riot

caption_prefix = '@riot.'

# trained using diffusion-pipe commit 6e95020cad0b3cd7dcb93ce42b358669051bf6d2

output_dir = '/home/txiaou/diffusion-pipe/output/riot_style'

dataset = '/home/txiaou/diffusion-pipe/examples/riot.toml'

eval_datasets = [

{name = 'eval_train', config = '/home/txiaou/diffusion-pipe/examples/riot_eval_train.toml'},

]

# training settings

epochs = 80

micro_batch_size_per_gpu = 1

pipeline_stages = 1

gradient_accumulation_steps = 4

gradient_clipping = 1

warmup_steps = 100

# eval settings

eval_every_n_epochs = 5

#eval_every_n_steps = 500

#eval_every_n_examples = 2048000

eval_before_first_step = true

eval_micro_batch_size_per_gpu = 1

eval_gradient_accumulation_steps = 1

# misc settings

save_every_n_epochs = 10

#save_every_n_steps = 1000

#save_every_n_examples = 4096000

#checkpoint_every_n_epochs = 1

#checkpoint_every_n_minutes = 120

activation_checkpointing = true

#reentrant_activation_checkpointing = true

partition_method = 'parameters'

# partition_method = 'manual'

# partition_split = [10]

save_dtype = 'bfloat16'

caching_batch_size = 1

map_num_proc = 8

steps_per_print = 1

compile = true

[model]

type = 'anima'

transformer_path = '/home/txiaou/diffusion-pipe/models/anima/anima-base-v1.0.safetensors'

vae_path = '/home/txiaou/diffusion-pipe/models/qwen/qwen_image_vae.safetensors'

llm_path = '/home/txiaou/diffusion-pipe/models/qwen/qwen_3_06b_base.safetensors'

dtype = 'bfloat16'

#cache_text_embeddings = false

llm_adapter_lr = 0

#timestep_sample_method = 'uniform'

#flux_shift = true

#multiscale_loss_weight = 0.5

sigmoid_scale = 1.3

[adapter]

type = 'lora'

rank = 32

dtype = 'bfloat16'

[optimizer]

type = 'adamw_optimi'

lr = 2e-5

betas = [0.9, 0.99]

weight_decay = 0.01

eps = 1e-8