manly20250529--01khmer001

Here's a human-readable version of the configuration, organized into clear sections with explanations:

Network Configuration

• U-Net Learning Rate: 0.0005
  (Controls how fast the U-Net model learns during training.)

• Text Encoder Learning Rate: 0.00005
  (Sets the learning rate for the text encoder, which processes text prompts.)

• Network Dimension: 32
  (Defines the size of the LoRA network layers.)

• Network Alpha: 16
  (Controls the scaling factor for LoRA weights.)

• Network Module: LoRA (networks.lora)
  (Uses LoRA, a lightweight fine-tuning method for efficient training.)

Optimizer Settings

• Learning Rate: 0.0005
  (The base learning rate for the optimizer.)

• Learning Rate Scheduler: Cosine with Restarts
  (Gradually reduces the learning rate following a cosine curve, with 3 cycles of restarts.)

• Learning Rate Warmup Steps: 0
  (No warmup period for the learning rate.)

• Optimizer Type: Adafactor
  (An adaptive optimizer designed for memory efficiency.)

• Optimizer Arguments:

  • Scale Parameter: Disabled

  • Relative Step: Disabled

  • Warmup Init: Disabled
    (Custom settings to tweak how Adafactor behaves.)

Training Settings

• Maximum Training Steps: 0
  (No fixed step limit; training is controlled by epochs instead.)

• Maximum Training Epochs: 33
  (The model will train for 33 passes over the dataset.)

• Save Model Every N Epochs: 1
  (Saves the model after every epoch.)

• Sample Generation Every N Epochs: 1
  (Generates sample outputs after every epoch.)

• Sample Prompts File: Located at /workspace/training/7dd6c905-0edb-4cd6-bb4c-39fa6c179726/text/sample_prompts.txt
  (Uses prompts from this file to generate samples during training.)

• Sample Sampler: Euler_a
  (Uses the Euler Ancestral sampling method for generating images.)

• Training Batch Size: 4
  (Processes 4 images per batch during training.)

• Noise Offset: 0.1
  (Adds slight noise to training data to improve stability.)

• Clip Skip: 1
  (Skips the last layer of the CLIP model for text encoding.)

• Weighted Captions: Disabled
  (Treats all captions equally, without assigning weights.)

• Maximum Token Length: 225
  (Allows up to 225 tokens for text prompts.)

• Low RAM Mode: Disabled
  (Uses full RAM capacity for faster training.)

• Data Loader Workers: 8
  (Uses 8 parallel workers to load data, speeding up training.)

• Persistent Data Loader Workers: Enabled
  (Keeps data loader workers active between batches for efficiency.)

• Save Precision: Bfloat16 (bf16)
  (Saves the model in bfloat16 format for reduced memory usage.)

• Mixed Precision Training: Bfloat16 (bf16)
  (Uses bfloat16 for calculations to balance speed and precision.)

• Output Directory: /workspace/training/7dd6c905-0edb-4cd6-bb4c-39fa6c179726/model
  (Where trained models are saved.)

• Logging Directory: /workspace/training/7dd6c905-0edb-4cd6-bb4c-39fa6c179726/logs
  (Where training logs are stored.)

• Output Model Name: manly20250529--01khmer001
  (The name of the saved model.)

• Save Training State: Disabled
  (Does not save the full training state, only the model weights.)

• Xformers: Enabled
  (Uses Xformers for optimized attention mechanisms, improving speed.)

• SDPA (Scaled Dot-Product Attention): Enabled
  (Enables efficient attention computation for better performance.)

• No Half VAE: Enabled
  (Disables half-precision for the Variational Autoencoder to maintain quality.)

• Gradient Checkpointing: Enabled
  (Reduces memory usage by recomputing gradients during backpropagation.)

• Gradient Accumulation Steps: 1
  (Processes gradients in a single step, no accumulation.)

Advanced Training Settings

• Multi-Resolution Noise Iterations: 6
  (Applies noise at multiple resolutions for 6 iterations to improve image quality.)

• Multi-Resolution Noise Discount: 0.3
  (Reduces noise impact by 30% across iterations.)

• Minimum SNR Gamma: 5.0
  (Enforces a minimum signal-to-noise ratio to stabilize training.)

Model Settings

• Pretrained Model Path: /model_cache/@civitai/889818/889818.safetensors
  (Uses this pre-trained model as the starting point.)

• V2 Model: Disabled
  (Not using a V2 model architecture.)

Saving Settings

• Save Model Format: Safetensors
  (Saves the model in the efficient Safetensors format.)

DreamBooth Settings

• Prior Loss Weight: 1.0
  (Balances the influence of prior preservation loss in DreamBooth training.)

Dataset Settings

• Cache Latents: Enabled
  (Precomputes and caches latent representations of images to speed up training.)

This configuration is tailored for fine-tuning a model (likely a Stable Diffusion model) using LoRA and DreamBooth techniques, with a focus on efficiency and quality. It uses bfloat16 precision, advanced optimization techniques, and noise management to produce a high-quality model named manly20250529--01khmer001.