Patrick Bateman版本v1.0 (ID: 1301167)

Patrick Bateman from American Psycho, trained on 36 short clips for [512] resolution bucket, and [24, 68] frame buckets. 5 repeats for 40 epochs. Trained with diffusion-pipe.

Trained on 2x 3090s. If you are training on 1x 3090, you can remove pipeline_stages=2

config.toml:

pipeline_stages = 2

output_dir = "/workspace/outputs/patrick-bateman"

dataset = "/workspace/configs/patrick-bateman/dataset.toml"

epochs = 150

micro_batch_size_per_gpu = 1

gradient_accumulation_steps = 4

gradient_clipping = 1

warmup_steps = 10

eval_every_n_epochs = 1

eval_before_first_step = true

eval_micro_batch_size_per_gpu = 1

eval_gradient_accumulation_steps = 1

save_every_n_epochs = 10

checkpoint_every_n_minutes = 120

activation_checkpointing = true

partition_method = "parameters"

save_dtype = "bfloat16"

caching_batch_size = 1

steps_per_print = 1

video_clip_mode = "single_middle"



[model]

type = "hunyuan-video"

transformer_path = "/workspace/models/hunyuan_video_720_cfgdistill_fp8_e4m3fn.safetensors"

vae_path = "/workspace/models/hunyuan_video_vae_fp32.safetensors"

llm_path = "/workspace/models/llava-llama-3-8b-text-encoder-tokenizer"

clip_path = "/workspace/models/clip-vit-large-patch14"

dtype = "bfloat16"

transformer_dtype = "float8"

timestep_sample_method = "logit_normal"



[adapter]

type = "lora"

rank = 32

dtype = "bfloat16"

only_double_blocks = false



[optimizer]

type = "adamw"

lr = 0.0001

betas = [ 0.9, 0.99,]

weight_decay = 0.01

eps = 1e-8



[monitoring]

enable_wandb = false

dataset.toml:

resolutions = [ 512, ]

enable_ar_bucket = true

min_ar = 0.5

max_ar = 2

num_ar_buckets = 7

frame_buckets = [ 24, 68,]

[[directory]]

path = "/workspace/datasets/patrick-bateman"

num_repeats = 5

描述:

Patrick Bateman from American Psycho, trained on 36 short clips for [512] resolution bucket, and [24, 36] frame buckets. 5 repeats for 40 epochs. Trained with diffusion-pipe.

训练词语: patrick bateman,man in suit and tie,smiles and nods