stable-diffusion-finetune/configs/stable-diffusion/dev.yaml

132 lines
3.3 KiB
YAML
Raw Normal View History

2022-05-29 21:38:08 +02:00
model:
base_learning_rate: 1.0e-04
target: ldm.models.diffusion.ddpm.LatentDiffusion
params:
linear_start: 0.00085
linear_end: 0.0120
num_timesteps_cond: 1
log_every_t: 200
timesteps: 1000
first_stage_key: "jpg"
cond_stage_key: "txt"
image_size: 32
channels: 4
cond_stage_trainable: true
conditioning_key: crossattn
monitor: val/loss_simple_ema
scale_factor: 0.18215
scheduler_config: # 10000 warmup steps
target: ldm.lr_scheduler.LambdaLinearScheduler
params:
warm_up_steps: [ 10000 ]
cycle_lengths: [ 10000000000000 ] # incredibly large number to prevent corner cases
f_start: [ 1.e-6 ]
f_max: [ 1. ]
f_min: [ 1. ]
unet_config:
target: ldm.modules.diffusionmodules.openaimodel.UNetModel
params:
image_size: 32
in_channels: 4
out_channels: 4
2022-05-29 22:52:31 +02:00
model_channels: 32 # 320 # TODO increase
attention_resolutions: [ ] # is equal to fixed spatial resolution: 32 , 16 , 8
2022-05-29 21:38:08 +02:00
num_res_blocks: 2
2022-05-29 22:52:31 +02:00
channel_mult: [ 1, ]
2022-05-29 21:38:08 +02:00
#num_head_channels: 32
num_heads: 8
use_spatial_transformer: True
transformer_depth: 1
2022-05-29 22:52:31 +02:00
context_dim: 32
use_checkpoint: False
2022-05-29 21:38:08 +02:00
first_stage_config:
target: ldm.models.autoencoder.AutoencoderKL
params:
embed_dim: 4
monitor: val/rec_loss
2022-05-29 21:51:43 +02:00
ckpt_path: "models/first_stage_models/kl-f8/model.ckpt"
2022-05-29 21:38:08 +02:00
ddconfig:
double_z: true
z_channels: 4
resolution: 256
in_channels: 3
out_ch: 3
ch: 128
ch_mult:
- 1
- 2
- 4
- 4
num_res_blocks: 2
attn_resolutions: []
dropout: 0.0
lossconfig:
target: torch.nn.Identity
cond_stage_config:
target: ldm.modules.encoders.modules.BERTEmbedder
params:
2022-05-29 22:52:31 +02:00
n_embed: 32
n_layer: 1 #32 # TODO: increase
2022-05-29 21:38:08 +02:00
data:
target: ldm.data.laion.WebDataModuleFromConfig
params:
tar_base: "pipe:aws s3 cp s3://s-datasets/laion5b/laion2B-data/"
batch_size: 10
num_workers: 4
n_nodes: 1
train:
shards: '{000000..000010}.tar -' # TODO: wild guess, change
image_key: jpg
image_transforms:
- target: torchvision.transforms.Resize
params:
size: 256
interpolation: 3
- target: torchvision.transforms.RandomCrop
params:
size: 256
shuffle: 5000
n_examples: 16519100 # TODO: find out
validation:
shards: '{000011..000012}.tar -' # TODO: wild guess, change
image_key: jpg
image_transforms:
- target: torchvision.transforms.Resize
params:
size: 256
interpolation: 3
- target: torchvision.transforms.CenterCrop
params:
size: 256
shuffle: 0
n_examples: 60000 # TODO: find out
val_num_workers: 2
lightning:
callbacks:
image_logger:
target: main.ImageLogger
params:
batch_frequency: 5000 # 5000
max_images: 0
increase_log_steps: False
log_first_step: True
trainer:
replace_sampler_ddp: False
benchmark: True
val_check_interval: 20000 # every 20k training steps
num_sanity_val_steps: 0