|
|
|
|
|
|
|
|
|
name: 6931_linguini_ortho |
|
manual_seed: 6931 |
|
mixed_precision: fp16 |
|
gradient_accumulation_steps: 1 |
|
datasets: |
|
train: |
|
name: LoraDataset |
|
concept_list: single-concept/data_configs/linguini.json |
|
use_caption: true |
|
use_mask: true |
|
instance_transform: |
|
- type: HumanResizeCropFinalV3 |
|
size: 512 |
|
crop_p: 0.5 |
|
- type: ToTensor |
|
- type: Normalize |
|
mean: |
|
- 0.5 |
|
std: |
|
- 0.5 |
|
- type: ShuffleCaption |
|
keep_token_num: 1 |
|
- type: EnhanceText |
|
enhance_type: human |
|
replace_mapping: |
|
<TOK>: <linguini1> <linguini2> |
|
batch_size_per_gpu: 2 |
|
dataset_enlarge_ratio: 500 |
|
val_vis: |
|
name: PromptDataset |
|
prompts: single-concept/validation_prompts/characters/test_man.txt |
|
num_samples_per_prompt: 8 |
|
latent_size: |
|
- 4 |
|
- 64 |
|
- 64 |
|
replace_mapping: |
|
<TOK>: <linguini1> <linguini2> |
|
batch_size_per_gpu: 4 |
|
models: |
|
pretrained_path: nitrosocke/mo-di-diffusion |
|
enable_edlora: true |
|
finetune_cfg: |
|
text_embedding: |
|
enable_tuning: true |
|
lr: 0.001 |
|
text_encoder: |
|
enable_tuning: true |
|
lora_cfg: |
|
rank: 5 |
|
alpha: 1.0 |
|
where: CLIPAttention |
|
lr: 1.0e-05 |
|
unet: |
|
enable_tuning: true |
|
lora_cfg: |
|
rank: 5 |
|
alpha: 1.0 |
|
where: Attention |
|
lr: 0.0001 |
|
new_concept_token: <linguini1>+<linguini2> |
|
initializer_token: <rand-0.013>+man |
|
noise_offset: 0.01 |
|
attn_reg_weight: 0.01 |
|
reg_full_identity: false |
|
use_mask_loss: true |
|
gradient_checkpoint: false |
|
enable_xformers: true |
|
path: |
|
pretrain_network: null |
|
train: |
|
optim_g: |
|
type: AdamW |
|
lr: 0.0 |
|
weight_decay: 0.01 |
|
betas: |
|
- 0.9 |
|
- 0.999 |
|
unet_kv_drop_rate: 0 |
|
scheduler: linear |
|
emb_norm_threshold: 0.55 |
|
val: |
|
val_during_save: false |
|
compose_visualize: false |
|
alpha_list: |
|
- 0 |
|
- 0.7 |
|
- 1.0 |
|
sample: |
|
num_inference_steps: 50 |
|
guidance_scale: 7.5 |
|
logger: |
|
print_freq: 10 |
|
save_checkpoint_freq: 10000.0 |
|
|
