File size: 1,345 Bytes
572676d |
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 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 |
model:
checkpoint_path: "./models/aura_flow_0.3.bnb_nf4.safetensors"
pretrained_model_name_or_path: fal/AuraFlow-v0.3
dtype: bfloat16
denoiser:
use_flash_attn: true
use_rope: True
rope_theta: 10000
rope_dim_sizes: [32, 112, 112]
peft:
type: lora
rank: 8
alpha: 1.0
dropout: 0.0
dtype: bfloat16
include_keys: [".attn."]
exclude_keys: ["text_encoder", "vae", "t_embedder", "final_linear"]
dataset:
folder: "data/pexels-1k-random"
num_repeats: 2
batch_size: 2
bucket_base_size: 1024
step: 128
min_size: 384
do_upscale: false
caption_processors: []
optimizer:
name: "schedulefree.RAdamScheduleFree"
# name: "bitsandbytes.optim.AdamW8bit"
args:
lr: 0.005
scheduler:
# name: "torch.optim.lr_scheduler.ConstantLR"
# args: {}
tracker:
project_name: "auraflow-rope-1"
loggers:
- wandb
saving:
strategy:
per_epochs: 1
per_steps: null
save_last: true
callbacks:
- type: "hf_hub" # or "hf_hub" to push to hub
name: "rope-1"
save_dir: "./output/rope-1"
hub_id: "p1atdev/afv03-lora"
dir_in_repo: "rope-1"
seed: 42
num_train_epochs: 10
trainer:
# debug_mode: "1step"
gradient_checkpointing: true
torch_compile: true
torch_compile_args:
mode: max-autotune
fullgraph: true
fp32_matmul_precision: "medium"
|