Here's a human-readable version of the configuration, organized into clear sections with explanations:<hr />Network Configuration<ul><li>U-Net Learning Rate: 0.0005 (Controls how fast the U-Net model learns during training.)</li><li>Text Encoder Learning Rate: 0.00005 (Sets the learning rate for the text encoder, which processes text prompts.)</li><li>Network Dimension: 32 (Defines the size of the LoRA network layers.)</li><li>Network Alpha: 16 (Controls the scaling factor for LoRA weights.)</li><li>Network Module: LoRA (networks.lora) (Uses LoRA, a lightweight fine-tuning method for efficient training.)</li></ul><hr />Optimizer Settings<ul><li>Learning Rate: 0.0005 (The base learning rate for the optimizer.)</li><li>Learning Rate Scheduler: Cosine with Restarts (Gradually reduces the learning rate following a cosine curve, with 3 cycles of restarts.)</li><li>Learning Rate Warmup Steps: 0 (No warmup period for the learning rate.)</li><li>Optimizer Type: Adafactor (An adaptive optimizer designed for memory efficiency.)</li><li>Optimizer Arguments:<ul><li>Scale Parameter: Disabled</li><li>Relative Step: Disabled</li><li>Warmup Init: Disabled (Custom settings to tweak how Adafactor behaves.)</li></ul></li></ul><hr />Training Settings<ul><li>Maximum Training Steps: 0 (No fixed step limit; training is controlled by epochs instead.)</li><li>Maximum Training Epochs: 33 (The model will train for 33 passes over the dataset.)</li><li>Save Model Every N Epochs: 1 (Saves the model after every epoch.)</li><li>Sample Generation Every N Epochs: 1 (Generates sample outputs after every epoch.)</li><li>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.)</li><li>Sample Sampler: Euler_a (Uses the Euler Ancestral sampling method for generating images.)</li><li>Training Batch Size: 4 (Processes 4 images per batch during training.)</li><li>Noise Offset: 0.1 (Adds slight noise to training data to improve stability.)</li><li>Clip Skip: 1 (Skips the last layer of the CLIP model for text encoding.)</li><li>Weighted Captions: Disabled (Treats all captions equally, without assigning weights.)</li><li>Maximum Token Length: 225 (Allows up to 225 tokens for text prompts.)</li><li>Low RAM Mode: Disabled (Uses full RAM capacity for faster training.)</li><li>Data Loader Workers: 8 (Uses 8 parallel workers to load data, speeding up training.)</li><li>Persistent Data Loader Workers: Enabled (Keeps data loader workers active between batches for efficiency.)</li><li>Save Precision: Bfloat16 (bf16) (Saves the model in bfloat16 format for reduced memory usage.)</li><li>Mixed Precision Training: Bfloat16 (bf16) (Uses bfloat16 for calculations to balance speed and precision.)</li><li>Output Directory: /workspace/training/7dd6c905-0edb-4cd6-bb4c-39fa6c179726/model (Where trained models are saved.)</li><li>Logging Directory: /workspace/training/7dd6c905-0edb-4cd6-bb4c-39fa6c179726/logs (Where training logs are stored.)</li><li>Output Model Name: manly20250529--01khmer001 (The name of the saved model.)</li><li>Save Training State: Disabled (Does not save the full training state, only the model weights.)</li><li>Xformers: Enabled (Uses Xformers for optimized attention mechanisms, improving speed.)</li><li>SDPA (Scaled Dot-Product Attention): Enabled (Enables efficient attention computation for better performance.)</li><li>No Half VAE: Enabled (Disables half-precision for the Variational Autoencoder to maintain quality.)</li><li>Gradient Checkpointing: Enabled (Reduces memory usage by recomputing gradients during backpropagation.)</li><li>Gradient Accumulation Steps: 1 (Processes gradients in a single step, no accumulation.)</li></ul><hr />Advanced Training Settings<ul><li>Multi-Resolution Noise Iterations: 6 (Applies noise at multiple resolutions for 6 iterations to improve image quality.)</li><li>Multi-Resolution Noise Discount: 0.3 (Reduces noise impact by 30% across iterations.)</li><li>Minimum SNR Gamma: 5.0 (Enforces a minimum signal-to-noise ratio to stabilize training.)</li></ul><hr />Model Settings<ul><li>Pretrained Model Path: /model_cache/@civitai/889818/889818.safetensors (Uses this pre-trained model as the starting point.)</li><li>V2 Model: Disabled (Not using a V2 model architecture.)</li></ul><hr />Saving Settings<ul><li>Save Model Format: Safetensors (Saves the model in the efficient Safetensors format.)</li></ul><hr />DreamBooth Settings<ul><li>Prior Loss Weight: 1.0 (Balances the influence of prior preservation loss in DreamBooth training.)</li></ul><hr />Dataset Settings<ul><li>Cache Latents: Enabled (Precomputes and caches latent representations of images to speed up training.)</li></ul><hr />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.

manly20250529--01khmer001

Here's a human-readable version of the provided configuration, organized into clear sections with explanations, followed by the sample prompts:Dataset Configuration<ul><li>Subsets:<ul><li>Number of Repeats: 4 (Each image in the dataset will be repeated 4 times during training to increase its influence.)</li><li>Image Directory: /workspace/training/7dd6c905-0edb-4cd6-bb4c-39fa6c179726/img (The folder containing the training images.)</li></ul></li></ul>General Settings<ul><li>Resolution: 1024 (Images will be processed at a resolution of 1024x1024 pixels.)</li><li>Shuffle Caption: Enabled (Randomly shuffles words in captions to improve generalization, while respecting the keep_tokens setting.)</li><li>Keep Tokens: 3 (Preserves the first 3 tokens of each caption in their original order during shuffling.)</li><li>Flip Augmentation: Enabled (Applies horizontal flipping to images during training to increase dataset variety.)</li><li>Caption Extension: .txt (Captions for images are stored in text files with a .txt extension.)</li><li>Enable Bucket: Enabled (Groups images into buckets based on their resolution for efficient training.)</li><li>Bucket Resolution Steps: 64 (Buckets are created in increments of 64 pixels to match image resolutions.)</li><li>Bucket No Upscale: Enabled (Prevents upscaling of images to fit bucket resolutions, preserving original sizes.)</li><li>Minimum Bucket Resolution: 256 (The smallest resolution bucket is 256x256 pixels.)</li><li>Maximum Bucket Resolution: 2048 (The largest resolution bucket is 2048x2048 pixels.)</li></ul>Sample PromptsThese are the prompts used to generate sample outputs during training (likely from the file referenced in the previous configuration). They describe the concepts the model is being trained to generate:<ol><li>Prompt: "father, rugby coach" (Generates images of a father who is a rugby coach.)</li><li>Prompt: "grandfather, rugby coach" (Generates images of a grandfather who is a rugby coach.)</li><li>Prompt: "father and grandfather went to mosque together" (Generates images of a father and grandfather together at a mosque.)</li></ol>

Files with SHA256: cc0f33e31ae6fd63500e064a67d621031df5f8ec1d3d996c1bd0f95901b8a991