 4, 8, 12, 51, 55, 59, 63]
|
| 8 |
+
controlnet_filter: SREffnetFilter
|
| 9 |
+
controlnet_filter_params:
|
| 10 |
+
scale_factor: 0.5
|
| 11 |
+
|
| 12 |
+
effnet_checkpoint_path: models/effnet_encoder.safetensors
|
| 13 |
+
previewer_checkpoint_path: models/previewer.safetensors
|
| 14 |
+
generator_checkpoint_path: models/stage_c_bf16.safetensors
|
| 15 |
+
controlnet_checkpoint_path: models/super_resolution.safetensors
|
configs/inference/lora_c_3b.yaml
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# GLOBAL STUFF
|
| 2 |
+
model_version: 3.6B
|
| 3 |
+
dtype: bfloat16
|
| 4 |
+
|
| 5 |
+
# LoRA specific
|
| 6 |
+
module_filters: ['.attn']
|
| 7 |
+
rank: 4
|
| 8 |
+
train_tokens:
|
| 9 |
+
# - ['^snail', null] # token starts with "snail" -> "snail" & "snails", don't need to be reinitialized
|
| 10 |
+
- ['[fernando]', '^dog</w>'] # custom token [snail], initialize as avg of snail & snails
|
| 11 |
+
|
| 12 |
+
effnet_checkpoint_path: models/effnet_encoder.safetensors
|
| 13 |
+
previewer_checkpoint_path: models/previewer.safetensors
|
| 14 |
+
generator_checkpoint_path: models/stage_c_bf16.safetensors
|
| 15 |
+
lora_checkpoint_path: models/lora_fernando_10k.safetensors
|
configs/inference/stage_b_1b.yaml
ADDED
|
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# GLOBAL STUFF
|
| 2 |
+
model_version: 700M
|
| 3 |
+
dtype: bfloat16
|
| 4 |
+
|
| 5 |
+
# For demonstration purposes in reconstruct_images.ipynb
|
| 6 |
+
webdataset_path: path to your dataset
|
| 7 |
+
batch_size: 1
|
| 8 |
+
image_size: 2048
|
| 9 |
+
grad_accum_steps: 1
|
| 10 |
+
|
| 11 |
+
effnet_checkpoint_path: models/effnet_encoder.safetensors
|
| 12 |
+
stage_a_checkpoint_path: models/stage_a.safetensors
|
| 13 |
+
generator_checkpoint_path: models/stage_b_lite_bf16.safetensors
|
configs/inference/stage_b_3b.yaml
ADDED
|
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# GLOBAL STUFF
|
| 2 |
+
model_version: 3B
|
| 3 |
+
dtype: bfloat16
|
| 4 |
+
|
| 5 |
+
# For demonstration purposes in reconstruct_images.ipynb
|
| 6 |
+
webdataset_path: path to your dataset
|
| 7 |
+
batch_size: 4
|
| 8 |
+
image_size: 1024
|
| 9 |
+
grad_accum_steps: 1
|
| 10 |
+
|
| 11 |
+
effnet_checkpoint_path: path to effnet of stablecascade / effnet_encoder.safetensors
|
| 12 |
+
stage_a_checkpoint_path: path to effnet of stablecascade stage a decoder/stage_a.safetensors
|
| 13 |
+
generator_checkpoint_path: path to effnet of stablecascade stage b decoer heavy version bf16/stage_b_lite_bf16.safetensors
|
configs/inference/stage_c_1b.yaml
ADDED
|
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# GLOBAL STUFF
|
| 2 |
+
model_version: 1B
|
| 3 |
+
dtype: bfloat16
|
| 4 |
+
|
| 5 |
+
effnet_checkpoint_path: path to effnet of stablecascade / effnet_encoder.safetensors
|
| 6 |
+
previewer_checkpoint_path: path to previewer of stablecascade/previewer.safetensors
|
| 7 |
+
generator_checkpoint_path: path to generator of stablecascade stage c lite version bf16 /stage_c_lite_bf16.safetensors
|
configs/inference/stage_c_3b.yaml
ADDED
|
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# GLOBAL STUFF
|
| 2 |
+
model_version: 3.6B
|
| 3 |
+
dtype: bfloat16
|
| 4 |
+
|
| 5 |
+
effnet_checkpoint_path: models/effnet_encoder.safetensors
|
| 6 |
+
previewer_checkpoint_path: models/previewer.safetensors
|
| 7 |
+
generator_checkpoint_path: models/stage_c_bf16.safetensors
|
configs/training/cfg_control_lr.yaml
ADDED
|
@@ -0,0 +1,48 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# GLOBAL STUFF
|
| 2 |
+
experiment_id: Ultrapixel_controlnet
|
| 3 |
+
|
| 4 |
+
checkpoint_path: checkpoint output path
|
| 5 |
+
output_path: visual results output path
|
| 6 |
+
model_version: 3.6B
|
| 7 |
+
dtype: float32
|
| 8 |
+
# # WandB
|
| 9 |
+
# wandb_project: StableCascade
|
| 10 |
+
# wandb_entity: wandb_username
|
| 11 |
+
#module_filters: ['.depthwise', '.mapper', '.attn', '.channelwise' ]
|
| 12 |
+
#rank: 32
|
| 13 |
+
# TRAINING PARAMS
|
| 14 |
+
lr: 1.0e-4
|
| 15 |
+
batch_size: 12
|
| 16 |
+
#image_size: [1536, 2048, 2560, 3072, 4096]
|
| 17 |
+
image_size: [1024, 2048, 2560, 3072, 3584, 3840, 4096, 4608]
|
| 18 |
+
#image_size: [ 1024, 1536, 2048, 2560, 3072, 3584, 3840, 4096, 4608]
|
| 19 |
+
#image_size: [ 1024, 1280]
|
| 20 |
+
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]
|
| 21 |
+
grad_accum_steps: 2
|
| 22 |
+
updates: 40000
|
| 23 |
+
backup_every: 5000
|
| 24 |
+
save_every: 256
|
| 25 |
+
warmup_updates: 1
|
| 26 |
+
use_fsdp: True
|
| 27 |
+
|
| 28 |
+
# ControlNet specific
|
| 29 |
+
controlnet_blocks: [0, 4, 8, 12, 51, 55, 59, 63]
|
| 30 |
+
controlnet_filter: CannyFilter
|
| 31 |
+
controlnet_filter_params:
|
| 32 |
+
resize: 224
|
| 33 |
+
# offset_noise: 0.1
|
| 34 |
+
|
| 35 |
+
# GDF
|
| 36 |
+
adaptive_loss_weight: True
|
| 37 |
+
|
| 38 |
+
ema_start_iters: 10
|
| 39 |
+
ema_iters: 50
|
| 40 |
+
ema_beta: 0.9
|
| 41 |
+
|
| 42 |
+
webdataset_path: path to your training dataset
|
| 43 |
+
effnet_checkpoint_path: models/effnet_encoder.safetensors
|
| 44 |
+
previewer_checkpoint_path: models/previewer.safetensors
|
| 45 |
+
generator_checkpoint_path: models/stage_c_bf16.safetensors
|
| 46 |
+
controlnet_checkpoint_path: models/canny.safetensors
|
| 47 |
+
|
| 48 |
+
|
configs/training/lora_personalization.yaml
ADDED
|
@@ -0,0 +1,38 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# GLOBAL STUFF
|
| 2 |
+
experiment_id: roubao_cat_personalized
|
| 3 |
+
|
| 4 |
+
checkpoint_path: checkpoint output path
|
| 5 |
+
output_path: visual results output path
|
| 6 |
+
model_version: 3.6B
|
| 7 |
+
dtype: float32
|
| 8 |
+
|
| 9 |
+
module_filters: [ '.attn']
|
| 10 |
+
rank: 4
|
| 11 |
+
train_tokens:
|
| 12 |
+
# - ['^snail', null] # token starts with "snail" -> "snail" & "snails", don't need to be reinitialized
|
| 13 |
+
- ['[roubaobao]', '^cat</w>'] # custom token [snail], initialize as avg of snail & snails
|
| 14 |
+
# TRAINING PARAMS
|
| 15 |
+
lr: 1.0e-4
|
| 16 |
+
batch_size: 4
|
| 17 |
+
|
| 18 |
+
image_size: [1024, 2048, 2560, 3072, 3584, 3840, 4096, 4608]
|
| 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 |
+
grad_accum_steps: 2
|
| 21 |
+
updates: 40000
|
| 22 |
+
backup_every: 5000
|
| 23 |
+
save_every: 512
|
| 24 |
+
warmup_updates: 1
|
| 25 |
+
use_ddp: True
|
| 26 |
+
|
| 27 |
+
# GDF
|
| 28 |
+
adaptive_loss_weight: True
|
| 29 |
+
|
| 30 |
+
|
| 31 |
+
tmp_prompt: a photo of a cat [roubaobao]
|
| 32 |
+
webdataset_path: path to your personalized training dataset
|
| 33 |
+
effnet_checkpoint_path: models/effnet_encoder.safetensors
|
| 34 |
+
previewer_checkpoint_path: models/previewer.safetensors
|
| 35 |
+
generator_checkpoint_path: models/stage_c_bf16.safetensors
|
| 36 |
+
ultrapixel_path: models/ultrapixel_t2i.safetensors
|
| 37 |
+
|
| 38 |
+
|
configs/training/t2i.yaml
ADDED
|
@@ -0,0 +1,29 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# GLOBAL STUFF
|
| 2 |
+
experiment_id: ultrapixel_t2i
|
| 3 |
+
#strc_fixlrt_norm3_lite_1024_hrft_newdata
|
| 4 |
+
checkpoint_path: checkpoint output path #output model directory
|
| 5 |
+
output_path: visual results output path #experiment output directory
|
| 6 |
+
model_version: 3.6B # finetune large stage c model of stablecascade
|
| 7 |
+
dtype: float32
|
| 8 |
+
|
| 9 |
+
|
| 10 |
+
# TRAINING PARAMS
|
| 11 |
+
lr: 1.0e-4
|
| 12 |
+
batch_size: 4 # gpu_number * num_per_gpu * grad_accum_steps
|
| 13 |
+
image_size: [1024, 2048, 2560, 3072, 3584, 3840, 4096, 4608] # possible image resolution
|
| 14 |
+
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]
|
| 15 |
+
grad_accum_steps: 2
|
| 16 |
+
updates: 40000
|
| 17 |
+
backup_every: 5000
|
| 18 |
+
save_every: 256
|
| 19 |
+
warmup_updates: 1
|
| 20 |
+
use_ddp: True
|
| 21 |
+
|
| 22 |
+
# GDF
|
| 23 |
+
adaptive_loss_weight: True
|
| 24 |
+
|
| 25 |
+
|
| 26 |
+
webdataset_path: path to your personalized training dataset
|
| 27 |
+
effnet_checkpoint_path: models/effnet_encoder.safetensors
|
| 28 |
+
previewer_checkpoint_path: models/previewer.safetensors
|
| 29 |
+
generator_checkpoint_path: models/stage_c_bf16.safetensors
|