kaggle / config_file.toml
Alterneko's picture
Update config_file.toml
7590cd1
[model_arguments]
v2 = true
v_parameterization = true
pretrained_model_name_or_path = "/kaggle/working/pretrained_model/model.ckpt"
[optimizer_arguments]
min_snr_gamma = 5
optimizer_type = "AdamW8bit"
learning_rate = 2e-6
max_grad_norm = 1.0
train_text_encoder = false
lr_scheduler = "constant"
lr_warmup_steps = 0
[dataset_arguments]
enable_bucket = true
debug_dataset = false
in_json = "/kaggle/working/config_file/kaggle_meta.json"
train_data_dir = "/kaggle/working/train_data/content/fine_tune/matsuriv2_dataset/train_data"
dataset_repeats = 1
shuffle_caption = true
keep_tokens = 0
resolution = "768,768"
caption_dropout_rate = 0
caption_tag_dropout_rate = 0
caption_dropout_every_n_epochs = 0
color_aug = false
token_warmup_min = 1
token_warmup_step = 0
[training_arguments]
output_dir = "/kaggle/working/output"
output_name = "Matsuriv2-75k-steps"
save_precision = "fp16"
save_every_n_steps = 36385
save_state = false
train_batch_size = 1
max_token_length = 225
mem_eff_attn = false
xformers = true
max_train_steps = 36384
max_data_loader_n_workers = 8
persistent_data_loader_workers = true
gradient_checkpointing = false
gradient_accumulation_steps = 1
mixed_precision = "fp16"
logging_dir = "/kaggle/working/output/logs"
log_prefix = "Matsuriv2-75k-steps"
[sample_prompt_arguments]
sample_every_n_steps = 250000
sample_sampler = "ddim"
[saving_arguments]
save_model_as = "ckpt"