Could you share your training parameters, expert? My training results are always poor. The loss of my LoRA training stabilizes around 0.05, but after saving the seventh checkpoint, the gradient explodes, and the loss stops decreasing and starts increasing instead. Additionally, when the LoRA weight is set to 1.0, the output becomes blurry, and I have to adjust it to 0.7 for it to be usable. Below are my training parameters:

# Processing

resolution**: 1024

caption_extension**: ".txt"

shuffle_tags**: True

shuffle_caption**: True

activation_tags**: "1"

keep_tokens**: 1

# Steps

num_repeats**: 100

preferred_unit**: "Epochs"

how_many**: 10

max_train_epochs**: 10

max_train_steps**: None

save_every_n_epochs**: 1

keep_only_last_n_epochs**: 10

# Learning

unet_lr**: 5e-5

text_encoder_lr**: 6e-5

lr_scheduler**: "cosine_with_restarts"

lr_scheduler_number**: 3

lr_warmup_ratio**: 0.05

lr_warmup_steps**: 0

min_snr_gamma**: 8.0

multinoise**: True

# Structure

lora_type**: "LoRA"

network_dim**: 8

network_alpha**: 4

conv_dim**: 4

conv_alpha**: 1

network_module**: "networks.lora"

network_args**: ["conv_dim=4", "conv_alpha=1"]

# Training

train_batch_size**: 4

cross_attention**: "sdpa"

mixed_precision**: "bf16"

cache_latents**: True

cache_latents_to_drive**: True

cache_text_encoder_outputs**: False

# Advanced

optimizer**: "Prodigy"

optimizer_args**: ["decouple=True", "weight_decay=0.01", "betas=[0.9,0.999]", "d_coef=2", "use_bias_correction=True", "safeguard_warmup=True"]

recommended_values**: True