VictorMylle/Kohya-ss-sd-scripts
1
0
Fork 1
mirror of https://github.com/kohya-ss/sd-scripts.git synced 2026-04-08 22:35:09 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add training metadata to output LoRA model

Browse Source
This commit is contained in:
space-nuko
2023-01-10 02:49:52 -08:00
parent a84ca297bd
commit 2e4ce0fdff
4 changed files with 72 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

45
train_network.py
View File

@@ -197,6 +197,47 @@ def train(args):
print(f" gradient ccumulation steps / 勾配を合計するステップ数 = {args.gradient_accumulation_steps}")
print(f" total optimization steps / 学習ステップ数: {args.max_train_steps}")
metadata = {
"ss_learning_rate": args.learning_rate,
"ss_text_encoder_lr": args.text_encoder_lr,
"ss_unet_lr": args.unet_lr,
"ss_num_train_images": train_dataset.num_train_images,
"ss_num_reg_images": train_dataset.num_reg_images,
"ss_num_batches_per_epoch": len(train_dataloader),
"ss_num_epochs": num_train_epochs,
"ss_batch_size_per_device": args.train_batch_size,
"ss_total_batch_size": total_batch_size,
"ss_gradient_accumulation_steps": args.gradient_accumulation_steps,
"ss_max_train_steps": args.max_train_steps,
"ss_lr_warmup_steps": args.lr_warmup_steps,
"ss_lr_scheduler": args.lr_scheduler,
"ss_network_module": args.network_module,
"ss_network_dim": 4 if args.network_dim is None else args.network_dim,
"ss_full_fp16": bool(args.full_fp16),
"ss_v2": bool(args.v2),
"ss_resolution": args.resolution,
"ss_clip_skip": args.clip_skip,
"ss_max_token_length": args.max_token_length,
"ss_color_aug": bool(args.color_aug),
"ss_flip_aug": bool(args.flip_aug),
"ss_random_crop": bool(args.random_crop),
"ss_shuffle_caption": bool(args.shuffle_caption),
"ss_cache_latents": bool(args.cache_latents),
"ss_enable_bucket": bool(args.enable_bucket),
"ss_min_bucket_reso": args.min_bucket_reso,
"ss_max_bucket_reso": args.max_bucket_reso,
"ss_seed": args.seed
}
if args.pretrained_model_name_or_path is not None:
sd_model_name = args.pretrained_model_name_or_path
if os.path.exists(sd_model_name):
metadata["ss_sd_model_hash"] = train_util.model_hash(sd_model_name)
sd_model_name = os.path.basename(sd_model_name)
metadata["ss_sd_model_name"] = sd_model_name
metadata = {k: str(v) for k, v in metadata.items()}
progress_bar = tqdm(range(args.max_train_steps), smoothing=0, disable=not accelerator.is_local_main_process, desc="steps")
global_step = 0
@@ -296,7 +337,7 @@ def train(args):
ckpt_name = train_util.EPOCH_FILE_NAME.format(model_name, epoch + 1) + '.' + args.save_model_as
ckpt_file = os.path.join(args.output_dir, ckpt_name)
print(f"saving checkpoint: {ckpt_file}")
unwrap_model(network).save_weights(ckpt_file, save_dtype)
unwrap_model(network).save_weights(ckpt_file, save_dtype, metadata)
def remove_old_func(old_epoch_no):
old_ckpt_name = train_util.EPOCH_FILE_NAME.format(model_name, old_epoch_no) + '.' + args.save_model_as
@@ -330,7 +371,7 @@ def train(args):
ckpt_file = os.path.join(args.output_dir, ckpt_name)
print(f"save trained model to {ckpt_file}")
network.save_weights(ckpt_file, save_dtype)
network.save_weights(ckpt_file, save_dtype, metadata)
print("model saved.")