prodigy_bf16_targeted

job: "extension"
config:
  name: "targeted_layers_bf16"
  process:
    - type: "diffusion_trainer"
      training_folder: "/app/ai-toolkit/output"
      sqlite_db_path: "./aitk_db.db"
      device: "cuda"
      trigger_word: null
      performance_log_every: 10
      network:
        type: "lora"
        linear: 64
        linear_alpha: 64
        conv: 32
        conv_alpha: 32
        lokr_full_rank: true
        lokr_factor: -1
        network_kwargs:
          only_if_contains:
            - "layers.14.attention.to_q"
            - "layers.14.attention.to_k"
            - "layers.14.attention.to_v"
            - "layers.14.attention.to_out.0"
            - "layers.14.feed_forward.w2"
            - "layers.14.feed_forward.w3"
            - "layers.15.attention.to_q"
            - "layers.15.attention.to_k"
            - "layers.15.attention.to_v"
            - "layers.15.attention.to_out.0"
            - "layers.15.feed_forward.w2"
            - "layers.15.feed_forward.w3"
            - "layers.16.attention.to_q"
            - "layers.16.attention.to_k"
            - "layers.16.attention.to_v"
            - "layers.16.attention.to_out.0"
            - "layers.16.feed_forward.w2"
            - "layers.16.feed_forward.w3"
            - "layers.17.attention.to_q"
            - "layers.17.attention.to_k"
            - "layers.17.attention.to_v"
            - "layers.17.attention.to_out.0"
            - "layers.17.feed_forward.w2"
            - "layers.17.feed_forward.w3"
            - "layers.18.attention.to_q"
            - "layers.18.attention.to_k"
            - "layers.18.attention.to_v"
            - "layers.18.attention.to_out.0"
            - "layers.18.feed_forward.w2"
            - "layers.18.feed_forward.w3"
            - "layers.19.attention.to_q"
            - "layers.19.attention.to_k"
            - "layers.19.attention.to_v"
            - "layers.19.attention.to_out.0"
            - "layers.19.feed_forward.w2"
            - "layers.19.feed_forward.w3"
            - "layers.20.attention.to_q"
            - "layers.20.attention.to_k"
            - "layers.20.attention.to_v"
            - "layers.20.attention.to_out.0"
            - "layers.20.feed_forward.w2"
            - "layers.20.feed_forward.w3"
            - "layers.21.attention.to_q"
            - "layers.21.attention.to_k"
            - "layers.21.attention.to_v"
            - "layers.21.attention.to_out.0"
            - "layers.21.feed_forward.w2"
            - "layers.21.feed_forward.w3"
            - "layers.22.attention.to_q"
            - "layers.22.attention.to_k"
            - "layers.22.attention.to_v"
            - "layers.22.attention.to_out.0"
            - "layers.22.feed_forward.w2"
            - "layers.22.feed_forward.w3"
            - "layers.23.attention.to_q"
            - "layers.23.attention.to_k"
            - "layers.23.attention.to_v"
            - "layers.23.attention.to_out.0"
            - "layers.23.feed_forward.w2"
            - "layers.23.feed_forward.w3"
            - "layers.24.attention.to_q"
            - "layers.24.attention.to_k"
            - "layers.24.attention.to_v"
            - "layers.24.attention.to_out.0"
            - "layers.24.feed_forward.w2"
            - "layers.24.feed_forward.w3"
            - "layers.25.attention.to_q"
            - "layers.25.attention.to_k"
            - "layers.25.attention.to_v"
            - "layers.25.attention.to_out.0"
            - "layers.25.feed_forward.w2"
            - "layers.25.feed_forward.w3"
      save:
        dtype: "bf16"
        save_every: 250
        max_step_saves_to_keep: 12
        save_format: "diffusers"
        push_to_hub: false
      datasets:
        - folder_path: "/app/ai-toolkit/datasets/new"
          mask_path: null
          mask_min_value: 0.1
          default_caption: ""
          caption_ext: "txt"
          caption_dropout_rate: 0.05
          cache_latents_to_disk: true
          is_reg: false
          network_weight: 1
          resolution:
            - 512
          controls: []
          shrink_video_to_frames: true
          num_frames: 1
          flip_x: false
          flip_y: false
          num_repeats: 1
      train:
        batch_size: 1
        bypass_guidance_embedding: false
        steps: 5000
        gradient_accumulation: 1
        train_unet: true
        train_text_encoder: false
        gradient_checkpointing: true
        noise_scheduler: "flowmatch"
        optimizer: "prodigy"
        timestep_type: "sigmoid"
        content_or_style: "balanced"
        optimizer_params:
          weight_decay: 0.01
          d_coef: 2
          safeguard_warmup: true
          use_bias_correction: true
          betas:
            - 0.9
            - 0.99
        unload_text_encoder: false
        cache_text_embeddings: true
        lr: 1
        ema_config:
          use_ema: false
          ema_decay: 0.99
        skip_first_sample: true
        force_first_sample: false
        disable_sampling: false
        dtype: "bf16"
        diff_output_preservation: false
        diff_output_preservation_multiplier: 1
        diff_output_preservation_class: "person"
        switch_boundary_every: 1
        loss_type: "mse"
      logging:
        log_every: 1
        use_ui_logger: true
      model:
        name_or_path: "Tongyi-MAI/Z-Image-Turbo"
        quantize: false
        qtype: "qfloat8"
        quantize_te: false
        qtype_te: "qfloat8"
        arch: "zimage:turbo"
        low_vram: false
        model_kwargs: {}
        layer_offloading: false
        layer_offloading_text_encoder_percent: 1
        layer_offloading_transformer_percent: 1
        assistant_lora_path: "ostris/zimage_turbo_training_adapter/zimage_turbo_training_adapter_v2.safetensors"
      sample:
        sampler: "flowmatch"
        sample_every: 250
        width: 1024
        height: 1024
        samples:
          - prompt: "from a close range we see, a beautiful woman staring at the camera"
            seed: 42
        walk_seed: true
        guidance_scale: 1
        sample_steps: 8
        num_frames: 1
        fps: 1
meta:
  name: "targeted_layers_bf16"
  version: "1.0"