Upload 2962460/3356275/jamie_h3_ref2v_v4.yml with huggingface_hub

2962460/3356275/jamie_h3_ref2v_v4.yml

@@ -0,0 +1,127 @@
+---
+job: "extension"
+meta:
+  name: "[name]"
+  version: "1.0"
+config:
+  name: "jamie_h3_ref2v_v4"
+  process:
+    - save:
+        dtype: "bf16"
+        save_every: 250
+        push_to_hub: false
+        save_format: "diffusers"
+        max_step_saves_to_keep: 4
+      type: "diffusion_trainer"
+      model:
+        arch: "minimax_h3_ref2va"
+        qtype: "convrot8"
+        compile: false
+        low_vram: true
+        qtype_te: "nvfp4"
+        quantize: true
+        quantize_te: true
+        model_kwargs: {}
+        name_or_path: "Comfy-Org/MiniMax-H3"
+        layer_offloading: false
+        assistant_lora_path: "ostris/minimax_h3_training_adapter/minimax_h3_ref2va_training_adapter_v1.safetensors"
+        layer_offloading_transformer_percent: 1
+        layer_offloading_text_encoder_percent: 1
+      train:
+        lr: 0.0001
+        dtype: "bf16"
+        steps: 500
+        loss_type: "mse"
+        optimizer: "adamw8bit"
+        batch_size: 1
+        ema_config:
+          use_ema: false
+          ema_decay: 0.99
+        train_unet: true
+        timestep_type: "shift"
+        noise_scheduler: "flowmatch"
+        content_or_style: "balanced"
+        disable_sampling: false
+        do_guidance_loss: true
+        optimizer_params:
+          weight_decay: 0.0001
+        skip_first_sample: false
+        force_first_sample: false
+        train_text_encoder: false
+        unload_text_encoder: false
+        guidance_loss_target: 3.5
+        audio_loss_multiplier: 1
+        cache_text_embeddings: true
+        gradient_accumulation: 1
+        switch_boundary_every: 1
+        gradient_checkpointing: true
+        diff_output_preservation: false
+        bypass_guidance_embedding: false
+        diff_output_preservation_class: "person"
+        diff_output_preservation_multiplier: 1
+      device: "cuda"
+      sample:
+        fps: 24
+        neg: ""
+        seed: 42
+        width: 768
+        height: 768
+        sampler: "flowmatch"
+        samples:
+          - prompt: "j-rogue, Jamie-style woman, close-up face, bedroom window light, natural expression"
+          - prompt: "j-rogue, woman with expressive face in a bedroom, soft window light, cinematic portrait"
+          - prompt: "j-rogue, Jamie-style woman by a bedroom window, detailed face, warm morning light"
+          - prompt: "j-rogue, woman sitting near a bedroom window, face in focus, soft natural light"
+          - prompt: "a bear building a log cabin in the snow covered mountains"
+          - prompt: "woman playing the guitar, on stage, singing a song, laser lights, punk rocker"
+          - prompt: "hipster man with a beard, building a chair, in a wood shop"
+          - prompt: "photo of a man, white background, medium shot, modeling clothing, studio lighting, white backdrop"
+          - prompt: "a man holding a sign that says, 'this is a sign'"
+          - prompt: "a bulldog, in a post apocalyptic world, with a shotgun, in a leather jacket, in a desert, with a motorcycle"
+        walk_seed: true
+        num_frames: 39
+        sample_every: 250
+        sample_steps: 28
+        guidance_scale: 1
+        sample_start_step: 0
+      logging:
+        log_every: 1
+        use_ui_logger: true
+      network:
+        conv: 16
+        type: "lora"
+        linear: 16
+        conv_alpha: 16
+        lokr_factor: -1
+        linear_alpha: 16
+        lokr_full_rank: true
+        network_kwargs:
+          ignore_if_contains:
+            - "adaln_proj"
+      datasets:
+        - fps: 24
+          flip_x: false
+          flip_y: false
+          is_reg: false
+          controls: []
+          mask_path: null
+          num_frames: 39
+          resolution:
+            - 768
+          caption_ext: "txt"
+          folder_path: "/app/ai-toolkit/datasets/jamie-h3-targets"
+          num_repeats: 1
+          control_path_1: "/app/ai-toolkit/datasets/jamie-h3-refs"
+          control_path_2: null
+          control_path_3: null
+          mask_min_value: 0.1
+          network_weight: 1
+          default_caption: ""
+          auto_frame_count: false
+          caption_dropout_rate: 0.05
+          cache_latents_to_disk: true
+          shrink_video_to_frames: true
+      trigger_word: "j-rogue"
+      sqlite_db_path: "./aitk_db.db"
+      training_folder: "/app/ai-toolkit/output"
+      performance_log_every: 10