stable-diffusion-finetune/configs/stable-diffusion/txt2img-multinode-clip-encoder-f16-256-pretraining.yaml
2022-06-10 11:08:42 +02:00

131 lines
3.6 KiB
YAML

model:
base_learning_rate: 1.0e-04
target: ldm.models.diffusion.ddpm.LatentDiffusion
params:
linear_start: 0.001
linear_end: 0.015
num_timesteps_cond: 1
log_every_t: 200
timesteps: 1000
first_stage_key: "jpg"
cond_stage_key: "txt"
image_size: 16
channels: 16
cond_stage_trainable: false # Note: different from the one we trained before
conditioning_key: crossattn
monitor: val/loss_simple_ema
scale_factor: 0.22765929 # magic number
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: 16 # not really needed
in_channels: 16
out_channels: 16
model_channels: 320 # TODO: scale model here
attention_resolutions: [ 4, 2, 1 ]
num_res_blocks: 2
channel_mult: [ 1, 2, 4, 4 ]
num_heads: 8
use_spatial_transformer: True
transformer_depth: 1
context_dim: 768
use_checkpoint: True
legacy: False
first_stage_config:
target: ldm.models.autoencoder.AutoencoderKL
params:
embed_dim: 4
monitor: val/rec_loss
ckpt_path: "models/first_stage_models/kl-f16/model.ckpt"
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.FrozenCLIPEmbedder
data:
target: ldm.data.laion.WebDataModuleFromConfig
params:
tar_base: "pipe:aws s3 cp s3://s-datasets/laion5b/laion2B-data/"
batch_size: 50 # TODO: max out
num_workers: 4
multinode: True
min_size: 256 # TODO: experiment. Note: for 2B, images are stored at max 384 resolution
train:
shards: '{000000..231317}.tar -'
shuffle: 10000
image_key: jpg
image_transforms:
- target: torchvision.transforms.Resize
params:
size: 256
interpolation: 3
- target: torchvision.transforms.RandomCrop
params:
size: 256
# NOTE use enough shards to avoid empty validation loops in workers
validation:
shards: '{231318..231349}.tar -'
shuffle: 0
image_key: jpg
image_transforms:
- target: torchvision.transforms.Resize
params:
size: 256
interpolation: 3
- target: torchvision.transforms.CenterCrop
params:
size: 256
lightning:
callbacks:
image_logger:
target: main.ImageLogger
params:
batch_frequency: 5000
max_images: 4
increase_log_steps: False
log_first_step: False
log_images_kwargs:
use_ema_scope: False
inpaint: False
plot_progressive_rows: False
plot_diffusion_rows: False
N: 4
unconditional_guidance_scale: 3.0
unconditional_guidance_label: [""]
trainer:
benchmark: True
val_check_interval: 5000000 # really sorry
num_sanity_val_steps: 0
accumulate_grad_batches: 2 # TODO: want accumulate on? --> wait for final batch-size