Add files using upload-large-folder tool

01_11_2025/31/models/net_g_40000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6f998f444707892e357e3ab32099e937a7e2afadc60dc2eb0f32212ae3049e8d
+size 61082991

01_11_2025/31_archived_20251101_183720/train_31_20251101_182408.log

@@ -0,0 +1,246 @@
+2025-11-01 18:24:08,512 INFO: 
+                ____                _       _____  ____
+               / __ ) ____ _ _____ (_)_____/ ___/ / __ \
+              / __  |/ __ `// ___// // ___/\__ \ / /_/ /
+             / /_/ // /_/ /(__  )/ // /__ ___/ // _, _/
+            /_____/ \__,_//____//_/ \___//____//_/ |_|
+     ______                   __   __                 __      __
+    / ____/____   ____   ____/ /  / /   __  __ _____ / /__   / /
+   / / __ / __ \ / __ \ / __  /  / /   / / / // ___// //_/  / /
+  / /_/ // /_/ // /_/ // /_/ /  / /___/ /_/ // /__ / /<    /_/
+  \____/ \____/ \____/ \____/  /_____/\____/ \___//_/|_|  (_)
+    
+Version Information: 
+	BasicSR: 1.4.2
+	PyTorch: 2.9.0+cu129
+	TorchVision: 0.24.0+cpu
+2025-11-01 18:24:08,512 INFO: 
+  model_type: SwinIRLatentModelMultiHead
+  primary_head: x4
+  scale: 4
+  num_gpu: 3
+  manual_seed: 0
+  find_unused_parameters: False
+  vae_sources:[
+    flux_vae:[
+      hf_repo: wolfgangblack/flux_vae
+      vae_kind: kl
+    ]
+  ]
+  datasets:[
+    train:[
+      name: gpt4_nanobana_midjourney_recraft_CropsFluxVAE
+      type: MultiScaleLatentCacheDataset
+      scales: [128, 256, 512]
+      cache_dirs: ['/data/kazanplova/datasets/full_latent_upscale_dataset_train/nano_banana_crops/embeddings/flux_vae', '/data/kazanplova/datasets/full_latent_upscale_dataset_train/midjourney_full_dataset_crops_new/embeddings/flux_vae', '/data/kazanplova/datasets/full_latent_upscale_dataset_train/gpt4_crops/embeddings/flux_vae', '/data/kazanplova/datasets/full_latent_upscale_dataset_train/recraft_data_crops/embeddings/flux_vae', '/data/kazanplova/datasets/full_latent_upscale_dataset_train/mjv5_crops_fix/embeddings/flux_vae']
+      vae_names: ['flux_vae']
+      phase: train
+      filename_tmpl: {}
+      io_backend:[
+        type: disk
+      ]
+      scale: 4
+      mean: None
+      std: None
+      num_worker_per_gpu: 16
+      batch_size_per_gpu: 256
+      pin_memory: True
+      persistent_workers: True
+    ]
+    val:[
+      name: sdxk_120_1024x1024
+      type: MultiScaleLatentCacheDataset
+      scales: [256, 512, 1024]
+      cache_dirs: ['/data/kazanplova/datasets/latent_upscale_validation_120_samples/embeddings/flux_vae']
+      vae_names: ['flux_vae']
+      phase: val
+      io_backend:[
+        type: disk
+      ]
+      scale: 4
+      mean: None
+      std: None
+      batch_size_per_gpu: 16
+      num_worker_per_gpu: 4
+      pin_memory: True
+    ]
+  ]
+  network_g:[
+    type: SwinIRMultiHead
+    in_chans: 16
+    img_size: 32
+    window_size: 8
+    img_range: 1.0
+    depths: [6, 6, 6, 6, 6, 6]
+    embed_dim: 180
+    num_heads: [6, 6, 6, 6, 6, 6]
+    mlp_ratio: 2
+    resi_connection: 1conv
+    head_num_feat: 128
+    primary_head: x4
+    heads: [OrderedDict({'name': 'x2', 'scale': 2, 'out_chans': 16}), OrderedDict({'name': 'x4', 'scale': 4, 'out_chans': 16, 'primary': True})]
+  ]
+  path:[
+    pretrain_network_g: None
+    strict_load_g: False
+    experiments_root: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/31
+    models: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/31/models
+    training_states: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/31/training_states
+    log: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/31
+    visualization: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/31/visualization
+  ]
+  compile:[
+    enabled: False
+    mode: max-autotune
+    dynamic: True
+    fullgraph: False
+    backend: None
+  ]
+  train:[
+    ema_decay: 0.999
+    head_inputs:[
+      x2:[
+        lq: 256
+        gt: 512
+      ]
+      x4:[
+        lq: 128
+        gt: 512
+      ]
+    ]
+    optim_g:[
+      type: Adam
+      lr: 0.0002
+      weight_decay: 0
+      betas: [0.9, 0.995]
+    ]
+    grad_clip:[
+      enabled: True
+      generator:[
+        type: norm
+        max_norm: 0.4
+        norm_type: 2.0
+      ]
+    ]
+    scheduler:[
+      type: MultiStepLR
+      milestones: [62500, 93750, 112500]
+      gamma: 0.5
+    ]
+    total_steps: 125000
+    warmup_iter: -1
+    l1_latent_x2_opt:[
+      type: L1Loss
+      loss_weight: 1.0
+      reduction: mean
+      space: latent
+      target: x2
+    ]
+    l1_latent_x4_opt:[
+      type: L1Loss
+      loss_weight: 1.0
+      reduction: mean
+      space: latent
+      target: x4
+    ]
+    fft_latent_x2_opt:[
+      type: FFTFrequencyLoss
+      loss_weight: 0.1
+      reduction: mean
+      space: latent
+      target: x2
+      norm: ortho
+      use_log_amplitude: False
+      alpha: 0.0
+      normalize_weight: True
+      eps: 1e-8
+    ]
+    fft_latent_x4_opt:[
+      type: FFTFrequencyLoss
+      loss_weight: 0.1
+      reduction: mean
+      space: latent
+      target: x4
+      norm: ortho
+      use_log_amplitude: False
+      alpha: 0.0
+      normalize_weight: True
+      eps: 1e-8
+    ]
+  ]
+  val:[
+    val_freq: 5000
+    save_img: True
+    head_evals:[
+      x2:[
+        save_img: True
+        label: val_x2
+        val_sizes:[
+          lq: 512
+          gt: 1024
+        ]
+        metrics:[
+          l1_latent:[
+            type: L1Loss
+            space: latent
+          ]
+          pixel_psnr_pt:[
+            type: calculate_psnr_pt
+            space: pixel
+            crop_border: 2
+            test_y_channel: False
+          ]
+        ]
+      ]
+      x4:[
+        save_img: True
+        label: val_x4
+        val_sizes:[
+          lq: 256
+          gt: 1024
+        ]
+        metrics:[
+          l1_latent:[
+            type: L1Loss
+            space: latent
+          ]
+          l2_latent:[
+            type: MSELoss
+            space: latent
+          ]
+          pixel_psnr_pt:[
+            type: calculate_psnr_pt
+            space: pixel
+            crop_border: 2
+            test_y_channel: False
+          ]
+        ]
+      ]
+    ]
+  ]
+  logger:[
+    print_freq: 100
+    save_checkpoint_freq: 5000
+    use_tb_logger: True
+    wandb:[
+      project: Swin2SR-Latent-SR
+      entity: kazanplova-it-more
+      resume_id: None
+      max_val_images: 10
+    ]
+  ]
+  dist_params:[
+    backend: nccl
+    port: 29500
+    dist: True
+  ]
+  load_networks_only: False
+  exp_name: 31
+  name: 31
+  dist: True
+  rank: 0
+  world_size: 3
+  auto_resume: False
+  is_train: True
+  root_path: /data/kazanplova/latent_vae_upscale_train
+

01_11_2025/32/basicsr_options.yaml

@@ -0,0 +1,220 @@
+# GENERATE TIME: Sat Nov  1 18:55:31 2025
+# CMD:
+# train_vae.py -opt /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/basicsr_options.yaml --launcher pytorch --local_rank 0
+
+model_type: SwinIRLatentModelMultiHead
+primary_head: x4
+scale: 4
+num_gpu: auto
+manual_seed: 0
+find_unused_parameters: false
+vae_sources:
+  flux_vae:
+    hf_repo: wolfgangblack/flux_vae
+    vae_kind: kl
+datasets:
+  train:
+    name: gpt4_nanobana_midjourney_recraft_CropsFluxVAE
+    type: MultiScaleLatentCacheDataset
+    scales:
+    - 128
+    - 256
+    - 512
+    cache_dirs:
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/nano_banana_crops/embeddings/flux_vae
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/midjourney_full_dataset_crops_new/embeddings/flux_vae
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/gpt4_crops/embeddings/flux_vae
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/recraft_data_crops/embeddings/flux_vae
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/mjv5_crops_fix/embeddings/flux_vae
+    vae_names:
+    - flux_vae
+    phase: train
+    filename_tmpl: '{}'
+    io_backend:
+      type: disk
+    scale: 4
+    mean: null
+    std: null
+    num_worker_per_gpu: 32
+    batch_size_per_gpu: 64
+    pin_memory: true
+    persistent_workers: true
+  val:
+    name: sdxk_120_1024x1024
+    type: MultiScaleLatentCacheDataset
+    scales:
+    - 256
+    - 512
+    - 1024
+    cache_dirs:
+    - /data/kazanplova/datasets/latent_upscale_validation_120_samples/embeddings/flux_vae
+    vae_names:
+    - flux_vae
+    phase: val
+    io_backend:
+      type: disk
+    scale: 4
+    mean: null
+    std: null
+    batch_size_per_gpu: 16
+    num_worker_per_gpu: 4
+    pin_memory: true
+network_g:
+  type: SwinIRMultiHead
+  in_chans: 16
+  img_size: 32
+  window_size: 16
+  img_range: 1.0
+  depths:
+  - 6
+  - 6
+  - 6
+  - 6
+  - 6
+  - 6
+  embed_dim: 360
+  num_heads:
+  - 12
+  - 12
+  - 12
+  - 12
+  - 12
+  - 12
+  mlp_ratio: 2
+  resi_connection: 1conv
+  primary_head: x4
+  head_num_feat: 256
+  heads:
+  - name: x2
+    scale: 2
+    out_chans: 16
+  - name: x4
+    scale: 4
+    out_chans: 16
+    primary: true
+compile:
+  enabled: false
+  mode: max-autotune
+  dynamic: true
+  fullgraph: false
+  backend: null
+train:
+  ema_decay: 0.999
+  head_inputs:
+    x2:
+      lq: 256
+      gt: 512
+    x4:
+      lq: 128
+      gt: 512
+  optim_g:
+    type: Adam
+    lr: 0.0002
+    weight_decay: 0
+    betas:
+    - 0.9
+    - 0.995
+  grad_clip:
+    enabled: true
+    generator:
+      type: norm
+      max_norm: 0.4
+      norm_type: 2.0
+  scheduler:
+    type: MultiStepLR
+    milestones:
+    - 62500
+    - 93750
+    - 112500
+    gamma: 0.5
+  total_steps: 125000
+  warmup_iter: -1
+  l1_latent_x2_opt:
+    type: L1Loss
+    loss_weight: 1.0
+    reduction: mean
+    space: latent
+    target: x2
+  l1_latent_x4_opt:
+    type: L1Loss
+    loss_weight: 1.0
+    reduction: mean
+    space: latent
+    target: x4
+  fft_latent_x2_opt:
+    type: FFTFrequencyLoss
+    loss_weight: 0.1
+    reduction: mean
+    space: latent
+    target: x2
+    norm: ortho
+    use_log_amplitude: false
+    alpha: 0.0
+    normalize_weight: true
+    eps: 1e-8
+  fft_latent_x4_opt:
+    type: FFTFrequencyLoss
+    loss_weight: 0.1
+    reduction: mean
+    space: latent
+    target: x4
+    norm: ortho
+    use_log_amplitude: false
+    alpha: 0.0
+    normalize_weight: true
+    eps: 1e-8
+val:
+  val_freq: 2500
+  save_img: true
+  head_evals:
+    x2:
+      save_img: true
+      label: val_x2
+      val_sizes:
+        lq: 512
+        gt: 1024
+      metrics:
+        l1_latent:
+          type: L1Loss
+          space: latent
+        pixel_psnr_pt:
+          type: calculate_psnr_pt
+          space: pixel
+          crop_border: 2
+          test_y_channel: false
+    x4:
+      save_img: true
+      label: val_x4
+      val_sizes:
+        lq: 256
+        gt: 1024
+      metrics:
+        l1_latent:
+          type: L1Loss
+          space: latent
+        l2_latent:
+          type: MSELoss
+          space: latent
+        pixel_psnr_pt:
+          type: calculate_psnr_pt
+          space: pixel
+          crop_border: 2
+          test_y_channel: false
+logger:
+  print_freq: 100
+  save_checkpoint_freq: 2500
+  use_tb_logger: true
+  wandb:
+    project: Swin2SR-Latent-SR
+    entity: kazanplova-it-more
+    resume_id: null
+    max_val_images: 10
+dist_params:
+  backend: nccl
+  port: 29500
+  dist: true
+load_networks_only: false
+exp_name: '32'
+name: '32'
+path:
+  experiments_root: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025

01_11_2025/32_4_archived_20251101_173315/basicsr_options.yaml

@@ -0,0 +1,220 @@
+# GENERATE TIME: Sat Nov  1 17:33:12 2025
+# CMD:
+# train_vae.py -opt /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/basicsr_options.yaml
+
+model_type: SwinIRLatentModelMultiHead
+primary_head: x4
+scale: 4
+num_gpu: auto
+manual_seed: 0
+find_unused_parameters: false
+vae_sources:
+  flux_vae:
+    hf_repo: wolfgangblack/flux_vae
+    vae_kind: kl
+datasets:
+  train:
+    name: gpt4_nanobana_midjourney_recraft_CropsFluxVAE
+    type: MultiScaleLatentCacheDataset
+    scales:
+    - 128
+    - 256
+    - 512
+    cache_dirs:
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/nano_banana_crops/embeddings/flux_vae
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/midjourney_full_dataset_crops_new/embeddings/flux_vae
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/gpt4_crops/embeddings/flux_vae
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/recraft_data_crops/embeddings/flux_vae
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/mjv5_crops_fix/embeddings/flux_vae
+    vae_names:
+    - flux_vae
+    phase: train
+    filename_tmpl: '{}'
+    io_backend:
+      type: disk
+    scale: 4
+    mean: null
+    std: null
+    num_worker_per_gpu: 12
+    batch_size_per_gpu: 48
+    pin_memory: true
+    persistent_workers: true
+  val:
+    name: sdxk_120_1024x1024
+    type: MultiScaleLatentCacheDataset
+    scales:
+    - 256
+    - 512
+    - 1024
+    cache_dirs:
+    - /data/kazanplova/datasets/latent_upscale_validation_120_samples/embeddings/flux_vae
+    vae_names:
+    - flux_vae
+    phase: val
+    io_backend:
+      type: disk
+    scale: 4
+    mean: null
+    std: null
+    batch_size_per_gpu: 16
+    num_worker_per_gpu: 4
+    pin_memory: true
+network_g:
+  type: SwinIRMultiHead
+  in_chans: 16
+  img_size: 32
+  window_size: 16
+  img_range: 1.0
+  depths:
+  - 6
+  - 6
+  - 6
+  - 6
+  - 6
+  - 6
+  embed_dim: 360
+  num_heads:
+  - 12
+  - 12
+  - 12
+  - 12
+  - 12
+  - 12
+  mlp_ratio: 2
+  resi_connection: 1conv
+  primary_head: x4
+  head_num_feat: 256
+  heads:
+  - name: x2
+    scale: 2
+    out_chans: 16
+  - name: x4
+    scale: 4
+    out_chans: 16
+    primary: true
+compile:
+  enabled: false
+  mode: max-autotune
+  dynamic: true
+  fullgraph: false
+  backend: null
+train:
+  ema_decay: 0.999
+  head_inputs:
+    x2:
+      lq: 256
+      gt: 512
+    x4:
+      lq: 128
+      gt: 512
+  optim_g:
+    type: Adam
+    lr: 0.0002
+    weight_decay: 0
+    betas:
+    - 0.9
+    - 0.995
+  grad_clip:
+    enabled: true
+    generator:
+      type: norm
+      max_norm: 0.4
+      norm_type: 2.0
+  scheduler:
+    type: MultiStepLR
+    milestones:
+    - 62500
+    - 93750
+    - 112500
+    gamma: 0.5
+  total_steps: 125000
+  warmup_iter: -1
+  l1_latent_x2_opt:
+    type: L1Loss
+    loss_weight: 1.0
+    reduction: mean
+    space: latent
+    target: x2
+  l1_latent_x4_opt:
+    type: L1Loss
+    loss_weight: 1.0
+    reduction: mean
+    space: latent
+    target: x4
+  fft_latent_x2_opt:
+    type: FFTFrequencyLoss
+    loss_weight: 0.1
+    reduction: mean
+    space: latent
+    target: x2
+    norm: ortho
+    use_log_amplitude: false
+    alpha: 0.0
+    normalize_weight: true
+    eps: 1e-8
+  fft_latent_x4_opt:
+    type: FFTFrequencyLoss
+    loss_weight: 0.1
+    reduction: mean
+    space: latent
+    target: x4
+    norm: ortho
+    use_log_amplitude: false
+    alpha: 0.0
+    normalize_weight: true
+    eps: 1e-8
+val:
+  val_freq: 5000
+  save_img: true
+  head_evals:
+    x2:
+      save_img: true
+      label: val_x2
+      val_sizes:
+        lq: 512
+        gt: 1024
+      metrics:
+        l1_latent:
+          type: L1Loss
+          space: latent
+        pixel_psnr_pt:
+          type: calculate_psnr_pt
+          space: pixel
+          crop_border: 2
+          test_y_channel: false
+    x4:
+      save_img: true
+      label: val_x4
+      val_sizes:
+        lq: 256
+        gt: 1024
+      metrics:
+        l1_latent:
+          type: L1Loss
+          space: latent
+        l2_latent:
+          type: MSELoss
+          space: latent
+        pixel_psnr_pt:
+          type: calculate_psnr_pt
+          space: pixel
+          crop_border: 2
+          test_y_channel: false
+logger:
+  print_freq: 100
+  save_checkpoint_freq: 5000
+  use_tb_logger: true
+  wandb:
+    project: Swin2SR-Latent-SR
+    entity: kazanplova-it-more
+    resume_id: null
+    max_val_images: 10
+dist_params:
+  backend: nccl
+  port: 29500
+  dist: true
+load_networks_only: false
+exp_name: '32'
+name: '32_4'
+path:
+  experiments_root: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025

01_11_2025/32_5/basicsr_options.yaml

@@ -0,0 +1,220 @@
+# GENERATE TIME: Sat Nov  1 17:52:16 2025
+# CMD:
+# train_vae.py -opt /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/basicsr_options.yaml
+
+model_type: SwinIRLatentModelMultiHead
+primary_head: x4
+scale: 4
+num_gpu: auto
+manual_seed: 0
+find_unused_parameters: false
+vae_sources:
+  flux_vae:
+    hf_repo: wolfgangblack/flux_vae
+    vae_kind: kl
+datasets:
+  train:
+    name: gpt4_nanobana_midjourney_recraft_CropsFluxVAE
+    type: MultiScaleLatentCacheDataset
+    scales:
+    - 128
+    - 256
+    - 512
+    cache_dirs:
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/nano_banana_crops/embeddings/flux_vae
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/midjourney_full_dataset_crops_new/embeddings/flux_vae
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/gpt4_crops/embeddings/flux_vae
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/recraft_data_crops/embeddings/flux_vae
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/mjv5_crops_fix/embeddings/flux_vae
+    vae_names:
+    - flux_vae
+    phase: train
+    filename_tmpl: '{}'
+    io_backend:
+      type: disk
+    scale: 4
+    mean: null
+    std: null
+    num_worker_per_gpu: 32
+    batch_size_per_gpu: 128
+    pin_memory: true
+    persistent_workers: true
+  val:
+    name: sdxk_120_1024x1024
+    type: MultiScaleLatentCacheDataset
+    scales:
+    - 256
+    - 512
+    - 1024
+    cache_dirs:
+    - /data/kazanplova/datasets/latent_upscale_validation_120_samples/embeddings/flux_vae
+    vae_names:
+    - flux_vae
+    phase: val
+    io_backend:
+      type: disk
+    scale: 4
+    mean: null
+    std: null
+    batch_size_per_gpu: 16
+    num_worker_per_gpu: 4
+    pin_memory: true
+network_g:
+  type: SwinIRMultiHead
+  in_chans: 16
+  img_size: 32
+  window_size: 16
+  img_range: 1.0
+  depths:
+  - 6
+  - 6
+  - 6
+  - 6
+  - 6
+  - 6
+  embed_dim: 360
+  num_heads:
+  - 12
+  - 12
+  - 12
+  - 12
+  - 12
+  - 12
+  mlp_ratio: 2
+  resi_connection: 1conv
+  primary_head: x4
+  head_num_feat: 256
+  heads:
+  - name: x2
+    scale: 2
+    out_chans: 16
+  - name: x4
+    scale: 4
+    out_chans: 16
+    primary: true
+compile:
+  enabled: false
+  mode: max-autotune
+  dynamic: true
+  fullgraph: false
+  backend: null
+train:
+  ema_decay: 0.999
+  head_inputs:
+    x2:
+      lq: 256
+      gt: 512
+    x4:
+      lq: 128
+      gt: 512
+  optim_g:
+    type: Adam
+    lr: 0.0002
+    weight_decay: 0
+    betas:
+    - 0.9
+    - 0.995
+  grad_clip:
+    enabled: true
+    generator:
+      type: norm
+      max_norm: 0.4
+      norm_type: 2.0
+  scheduler:
+    type: MultiStepLR
+    milestones:
+    - 62500
+    - 93750
+    - 112500
+    gamma: 0.5
+  total_steps: 125000
+  warmup_iter: -1
+  l1_latent_x2_opt:
+    type: L1Loss
+    loss_weight: 1.0
+    reduction: mean
+    space: latent
+    target: x2
+  l1_latent_x4_opt:
+    type: L1Loss
+    loss_weight: 1.0
+    reduction: mean
+    space: latent
+    target: x4
+  fft_latent_x2_opt:
+    type: FFTFrequencyLoss
+    loss_weight: 0.1
+    reduction: mean
+    space: latent
+    target: x2
+    norm: ortho
+    use_log_amplitude: false
+    alpha: 0.0
+    normalize_weight: true
+    eps: 1e-8
+  fft_latent_x4_opt:
+    type: FFTFrequencyLoss
+    loss_weight: 0.1
+    reduction: mean
+    space: latent
+    target: x4
+    norm: ortho
+    use_log_amplitude: false
+    alpha: 0.0
+    normalize_weight: true
+    eps: 1e-8
+val:
+  val_freq: 5000
+  save_img: true
+  head_evals:
+    x2:
+      save_img: true
+      label: val_x2
+      val_sizes:
+        lq: 512
+        gt: 1024
+      metrics:
+        l1_latent:
+          type: L1Loss
+          space: latent
+        pixel_psnr_pt:
+          type: calculate_psnr_pt
+          space: pixel
+          crop_border: 2
+          test_y_channel: false
+    x4:
+      save_img: true
+      label: val_x4
+      val_sizes:
+        lq: 256
+        gt: 1024
+      metrics:
+        l1_latent:
+          type: L1Loss
+          space: latent
+        l2_latent:
+          type: MSELoss
+          space: latent
+        pixel_psnr_pt:
+          type: calculate_psnr_pt
+          space: pixel
+          crop_border: 2
+          test_y_channel: false
+logger:
+  print_freq: 100
+  save_checkpoint_freq: 5000
+  use_tb_logger: true
+  wandb:
+    project: Swin2SR-Latent-SR
+    entity: kazanplova-it-more
+    resume_id: null
+    max_val_images: 10
+dist_params:
+  backend: nccl
+  port: 29500
+  dist: true
+load_networks_only: false
+exp_name: '32'
+name: '32_5'
+path:
+  experiments_root: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025

01_11_2025/32_6/train_32_6_20251101_175325.log

@@ -0,0 +1,569 @@
+2025-11-01 17:53:25,678 INFO: 
+                ____                _       _____  ____
+               / __ ) ____ _ _____ (_)_____/ ___/ / __ \
+              / __  |/ __ `// ___// // ___/\__ \ / /_/ /
+             / /_/ // /_/ /(__  )/ // /__ ___/ // _, _/
+            /_____/ \__,_//____//_/ \___//____//_/ |_|
+     ______                   __   __                 __      __
+    / ____/____   ____   ____/ /  / /   __  __ _____ / /__   / /
+   / / __ / __ \ / __ \ / __  /  / /   / / / // ___// //_/  / /
+  / /_/ // /_/ // /_/ // /_/ /  / /___/ /_/ // /__ / /<    /_/
+  \____/ \____/ \____/ \____/  /_____/\____/ \___//_/|_|  (_)
+    
+Version Information: 
+	BasicSR: 1.4.2
+	PyTorch: 2.9.0+cu129
+	TorchVision: 0.24.0+cpu
+2025-11-01 17:53:25,679 INFO: 
+  model_type: SwinIRLatentModelMultiHead
+  primary_head: x4
+  scale: 4
+  num_gpu: 1
+  manual_seed: 0
+  find_unused_parameters: False
+  vae_sources:[
+    flux_vae:[
+      hf_repo: wolfgangblack/flux_vae
+      vae_kind: kl
+    ]
+  ]
+  datasets:[
+    train:[
+      name: gpt4_nanobana_midjourney_recraft_CropsFluxVAE
+      type: MultiScaleLatentCacheDataset
+      scales: [128, 256, 512]
+      cache_dirs: ['/data/kazanplova/datasets/full_latent_upscale_dataset_train/nano_banana_crops/embeddings/flux_vae', '/data/kazanplova/datasets/full_latent_upscale_dataset_train/midjourney_full_dataset_crops_new/embeddings/flux_vae', '/data/kazanplova/datasets/full_latent_upscale_dataset_train/gpt4_crops/embeddings/flux_vae', '/data/kazanplova/datasets/full_latent_upscale_dataset_train/recraft_data_crops/embeddings/flux_vae', '/data/kazanplova/datasets/full_latent_upscale_dataset_train/mjv5_crops_fix/embeddings/flux_vae']
+      vae_names: ['flux_vae']
+      phase: train
+      filename_tmpl: {}
+      io_backend:[
+        type: disk
+      ]
+      scale: 4
+      mean: None
+      std: None
+      num_worker_per_gpu: 16
+      batch_size_per_gpu: 64
+      pin_memory: True
+      persistent_workers: True
+    ]
+    val:[
+      name: sdxk_120_1024x1024
+      type: MultiScaleLatentCacheDataset
+      scales: [256, 512, 1024]
+      cache_dirs: ['/data/kazanplova/datasets/latent_upscale_validation_120_samples/embeddings/flux_vae']
+      vae_names: ['flux_vae']
+      phase: val
+      io_backend:[
+        type: disk
+      ]
+      scale: 4
+      mean: None
+      std: None
+      batch_size_per_gpu: 16
+      num_worker_per_gpu: 4
+      pin_memory: True
+    ]
+  ]
+  network_g:[
+    type: SwinIRMultiHead
+    in_chans: 16
+    img_size: 32
+    window_size: 16
+    img_range: 1.0
+    depths: [6, 6, 6, 6, 6, 6]
+    embed_dim: 360
+    num_heads: [12, 12, 12, 12, 12, 12]
+    mlp_ratio: 2
+    resi_connection: 1conv
+    primary_head: x4
+    head_num_feat: 256
+    heads: [OrderedDict({'name': 'x2', 'scale': 2, 'out_chans': 16}), OrderedDict({'name': 'x4', 'scale': 4, 'out_chans': 16, 'primary': True})]
+  ]
+  compile:[
+    enabled: False
+    mode: max-autotune
+    dynamic: True
+    fullgraph: False
+    backend: None
+  ]
+  train:[
+    ema_decay: 0.999
+    head_inputs:[
+      x2:[
+        lq: 256
+        gt: 512
+      ]
+      x4:[
+        lq: 128
+        gt: 512
+      ]
+    ]
+    optim_g:[
+      type: Adam
+      lr: 0.0002
+      weight_decay: 0
+      betas: [0.9, 0.995]
+    ]
+    grad_clip:[
+      enabled: True
+      generator:[
+        type: norm
+        max_norm: 0.4
+        norm_type: 2.0
+      ]
+    ]
+    scheduler:[
+      type: MultiStepLR
+      milestones: [62500, 93750, 112500]
+      gamma: 0.5
+    ]
+    total_steps: 125000
+    warmup_iter: -1
+    l1_latent_x2_opt:[
+      type: L1Loss
+      loss_weight: 1.0
+      reduction: mean
+      space: latent
+      target: x2
+    ]
+    l1_latent_x4_opt:[
+      type: L1Loss
+      loss_weight: 1.0
+      reduction: mean
+      space: latent
+      target: x4
+    ]
+    fft_latent_x2_opt:[
+      type: FFTFrequencyLoss
+      loss_weight: 0.1
+      reduction: mean
+      space: latent
+      target: x2
+      norm: ortho
+      use_log_amplitude: False
+      alpha: 0.0
+      normalize_weight: True
+      eps: 1e-8
+    ]
+    fft_latent_x4_opt:[
+      type: FFTFrequencyLoss
+      loss_weight: 0.1
+      reduction: mean
+      space: latent
+      target: x4
+      norm: ortho
+      use_log_amplitude: False
+      alpha: 0.0
+      normalize_weight: True
+      eps: 1e-8
+    ]
+  ]
+  val:[
+    val_freq: 5000
+    save_img: True
+    head_evals:[
+      x2:[
+        save_img: True
+        label: val_x2
+        val_sizes:[
+          lq: 512
+          gt: 1024
+        ]
+        metrics:[
+          l1_latent:[
+            type: L1Loss
+            space: latent
+          ]
+          pixel_psnr_pt:[
+            type: calculate_psnr_pt
+            space: pixel
+            crop_border: 2
+            test_y_channel: False
+          ]
+        ]
+      ]
+      x4:[
+        save_img: True
+        label: val_x4
+        val_sizes:[
+          lq: 256
+          gt: 1024
+        ]
+        metrics:[
+          l1_latent:[
+            type: L1Loss
+            space: latent
+          ]
+          l2_latent:[
+            type: MSELoss
+            space: latent
+          ]
+          pixel_psnr_pt:[
+            type: calculate_psnr_pt
+            space: pixel
+            crop_border: 2
+            test_y_channel: False
+          ]
+        ]
+      ]
+    ]
+  ]
+  logger:[
+    print_freq: 100
+    save_checkpoint_freq: 5000
+    use_tb_logger: True
+    wandb:[
+      project: Swin2SR-Latent-SR
+      entity: kazanplova-it-more
+      resume_id: None
+      max_val_images: 10
+    ]
+  ]
+  dist_params:[
+    backend: nccl
+    port: 29500
+    dist: True
+  ]
+  load_networks_only: False
+  exp_name: 32
+  name: 32_6
+  path:[
+    experiments_root: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/32_6
+    models: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/32_6/models
+    training_states: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/32_6/training_states
+    log: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/32_6
+    visualization: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/32_6/visualization
+  ]
+  dist: False
+  rank: 0
+  world_size: 1
+  auto_resume: False
+  is_train: True
+  root_path: /data/kazanplova/latent_vae_upscale_train
+
+2025-11-01 17:53:27,295 INFO: Use wandb logger with id=lip43i3d; project=Swin2SR-Latent-SR.
+2025-11-01 17:53:40,017 INFO: Dataset [MultiScaleLatentCacheDataset] - gpt4_nanobana_midjourney_recraft_CropsFluxVAE is built.
+2025-11-01 17:53:40,018 INFO: Training statistics:
+	Number of train images: 4858507
+	Dataset enlarge ratio: 1
+	Batch size per gpu: 64
+	World size (gpu number): 1
+	Steps per epoch: 75915
+	Configured training steps: 125000
+	Approximate epochs to cover: 2.
+2025-11-01 17:53:40,023 INFO: Dataset [MultiScaleLatentCacheDataset] - sdxk_120_1024x1024 is built.
+2025-11-01 17:53:40,023 INFO: Number of val images/folders in sdxk_120_1024x1024: 153
+2025-11-01 17:53:40,629 INFO: Network [SwinIRMultiHead] is created.
+2025-11-01 17:53:40,891 INFO: Network: SwinIRMultiHead, with parameters: 54,917,584
+2025-11-01 17:53:40,891 INFO: SwinIRMultiHead(
+  (conv_first): Conv2d(16, 360, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+  (patch_embed): PatchEmbed(
+    (norm): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+  )
+  (patch_unembed): PatchUnEmbed()
+  (pos_drop): Dropout(p=0.0, inplace=False)
+  (layers): ModuleList(
+    (0): RSTB(
+      (residual_group): BasicLayer(
+        dim=360, input_resolution=(32, 32), depth=6
+        (blocks): ModuleList(
+          (0): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=0, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): Identity()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (1): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=8, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (2): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=0, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (3): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=8, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (4): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=0, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (5): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=8, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+        )
+      )
+      (conv): Conv2d(360, 360, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+      (patch_embed): PatchEmbed()
+      (patch_unembed): PatchUnEmbed()
+    )
+    (1-5): 5 x RSTB(
+      (residual_group): BasicLayer(
+        dim=360, input_resolution=(32, 32), depth=6
+        (blocks): ModuleList(
+          (0): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=0, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (1): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=8, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (2): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=0, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (3): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=8, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (4): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=0, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (5): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=8, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+        )
+      )
+      (conv): Conv2d(360, 360, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+      (patch_embed): PatchEmbed()
+      (patch_unembed): PatchUnEmbed()
+    )
+  )
+  (norm): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+  (conv_after_body): Conv2d(360, 360, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+  (heads): ModuleDict(
+    (x2): _SwinIRPixelShuffleHead(
+      (conv_before): Sequential(
+        (0): Conv2d(360, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        (1): LeakyReLU(negative_slope=0.01, inplace=True)
+      )
+      (upsample): Upsample(
+        (0): Conv2d(256, 1024, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        (1): PixelShuffle(upscale_factor=2)
+      )
+      (conv_last): Conv2d(256, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+    )
+    (x4): _SwinIRPixelShuffleHead(
+      (conv_before): Sequential(
+        (0): Conv2d(360, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        (1): LeakyReLU(negative_slope=0.01, inplace=True)
+      )
+      (upsample): Upsample(
+        (0): Conv2d(256, 1024, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        (1): PixelShuffle(upscale_factor=2)
+        (2): Conv2d(256, 1024, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        (3): PixelShuffle(upscale_factor=2)
+      )
+      (conv_last): Conv2d(256, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+    )
+  )
+)
+2025-11-01 17:53:40,894 INFO: Use EMA with decay: 0.999
+2025-11-01 17:53:41,442 INFO: Network [SwinIRMultiHead] is created.
+2025-11-01 17:53:41,509 INFO: Loss [L1Loss] is created.
+2025-11-01 17:53:41,510 INFO: Initialized l1_latent_x2_opt in latent space (w=1.0).
+2025-11-01 17:53:41,511 INFO: Loss [L1Loss] is created.
+2025-11-01 17:53:41,511 INFO: Initialized l1_latent_x4_opt in latent space (w=1.0).
+2025-11-01 17:53:41,511 INFO: Loss [FFTFrequencyLoss] is created.
+2025-11-01 17:53:41,512 INFO: Initialized fft_latent_x2_opt in latent space (w=0.1).
+2025-11-01 17:53:41,513 INFO: Loss [FFTFrequencyLoss] is created.
+2025-11-01 17:53:41,514 INFO: Initialized fft_latent_x4_opt in latent space (w=0.1).
+2025-11-01 17:53:41,516 INFO: Gradient clipping enabled (G:norm(max_norm=0.4, norm_type=2.0)).
+2025-11-01 17:53:41,517 INFO: Model [SwinIRLatentModelMultiHead] is created.
+2025-11-01 17:53:42,378 INFO: Start training from epoch: 0, step: 0

01_11_2025/32_7/basicsr_options.yaml

@@ -0,0 +1,220 @@
+# GENERATE TIME: Sat Nov  1 18:37:02 2025
+# CMD:
+# train_vae.py -opt /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/basicsr_options.yaml
+
+model_type: SwinIRLatentModelMultiHead
+primary_head: x4
+scale: 4
+num_gpu: auto
+manual_seed: 0
+find_unused_parameters: false
+vae_sources:
+  flux_vae:
+    hf_repo: wolfgangblack/flux_vae
+    vae_kind: kl
+datasets:
+  train:
+    name: gpt4_nanobana_midjourney_recraft_CropsFluxVAE
+    type: MultiScaleLatentCacheDataset
+    scales:
+    - 128
+    - 256
+    - 512
+    cache_dirs:
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/nano_banana_crops/embeddings/flux_vae
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/midjourney_full_dataset_crops_new/embeddings/flux_vae
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/gpt4_crops/embeddings/flux_vae
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/recraft_data_crops/embeddings/flux_vae
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/mjv5_crops_fix/embeddings/flux_vae
+    vae_names:
+    - flux_vae
+    phase: train
+    filename_tmpl: '{}'
+    io_backend:
+      type: disk
+    scale: 4
+    mean: null
+    std: null
+    num_worker_per_gpu: 32
+    batch_size_per_gpu: 64
+    pin_memory: true
+    persistent_workers: true
+  val:
+    name: sdxk_120_1024x1024
+    type: MultiScaleLatentCacheDataset
+    scales:
+    - 256
+    - 512
+    - 1024
+    cache_dirs:
+    - /data/kazanplova/datasets/latent_upscale_validation_120_samples/embeddings/flux_vae
+    vae_names:
+    - flux_vae
+    phase: val
+    io_backend:
+      type: disk
+    scale: 4
+    mean: null
+    std: null
+    batch_size_per_gpu: 16
+    num_worker_per_gpu: 4
+    pin_memory: true
+network_g:
+  type: SwinIRMultiHead
+  in_chans: 16
+  img_size: 32
+  window_size: 16
+  img_range: 1.0
+  depths:
+  - 6
+  - 6
+  - 6
+  - 6
+  - 6
+  - 6
+  embed_dim: 360
+  num_heads:
+  - 12
+  - 12
+  - 12
+  - 12
+  - 12
+  - 12
+  mlp_ratio: 2
+  resi_connection: 1conv
+  primary_head: x4
+  head_num_feat: 256
+  heads:
+  - name: x2
+    scale: 2
+    out_chans: 16
+  - name: x4
+    scale: 4
+    out_chans: 16
+    primary: true
+compile:
+  enabled: false
+  mode: max-autotune
+  dynamic: true
+  fullgraph: false
+  backend: null
+train:
+  ema_decay: 0.999
+  head_inputs:
+    x2:
+      lq: 256
+      gt: 512
+    x4:
+      lq: 128
+      gt: 512
+  optim_g:
+    type: Adam
+    lr: 0.0002
+    weight_decay: 0
+    betas:
+    - 0.9
+    - 0.995
+  grad_clip:
+    enabled: true
+    generator:
+      type: norm
+      max_norm: 0.4
+      norm_type: 2.0
+  scheduler:
+    type: MultiStepLR
+    milestones:
+    - 62500
+    - 93750
+    - 112500
+    gamma: 0.5
+  total_steps: 125000
+  warmup_iter: -1
+  l1_latent_x2_opt:
+    type: L1Loss
+    loss_weight: 1.0
+    reduction: mean
+    space: latent
+    target: x2
+  l1_latent_x4_opt:
+    type: L1Loss
+    loss_weight: 1.0
+    reduction: mean
+    space: latent
+    target: x4
+  fft_latent_x2_opt:
+    type: FFTFrequencyLoss
+    loss_weight: 0.1
+    reduction: mean
+    space: latent
+    target: x2
+    norm: ortho
+    use_log_amplitude: false
+    alpha: 0.0
+    normalize_weight: true
+    eps: 1e-8
+  fft_latent_x4_opt:
+    type: FFTFrequencyLoss
+    loss_weight: 0.1
+    reduction: mean
+    space: latent
+    target: x4
+    norm: ortho
+    use_log_amplitude: false
+    alpha: 0.0
+    normalize_weight: true
+    eps: 1e-8
+val:
+  val_freq: 5000
+  save_img: true
+  head_evals:
+    x2:
+      save_img: true
+      label: val_x2
+      val_sizes:
+        lq: 512
+        gt: 1024
+      metrics:
+        l1_latent:
+          type: L1Loss
+          space: latent
+        pixel_psnr_pt:
+          type: calculate_psnr_pt
+          space: pixel
+          crop_border: 2
+          test_y_channel: false
+    x4:
+      save_img: true
+      label: val_x4
+      val_sizes:
+        lq: 256
+        gt: 1024
+      metrics:
+        l1_latent:
+          type: L1Loss
+          space: latent
+        l2_latent:
+          type: MSELoss
+          space: latent
+        pixel_psnr_pt:
+          type: calculate_psnr_pt
+          space: pixel
+          crop_border: 2
+          test_y_channel: false
+logger:
+  print_freq: 100
+  save_checkpoint_freq: 5000
+  use_tb_logger: true
+  wandb:
+    project: Swin2SR-Latent-SR
+    entity: kazanplova-it-more
+    resume_id: null
+    max_val_images: 10
+dist_params:
+  backend: nccl
+  port: 29500
+  dist: true
+load_networks_only: false
+exp_name: '32'
+name: '32_7'
+path:
+  experiments_root: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025

01_11_2025/32_7/train_32_7_20251101_183702.log

@@ -0,0 +1,571 @@
+2025-11-01 18:37:02,704 INFO: 
+                ____                _       _____  ____
+               / __ ) ____ _ _____ (_)_____/ ___/ / __ \
+              / __  |/ __ `// ___// // ___/\__ \ / /_/ /
+             / /_/ // /_/ /(__  )/ // /__ ___/ // _, _/
+            /_____/ \__,_//____//_/ \___//____//_/ |_|
+     ______                   __   __                 __      __
+    / ____/____   ____   ____/ /  / /   __  __ _____ / /__   / /
+   / / __ / __ \ / __ \ / __  /  / /   / / / // ___// //_/  / /
+  / /_/ // /_/ // /_/ // /_/ /  / /___/ /_/ // /__ / /<    /_/
+  \____/ \____/ \____/ \____/  /_____/\____/ \___//_/|_|  (_)
+    
+Version Information: 
+	BasicSR: 1.4.2
+	PyTorch: 2.9.0+cu129
+	TorchVision: 0.24.0+cpu
+2025-11-01 18:37:02,705 INFO: 
+  model_type: SwinIRLatentModelMultiHead
+  primary_head: x4
+  scale: 4
+  num_gpu: 1
+  manual_seed: 0
+  find_unused_parameters: False
+  vae_sources:[
+    flux_vae:[
+      hf_repo: wolfgangblack/flux_vae
+      vae_kind: kl
+    ]
+  ]
+  datasets:[
+    train:[
+      name: gpt4_nanobana_midjourney_recraft_CropsFluxVAE
+      type: MultiScaleLatentCacheDataset
+      scales: [128, 256, 512]
+      cache_dirs: ['/data/kazanplova/datasets/full_latent_upscale_dataset_train/nano_banana_crops/embeddings/flux_vae', '/data/kazanplova/datasets/full_latent_upscale_dataset_train/midjourney_full_dataset_crops_new/embeddings/flux_vae', '/data/kazanplova/datasets/full_latent_upscale_dataset_train/gpt4_crops/embeddings/flux_vae', '/data/kazanplova/datasets/full_latent_upscale_dataset_train/recraft_data_crops/embeddings/flux_vae', '/data/kazanplova/datasets/full_latent_upscale_dataset_train/mjv5_crops_fix/embeddings/flux_vae']
+      vae_names: ['flux_vae']
+      phase: train
+      filename_tmpl: {}
+      io_backend:[
+        type: disk
+      ]
+      scale: 4
+      mean: None
+      std: None
+      num_worker_per_gpu: 32
+      batch_size_per_gpu: 64
+      pin_memory: True
+      persistent_workers: True
+    ]
+    val:[
+      name: sdxk_120_1024x1024
+      type: MultiScaleLatentCacheDataset
+      scales: [256, 512, 1024]
+      cache_dirs: ['/data/kazanplova/datasets/latent_upscale_validation_120_samples/embeddings/flux_vae']
+      vae_names: ['flux_vae']
+      phase: val
+      io_backend:[
+        type: disk
+      ]
+      scale: 4
+      mean: None
+      std: None
+      batch_size_per_gpu: 16
+      num_worker_per_gpu: 4
+      pin_memory: True
+    ]
+  ]
+  network_g:[
+    type: SwinIRMultiHead
+    in_chans: 16
+    img_size: 32
+    window_size: 16
+    img_range: 1.0
+    depths: [6, 6, 6, 6, 6, 6]
+    embed_dim: 360
+    num_heads: [12, 12, 12, 12, 12, 12]
+    mlp_ratio: 2
+    resi_connection: 1conv
+    primary_head: x4
+    head_num_feat: 256
+    heads: [OrderedDict({'name': 'x2', 'scale': 2, 'out_chans': 16}), OrderedDict({'name': 'x4', 'scale': 4, 'out_chans': 16, 'primary': True})]
+  ]
+  compile:[
+    enabled: False
+    mode: max-autotune
+    dynamic: True
+    fullgraph: False
+    backend: None
+  ]
+  train:[
+    ema_decay: 0.999
+    head_inputs:[
+      x2:[
+        lq: 256
+        gt: 512
+      ]
+      x4:[
+        lq: 128
+        gt: 512
+      ]
+    ]
+    optim_g:[
+      type: Adam
+      lr: 0.0002
+      weight_decay: 0
+      betas: [0.9, 0.995]
+    ]
+    grad_clip:[
+      enabled: True
+      generator:[
+        type: norm
+        max_norm: 0.4
+        norm_type: 2.0
+      ]
+    ]
+    scheduler:[
+      type: MultiStepLR
+      milestones: [62500, 93750, 112500]
+      gamma: 0.5
+    ]
+    total_steps: 125000
+    warmup_iter: -1
+    l1_latent_x2_opt:[
+      type: L1Loss
+      loss_weight: 1.0
+      reduction: mean
+      space: latent
+      target: x2
+    ]
+    l1_latent_x4_opt:[
+      type: L1Loss
+      loss_weight: 1.0
+      reduction: mean
+      space: latent
+      target: x4
+    ]
+    fft_latent_x2_opt:[
+      type: FFTFrequencyLoss
+      loss_weight: 0.1
+      reduction: mean
+      space: latent
+      target: x2
+      norm: ortho
+      use_log_amplitude: False
+      alpha: 0.0
+      normalize_weight: True
+      eps: 1e-8
+    ]
+    fft_latent_x4_opt:[
+      type: FFTFrequencyLoss
+      loss_weight: 0.1
+      reduction: mean
+      space: latent
+      target: x4
+      norm: ortho
+      use_log_amplitude: False
+      alpha: 0.0
+      normalize_weight: True
+      eps: 1e-8
+    ]
+  ]
+  val:[
+    val_freq: 5000
+    save_img: True
+    head_evals:[
+      x2:[
+        save_img: True
+        label: val_x2
+        val_sizes:[
+          lq: 512
+          gt: 1024
+        ]
+        metrics:[
+          l1_latent:[
+            type: L1Loss
+            space: latent
+          ]
+          pixel_psnr_pt:[
+            type: calculate_psnr_pt
+            space: pixel
+            crop_border: 2
+            test_y_channel: False
+          ]
+        ]
+      ]
+      x4:[
+        save_img: True
+        label: val_x4
+        val_sizes:[
+          lq: 256
+          gt: 1024
+        ]
+        metrics:[
+          l1_latent:[
+            type: L1Loss
+            space: latent
+          ]
+          l2_latent:[
+            type: MSELoss
+            space: latent
+          ]
+          pixel_psnr_pt:[
+            type: calculate_psnr_pt
+            space: pixel
+            crop_border: 2
+            test_y_channel: False
+          ]
+        ]
+      ]
+    ]
+  ]
+  logger:[
+    print_freq: 100
+    save_checkpoint_freq: 5000
+    use_tb_logger: True
+    wandb:[
+      project: Swin2SR-Latent-SR
+      entity: kazanplova-it-more
+      resume_id: None
+      max_val_images: 10
+    ]
+  ]
+  dist_params:[
+    backend: nccl
+    port: 29500
+    dist: True
+  ]
+  load_networks_only: False
+  exp_name: 32
+  name: 32_7
+  path:[
+    experiments_root: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/32_7
+    models: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/32_7/models
+    training_states: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/32_7/training_states
+    log: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/32_7
+    visualization: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/32_7/visualization
+  ]
+  dist: False
+  rank: 0
+  world_size: 1
+  auto_resume: False
+  is_train: True
+  root_path: /data/kazanplova/latent_vae_upscale_train
+
+2025-11-01 18:37:04,406 INFO: Use wandb logger with id=1x0wne5b; project=Swin2SR-Latent-SR.
+2025-11-01 18:37:16,610 INFO: Dataset [MultiScaleLatentCacheDataset] - gpt4_nanobana_midjourney_recraft_CropsFluxVAE is built.
+2025-11-01 18:37:16,611 INFO: Training statistics:
+	Number of train images: 4858507
+	Dataset enlarge ratio: 1
+	Batch size per gpu: 64
+	World size (gpu number): 1
+	Steps per epoch: 75915
+	Configured training steps: 125000
+	Approximate epochs to cover: 2.
+2025-11-01 18:37:16,614 INFO: Dataset [MultiScaleLatentCacheDataset] - sdxk_120_1024x1024 is built.
+2025-11-01 18:37:16,614 INFO: Number of val images/folders in sdxk_120_1024x1024: 153
+2025-11-01 18:37:16,615 INFO: Enabled find_unused_parameters=True for multi-head training overrides.
+2025-11-01 18:37:17,515 INFO: Network [SwinIRMultiHead] is created.
+2025-11-01 18:37:17,724 INFO: Network: SwinIRMultiHead, with parameters: 54,917,584
+2025-11-01 18:37:17,725 INFO: SwinIRMultiHead(
+  (conv_first): Conv2d(16, 360, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+  (patch_embed): PatchEmbed(
+    (norm): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+  )
+  (patch_unembed): PatchUnEmbed()
+  (pos_drop): Dropout(p=0.0, inplace=False)
+  (layers): ModuleList(
+    (0): RSTB(
+      (residual_group): BasicLayer(
+        dim=360, input_resolution=(32, 32), depth=6
+        (blocks): ModuleList(
+          (0): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=0, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): Identity()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (1): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=8, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (2): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=0, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (3): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=8, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (4): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=0, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (5): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=8, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+        )
+      )
+      (conv): Conv2d(360, 360, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+      (patch_embed): PatchEmbed()
+      (patch_unembed): PatchUnEmbed()
+    )
+    (1-5): 5 x RSTB(
+      (residual_group): BasicLayer(
+        dim=360, input_resolution=(32, 32), depth=6
+        (blocks): ModuleList(
+          (0): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=0, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (1): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=8, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (2): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=0, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (3): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=8, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (4): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=0, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (5): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=8, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+        )
+      )
+      (conv): Conv2d(360, 360, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+      (patch_embed): PatchEmbed()
+      (patch_unembed): PatchUnEmbed()
+    )
+  )
+  (norm): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+  (conv_after_body): Conv2d(360, 360, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+  (heads): ModuleDict(
+    (x2): _SwinIRPixelShuffleHead(
+      (conv_before): Sequential(
+        (0): Conv2d(360, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        (1): LeakyReLU(negative_slope=0.01, inplace=True)
+      )
+      (upsample): Upsample(
+        (0): Conv2d(256, 1024, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        (1): PixelShuffle(upscale_factor=2)
+      )
+      (conv_last): Conv2d(256, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+    )
+    (x4): _SwinIRPixelShuffleHead(
+      (conv_before): Sequential(
+        (0): Conv2d(360, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        (1): LeakyReLU(negative_slope=0.01, inplace=True)
+      )
+      (upsample): Upsample(
+        (0): Conv2d(256, 1024, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        (1): PixelShuffle(upscale_factor=2)
+        (2): Conv2d(256, 1024, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        (3): PixelShuffle(upscale_factor=2)
+      )
+      (conv_last): Conv2d(256, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+    )
+  )
+)
+2025-11-01 18:37:17,728 INFO: Use EMA with decay: 0.999
+2025-11-01 18:37:18,275 INFO: Network [SwinIRMultiHead] is created.
+2025-11-01 18:37:18,341 INFO: Loss [L1Loss] is created.
+2025-11-01 18:37:18,342 INFO: Initialized l1_latent_x2_opt in latent space (w=1.0).
+2025-11-01 18:37:18,343 INFO: Loss [L1Loss] is created.
+2025-11-01 18:37:18,343 INFO: Initialized l1_latent_x4_opt in latent space (w=1.0).
+2025-11-01 18:37:18,343 INFO: Loss [FFTFrequencyLoss] is created.
+2025-11-01 18:37:18,344 INFO: Initialized fft_latent_x2_opt in latent space (w=0.1).
+2025-11-01 18:37:18,345 INFO: Loss [FFTFrequencyLoss] is created.
+2025-11-01 18:37:18,346 INFO: Initialized fft_latent_x4_opt in latent space (w=0.1).
+2025-11-01 18:37:18,347 INFO: Gradient clipping enabled (G:norm(max_norm=0.4, norm_type=2.0)).
+2025-11-01 18:37:18,348 INFO: Model [SwinIRLatentModelMultiHead] is created.
+2025-11-01 18:37:19,528 INFO: Start training from epoch: 0, step: 0
+2025-11-01 18:39:15,803 INFO: [32_7..][epoch:  0, step:     100, lr:(2.000e-04,)] [eta: 1 day, 13:56:24, time (data): 1.163 (0.022)] l1_latent_x2_opt: 9.2823e-01 fft_latent_x2_opt: 8.1074e-01 l1_latent_x4_opt: 1.0574e+00 fft_latent_x4_opt: 9.1827e-01 

01_11_2025/32_archived_20251101_095943/basicsr_options.yaml

@@ -0,0 +1,205 @@
+# GENERATE TIME: Sat Nov  1 09:58:39 2025
+# CMD:
+# train_vae.py -opt /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/basicsr_options.yaml --launcher pytorch --local_rank 0
+
+model_type: SwinIRLatentModelMultiHead
+primary_head: x4
+scale: 4
+num_gpu: auto
+manual_seed: 0
+find_unused_parameters: false
+vae_sources:
+  flux_vae:
+    hf_repo: wolfgangblack/flux_vae
+    vae_kind: kl
+datasets:
+  train:
+    name: gpt4_nanobana_midjourney_recraft_CropsFluxVAE
+    type: MultiScaleLatentCacheDataset
+    scales:
+    - 128
+    - 256
+    - 512
+    cache_dirs:
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/nano_banana_crops/embeddings/flux_vae
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/midjourney_full_dataset_crops_new/embeddings/flux_vae
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/gpt4_crops/embeddings/flux_vae
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/recraft_data_crops/embeddings/flux_vae
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/mjv5_crops_fix/embeddings/flux_vae
+    vae_names:
+    - flux_vae
+    phase: train
+    filename_tmpl: '{}'
+    io_backend:
+      type: disk
+    scale: 4
+    mean: null
+    std: null
+    num_worker_per_gpu: 32
+    batch_size_per_gpu: 256
+    pin_memory: true
+    persistent_workers: true
+  val:
+    name: sdxk_120_1024x1024
+    type: MultiScaleLatentCacheDataset
+    scales:
+    - 256
+    - 512
+    - 1024
+    cache_dirs:
+    - /data/kazanplova/datasets/latent_upscale_validation_120_samples/embeddings/flux_vae
+    vae_names:
+    - flux_vae
+    phase: val
+    io_backend:
+      type: disk
+    scale: 4
+    mean: null
+    std: null
+    batch_size_per_gpu: 16
+    num_worker_per_gpu: 4
+    pin_memory: true
+network_g:
+  type: SwinIRMultiHead
+  in_chans: 16
+  img_size: 32
+  window_size: 16
+  img_range: 1.0
+  depths:
+  - 6
+  - 6
+  - 6
+  - 6
+  - 6
+  - 6
+  embed_dim: 360
+  num_heads:
+  - 12
+  - 12
+  - 12
+  - 12
+  - 12
+  - 12
+  mlp_ratio: 2
+  resi_connection: 1conv
+  primary_head: x4
+  head_num_feat: 256
+  heads:
+  - name: x2
+    scale: 2
+    out_chans: 16
+  - name: x4
+    scale: 4
+    out_chans: 16
+    primary: true
+compile:
+  enabled: false
+  mode: max-autotune
+  dynamic: true
+  fullgraph: false
+  backend: null
+train:
+  ema_decay: 0.999
+  optim_g:
+    type: Adam
+    lr: 0.00015
+    weight_decay: 0
+    betas:
+    - 0.9
+    - 0.995
+  grad_clip:
+    enabled: true
+    generator:
+      type: norm
+      max_norm: 0.4
+      norm_type: 2.0
+  scheduler:
+    type: MultiStepLR
+    milestones:
+    - 75000
+    - 90000
+    - 110000
+    gamma: 0.5
+  total_steps: 125000
+  warmup_iter: -1
+  l1_latent_x2_opt:
+    type: L1Loss
+    loss_weight: 0.5
+    reduction: mean
+    space: latent
+    target: x2
+  fft_latent_x2_opt:
+    type: FFTFrequencyLoss
+    loss_weight: 0.1
+    reduction: mean
+    space: latent
+    target: x2
+    norm: ortho
+    use_log_amplitude: false
+    alpha: 0.0
+    normalize_weight: true
+  l1_latent_x4_opt:
+    type: L1Loss
+    loss_weight: 0.5
+    reduction: mean
+    space: latent
+    target: x4
+  fft_latent_x4_opt:
+    type: FFTFrequencyLoss
+    loss_weight: 0.1
+    reduction: mean
+    space: latent
+    target: x4
+    norm: ortho
+    use_log_amplitude: false
+    alpha: 0.0
+    normalize_weight: true
+val:
+  val_freq: 5000
+  save_img: true
+  head_evals:
+    x2:
+      save_img: true
+      label: val_x2
+      metrics:
+        l1_latent:
+          type: L1Loss
+          space: latent
+        pixel_psnr_pt:
+          type: calculate_psnr_pt
+          space: pixel
+          crop_border: 2
+          test_y_channel: false
+    x4:
+      save_img: true
+      label: val_x4
+      metrics:
+        l1_latent:
+          type: L1Loss
+          space: latent
+        l2_latent:
+          type: MSELoss
+          space: latent
+        pixel_psnr_pt:
+          type: calculate_psnr_pt
+          space: pixel
+          crop_border: 2
+          test_y_channel: false
+logger:
+  print_freq: 100
+  save_checkpoint_freq: 5000
+  use_tb_logger: true
+  wandb:
+    project: Swin2SR-Latent-SR
+    entity: kazanplova-it-more
+    resume_id: null
+    max_val_images: 10
+dist_params:
+  backend: nccl
+  port: 29500
+  dist: true
+load_networks_only: false
+exp_name: '32'
+name: '32'
+path:
+  experiments_root: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025

01_11_2025/32_archived_20251101_095943/train_32_20251101_095839.log

@@ -0,0 +1,548 @@
+2025-11-01 09:58:39,616 INFO: 
+                ____                _       _____  ____
+               / __ ) ____ _ _____ (_)_____/ ___/ / __ \
+              / __  |/ __ `// ___// // ___/\__ \ / /_/ /
+             / /_/ // /_/ /(__  )/ // /__ ___/ // _, _/
+            /_____/ \__,_//____//_/ \___//____//_/ |_|
+     ______                   __   __                 __      __
+    / ____/____   ____   ____/ /  / /   __  __ _____ / /__   / /
+   / / __ / __ \ / __ \ / __  /  / /   / / / // ___// //_/  / /
+  / /_/ // /_/ // /_/ // /_/ /  / /___/ /_/ // /__ / /<    /_/
+  \____/ \____/ \____/ \____/  /_____/\____/ \___//_/|_|  (_)
+    
+Version Information: 
+	BasicSR: 1.4.2
+	PyTorch: 2.9.0+cu129
+	TorchVision: 0.24.0+cpu
+2025-11-01 09:58:39,616 INFO: 
+  model_type: SwinIRLatentModelMultiHead
+  primary_head: x4
+  scale: 4
+  num_gpu: 3
+  manual_seed: 0
+  find_unused_parameters: False
+  vae_sources:[
+    flux_vae:[
+      hf_repo: wolfgangblack/flux_vae
+      vae_kind: kl
+    ]
+  ]
+  datasets:[
+    train:[
+      name: gpt4_nanobana_midjourney_recraft_CropsFluxVAE
+      type: MultiScaleLatentCacheDataset
+      scales: [128, 256, 512]
+      cache_dirs: ['/data/kazanplova/datasets/full_latent_upscale_dataset_train/nano_banana_crops/embeddings/flux_vae', '/data/kazanplova/datasets/full_latent_upscale_dataset_train/midjourney_full_dataset_crops_new/embeddings/flux_vae', '/data/kazanplova/datasets/full_latent_upscale_dataset_train/gpt4_crops/embeddings/flux_vae', '/data/kazanplova/datasets/full_latent_upscale_dataset_train/recraft_data_crops/embeddings/flux_vae', '/data/kazanplova/datasets/full_latent_upscale_dataset_train/mjv5_crops_fix/embeddings/flux_vae']
+      vae_names: ['flux_vae']
+      phase: train
+      filename_tmpl: {}
+      io_backend:[
+        type: disk
+      ]
+      scale: 4
+      mean: None
+      std: None
+      num_worker_per_gpu: 32
+      batch_size_per_gpu: 256
+      pin_memory: True
+      persistent_workers: True
+    ]
+    val:[
+      name: sdxk_120_1024x1024
+      type: MultiScaleLatentCacheDataset
+      scales: [256, 512, 1024]
+      cache_dirs: ['/data/kazanplova/datasets/latent_upscale_validation_120_samples/embeddings/flux_vae']
+      vae_names: ['flux_vae']
+      phase: val
+      io_backend:[
+        type: disk
+      ]
+      scale: 4
+      mean: None
+      std: None
+      batch_size_per_gpu: 16
+      num_worker_per_gpu: 4
+      pin_memory: True
+    ]
+  ]
+  network_g:[
+    type: SwinIRMultiHead
+    in_chans: 16
+    img_size: 32
+    window_size: 16
+    img_range: 1.0
+    depths: [6, 6, 6, 6, 6, 6]
+    embed_dim: 360
+    num_heads: [12, 12, 12, 12, 12, 12]
+    mlp_ratio: 2
+    resi_connection: 1conv
+    primary_head: x4
+    head_num_feat: 256
+    heads: [OrderedDict({'name': 'x2', 'scale': 2, 'out_chans': 16}), OrderedDict({'name': 'x4', 'scale': 4, 'out_chans': 16, 'primary': True})]
+  ]
+  compile:[
+    enabled: False
+    mode: max-autotune
+    dynamic: True
+    fullgraph: False
+    backend: None
+  ]
+  train:[
+    ema_decay: 0.999
+    optim_g:[
+      type: Adam
+      lr: 0.00015
+      weight_decay: 0
+      betas: [0.9, 0.995]
+    ]
+    grad_clip:[
+      enabled: True
+      generator:[
+        type: norm
+        max_norm: 0.4
+        norm_type: 2.0
+      ]
+    ]
+    scheduler:[
+      type: MultiStepLR
+      milestones: [75000, 90000, 110000]
+      gamma: 0.5
+    ]
+    total_steps: 125000
+    warmup_iter: -1
+    l1_latent_x2_opt:[
+      type: L1Loss
+      loss_weight: 0.5
+      reduction: mean
+      space: latent
+      target: x2
+    ]
+    fft_latent_x2_opt:[
+      type: FFTFrequencyLoss
+      loss_weight: 0.1
+      reduction: mean
+      space: latent
+      target: x2
+      norm: ortho
+      use_log_amplitude: False
+      alpha: 0.0
+      normalize_weight: True
+    ]
+    l1_latent_x4_opt:[
+      type: L1Loss
+      loss_weight: 0.5
+      reduction: mean
+      space: latent
+      target: x4
+    ]
+    fft_latent_x4_opt:[
+      type: FFTFrequencyLoss
+      loss_weight: 0.1
+      reduction: mean
+      space: latent
+      target: x4
+      norm: ortho
+      use_log_amplitude: False
+      alpha: 0.0
+      normalize_weight: True
+    ]
+  ]
+  val:[
+    val_freq: 5000
+    save_img: True
+    head_evals:[
+      x2:[
+        save_img: True
+        label: val_x2
+        metrics:[
+          l1_latent:[
+            type: L1Loss
+            space: latent
+          ]
+          pixel_psnr_pt:[
+            type: calculate_psnr_pt
+            space: pixel
+            crop_border: 2
+            test_y_channel: False
+          ]
+        ]
+      ]
+      x4:[
+        save_img: True
+        label: val_x4
+        metrics:[
+          l1_latent:[
+            type: L1Loss
+            space: latent
+          ]
+          l2_latent:[
+            type: MSELoss
+            space: latent
+          ]
+          pixel_psnr_pt:[
+            type: calculate_psnr_pt
+            space: pixel
+            crop_border: 2
+            test_y_channel: False
+          ]
+        ]
+      ]
+    ]
+  ]
+  logger:[
+    print_freq: 100
+    save_checkpoint_freq: 5000
+    use_tb_logger: True
+    wandb:[
+      project: Swin2SR-Latent-SR
+      entity: kazanplova-it-more
+      resume_id: None
+      max_val_images: 10
+    ]
+  ]
+  dist_params:[
+    backend: nccl
+    port: 29500
+    dist: True
+  ]
+  load_networks_only: False
+  exp_name: 32
+  name: 32
+  path:[
+    experiments_root: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/32
+    models: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/32/models
+    training_states: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/32/training_states
+    log: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/32
+    visualization: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/32/visualization
+  ]
+  dist: True
+  rank: 0
+  world_size: 3
+  auto_resume: False
+  is_train: True
+  root_path: /data/kazanplova/latent_vae_upscale_train
+
+2025-11-01 09:58:41,357 INFO: Use wandb logger with id=st3iv9sy; project=Swin2SR-Latent-SR.
+2025-11-01 09:58:54,358 INFO: Dataset [MultiScaleLatentCacheDataset] - gpt4_nanobana_midjourney_recraft_CropsFluxVAE is built.
+2025-11-01 09:58:54,359 INFO: Training statistics:
+	Number of train images: 4858507
+	Dataset enlarge ratio: 1
+	Batch size per gpu: 256
+	World size (gpu number): 3
+	Steps per epoch: 6327
+	Configured training steps: 125000
+	Approximate epochs to cover: 20.
+2025-11-01 09:58:54,362 INFO: Dataset [MultiScaleLatentCacheDataset] - sdxk_120_1024x1024 is built.
+2025-11-01 09:58:54,363 INFO: Number of val images/folders in sdxk_120_1024x1024: 153
+2025-11-01 09:58:54,831 INFO: Network [SwinIRMultiHead] is created.
+2025-11-01 09:58:56,417 INFO: Network: DistributedDataParallel - SwinIRMultiHead, with parameters: 54,917,584
+2025-11-01 09:58:56,418 INFO: SwinIRMultiHead(
+  (conv_first): Conv2d(16, 360, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+  (patch_embed): PatchEmbed(
+    (norm): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+  )
+  (patch_unembed): PatchUnEmbed()
+  (pos_drop): Dropout(p=0.0, inplace=False)
+  (layers): ModuleList(
+    (0): RSTB(
+      (residual_group): BasicLayer(
+        dim=360, input_resolution=(32, 32), depth=6
+        (blocks): ModuleList(
+          (0): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=0, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): Identity()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (1): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=8, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (2): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=0, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (3): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=8, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (4): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=0, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (5): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=8, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+        )
+      )
+      (conv): Conv2d(360, 360, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+      (patch_embed): PatchEmbed()
+      (patch_unembed): PatchUnEmbed()
+    )
+    (1-5): 5 x RSTB(
+      (residual_group): BasicLayer(
+        dim=360, input_resolution=(32, 32), depth=6
+        (blocks): ModuleList(
+          (0): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=0, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (1): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=8, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (2): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=0, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (3): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=8, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (4): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=0, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (5): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=8, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+        )
+      )
+      (conv): Conv2d(360, 360, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+      (patch_embed): PatchEmbed()
+      (patch_unembed): PatchUnEmbed()
+    )
+  )
+  (norm): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+  (conv_after_body): Conv2d(360, 360, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+  (heads): ModuleDict(
+    (x2): _SwinIRPixelShuffleHead(
+      (conv_before): Sequential(
+        (0): Conv2d(360, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        (1): LeakyReLU(negative_slope=0.01, inplace=True)
+      )
+      (upsample): Upsample(
+        (0): Conv2d(256, 1024, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        (1): PixelShuffle(upscale_factor=2)
+      )
+      (conv_last): Conv2d(256, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+    )
+    (x4): _SwinIRPixelShuffleHead(
+      (conv_before): Sequential(
+        (0): Conv2d(360, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        (1): LeakyReLU(negative_slope=0.01, inplace=True)
+      )
+      (upsample): Upsample(
+        (0): Conv2d(256, 1024, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        (1): PixelShuffle(upscale_factor=2)
+        (2): Conv2d(256, 1024, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        (3): PixelShuffle(upscale_factor=2)
+      )
+      (conv_last): Conv2d(256, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+    )
+  )
+)
+2025-11-01 09:58:56,421 INFO: Use EMA with decay: 0.999
+2025-11-01 09:58:56,865 INFO: Network [SwinIRMultiHead] is created.
+2025-11-01 09:58:56,933 INFO: Loss [L1Loss] is created.
+2025-11-01 09:58:56,934 INFO: Initialized l1_latent_x2_opt in latent space (w=0.5).
+2025-11-01 09:58:56,935 INFO: Loss [FFTFrequencyLoss] is created.
+2025-11-01 09:58:56,936 INFO: Initialized fft_latent_x2_opt in latent space (w=0.1).
+2025-11-01 09:58:56,937 INFO: Loss [L1Loss] is created.
+2025-11-01 09:58:56,937 INFO: Initialized l1_latent_x4_opt in latent space (w=0.5).
+2025-11-01 09:58:56,938 INFO: Loss [FFTFrequencyLoss] is created.
+2025-11-01 09:58:56,939 INFO: Initialized fft_latent_x4_opt in latent space (w=0.1).
+2025-11-01 09:58:56,942 INFO: Gradient clipping enabled (G:norm(max_norm=0.4, norm_type=2.0)).
+2025-11-01 09:58:56,942 INFO: Model [SwinIRLatentModelMultiHead] is created.

01_11_2025/32_archived_20251101_104933/basicsr_options.yaml

@@ -0,0 +1,205 @@
+# GENERATE TIME: Sat Nov  1 09:59:43 2025
+# CMD:
+# train_vae.py -opt /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/basicsr_options.yaml --launcher pytorch --local_rank 0
+
+model_type: SwinIRLatentModelMultiHead
+primary_head: x4
+scale: 4
+num_gpu: auto
+manual_seed: 0
+find_unused_parameters: false
+vae_sources:
+  flux_vae:
+    hf_repo: wolfgangblack/flux_vae
+    vae_kind: kl
+datasets:
+  train:
+    name: gpt4_nanobana_midjourney_recraft_CropsFluxVAE
+    type: MultiScaleLatentCacheDataset
+    scales:
+    - 128
+    - 256
+    - 512
+    cache_dirs:
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/nano_banana_crops/embeddings/flux_vae
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/midjourney_full_dataset_crops_new/embeddings/flux_vae
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/gpt4_crops/embeddings/flux_vae
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/recraft_data_crops/embeddings/flux_vae
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/mjv5_crops_fix/embeddings/flux_vae
+    vae_names:
+    - flux_vae
+    phase: train
+    filename_tmpl: '{}'
+    io_backend:
+      type: disk
+    scale: 4
+    mean: null
+    std: null
+    num_worker_per_gpu: 32
+    batch_size_per_gpu: 256
+    pin_memory: true
+    persistent_workers: true
+  val:
+    name: sdxk_120_1024x1024
+    type: MultiScaleLatentCacheDataset
+    scales:
+    - 256
+    - 512
+    - 1024
+    cache_dirs:
+    - /data/kazanplova/datasets/latent_upscale_validation_120_samples/embeddings/flux_vae
+    vae_names:
+    - flux_vae
+    phase: val
+    io_backend:
+      type: disk
+    scale: 4
+    mean: null
+    std: null
+    batch_size_per_gpu: 16
+    num_worker_per_gpu: 4
+    pin_memory: true
+network_g:
+  type: SwinIRMultiHead
+  in_chans: 16
+  img_size: 32
+  window_size: 16
+  img_range: 1.0
+  depths:
+  - 6
+  - 6
+  - 6
+  - 6
+  - 6
+  - 6
+  embed_dim: 360
+  num_heads:
+  - 12
+  - 12
+  - 12
+  - 12
+  - 12
+  - 12
+  mlp_ratio: 2
+  resi_connection: 1conv
+  primary_head: x4
+  head_num_feat: 256
+  heads:
+  - name: x2
+    scale: 2
+    out_chans: 16
+  - name: x4
+    scale: 4
+    out_chans: 16
+    primary: true
+compile:
+  enabled: false
+  mode: max-autotune
+  dynamic: true
+  fullgraph: false
+  backend: null
+train:
+  ema_decay: 0.999
+  optim_g:
+    type: Adam
+    lr: 0.00015
+    weight_decay: 0
+    betas:
+    - 0.9
+    - 0.995
+  grad_clip:
+    enabled: true
+    generator:
+      type: norm
+      max_norm: 0.4
+      norm_type: 2.0
+  scheduler:
+    type: MultiStepLR
+    milestones:
+    - 75000
+    - 90000
+    - 110000
+    gamma: 0.5
+  total_steps: 125000
+  warmup_iter: -1
+  l1_latent_x2_opt:
+    type: L1Loss
+    loss_weight: 0.5
+    reduction: mean
+    space: latent
+    target: x2
+  fft_latent_x2_opt:
+    type: FFTFrequencyLoss
+    loss_weight: 0.1
+    reduction: mean
+    space: latent
+    target: x2
+    norm: ortho
+    use_log_amplitude: false
+    alpha: 0.0
+    normalize_weight: true
+  l1_latent_x4_opt:
+    type: L1Loss
+    loss_weight: 0.5
+    reduction: mean
+    space: latent
+    target: x4
+  fft_latent_x4_opt:
+    type: FFTFrequencyLoss
+    loss_weight: 0.1
+    reduction: mean
+    space: latent
+    target: x4
+    norm: ortho
+    use_log_amplitude: false
+    alpha: 0.0
+    normalize_weight: true
+val:
+  val_freq: 5000
+  save_img: true
+  head_evals:
+    x2:
+      save_img: true
+      label: val_x2
+      metrics:
+        l1_latent:
+          type: L1Loss
+          space: latent
+        pixel_psnr_pt:
+          type: calculate_psnr_pt
+          space: pixel
+          crop_border: 2
+          test_y_channel: false
+    x4:
+      save_img: true
+      label: val_x4
+      metrics:
+        l1_latent:
+          type: L1Loss
+          space: latent
+        l2_latent:
+          type: MSELoss
+          space: latent
+        pixel_psnr_pt:
+          type: calculate_psnr_pt
+          space: pixel
+          crop_border: 2
+          test_y_channel: false
+logger:
+  print_freq: 100
+  save_checkpoint_freq: 5000
+  use_tb_logger: true
+  wandb:
+    project: Swin2SR-Latent-SR
+    entity: kazanplova-it-more
+    resume_id: null
+    max_val_images: 10
+dist_params:
+  backend: nccl
+  port: 29500
+  dist: true
+load_networks_only: false
+exp_name: '32'
+name: '32'
+path:
+  experiments_root: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025

01_11_2025/32_archived_20251101_104933/train_32_20251101_095943.log

@@ -0,0 +1,548 @@
+2025-11-01 09:59:43,922 INFO: 
+                ____                _       _____  ____
+               / __ ) ____ _ _____ (_)_____/ ___/ / __ \
+              / __  |/ __ `// ___// // ___/\__ \ / /_/ /
+             / /_/ // /_/ /(__  )/ // /__ ___/ // _, _/
+            /_____/ \__,_//____//_/ \___//____//_/ |_|
+     ______                   __   __                 __      __
+    / ____/____   ____   ____/ /  / /   __  __ _____ / /__   / /
+   / / __ / __ \ / __ \ / __  /  / /   / / / // ___// //_/  / /
+  / /_/ // /_/ // /_/ // /_/ /  / /___/ /_/ // /__ / /<    /_/
+  \____/ \____/ \____/ \____/  /_____/\____/ \___//_/|_|  (_)
+    
+Version Information: 
+	BasicSR: 1.4.2
+	PyTorch: 2.9.0+cu129
+	TorchVision: 0.24.0+cpu
+2025-11-01 09:59:43,923 INFO: 
+  model_type: SwinIRLatentModelMultiHead
+  primary_head: x4
+  scale: 4
+  num_gpu: 3
+  manual_seed: 0
+  find_unused_parameters: False
+  vae_sources:[
+    flux_vae:[
+      hf_repo: wolfgangblack/flux_vae
+      vae_kind: kl
+    ]
+  ]
+  datasets:[
+    train:[
+      name: gpt4_nanobana_midjourney_recraft_CropsFluxVAE
+      type: MultiScaleLatentCacheDataset
+      scales: [128, 256, 512]
+      cache_dirs: ['/data/kazanplova/datasets/full_latent_upscale_dataset_train/nano_banana_crops/embeddings/flux_vae', '/data/kazanplova/datasets/full_latent_upscale_dataset_train/midjourney_full_dataset_crops_new/embeddings/flux_vae', '/data/kazanplova/datasets/full_latent_upscale_dataset_train/gpt4_crops/embeddings/flux_vae', '/data/kazanplova/datasets/full_latent_upscale_dataset_train/recraft_data_crops/embeddings/flux_vae', '/data/kazanplova/datasets/full_latent_upscale_dataset_train/mjv5_crops_fix/embeddings/flux_vae']
+      vae_names: ['flux_vae']
+      phase: train
+      filename_tmpl: {}
+      io_backend:[
+        type: disk
+      ]
+      scale: 4
+      mean: None
+      std: None
+      num_worker_per_gpu: 32
+      batch_size_per_gpu: 256
+      pin_memory: True
+      persistent_workers: True
+    ]
+    val:[
+      name: sdxk_120_1024x1024
+      type: MultiScaleLatentCacheDataset
+      scales: [256, 512, 1024]
+      cache_dirs: ['/data/kazanplova/datasets/latent_upscale_validation_120_samples/embeddings/flux_vae']
+      vae_names: ['flux_vae']
+      phase: val
+      io_backend:[
+        type: disk
+      ]
+      scale: 4
+      mean: None
+      std: None
+      batch_size_per_gpu: 16
+      num_worker_per_gpu: 4
+      pin_memory: True
+    ]
+  ]
+  network_g:[
+    type: SwinIRMultiHead
+    in_chans: 16
+    img_size: 32
+    window_size: 16
+    img_range: 1.0
+    depths: [6, 6, 6, 6, 6, 6]
+    embed_dim: 360
+    num_heads: [12, 12, 12, 12, 12, 12]
+    mlp_ratio: 2
+    resi_connection: 1conv
+    primary_head: x4
+    head_num_feat: 256
+    heads: [OrderedDict({'name': 'x2', 'scale': 2, 'out_chans': 16}), OrderedDict({'name': 'x4', 'scale': 4, 'out_chans': 16, 'primary': True})]
+  ]
+  compile:[
+    enabled: False
+    mode: max-autotune
+    dynamic: True
+    fullgraph: False
+    backend: None
+  ]
+  train:[
+    ema_decay: 0.999
+    optim_g:[
+      type: Adam
+      lr: 0.00015
+      weight_decay: 0
+      betas: [0.9, 0.995]
+    ]
+    grad_clip:[
+      enabled: True
+      generator:[
+        type: norm
+        max_norm: 0.4
+        norm_type: 2.0
+      ]
+    ]
+    scheduler:[
+      type: MultiStepLR
+      milestones: [75000, 90000, 110000]
+      gamma: 0.5
+    ]
+    total_steps: 125000
+    warmup_iter: -1
+    l1_latent_x2_opt:[
+      type: L1Loss
+      loss_weight: 0.5
+      reduction: mean
+      space: latent
+      target: x2
+    ]
+    fft_latent_x2_opt:[
+      type: FFTFrequencyLoss
+      loss_weight: 0.1
+      reduction: mean
+      space: latent
+      target: x2
+      norm: ortho
+      use_log_amplitude: False
+      alpha: 0.0
+      normalize_weight: True
+    ]
+    l1_latent_x4_opt:[
+      type: L1Loss
+      loss_weight: 0.5
+      reduction: mean
+      space: latent
+      target: x4
+    ]
+    fft_latent_x4_opt:[
+      type: FFTFrequencyLoss
+      loss_weight: 0.1
+      reduction: mean
+      space: latent
+      target: x4
+      norm: ortho
+      use_log_amplitude: False
+      alpha: 0.0
+      normalize_weight: True
+    ]
+  ]
+  val:[
+    val_freq: 5000
+    save_img: True
+    head_evals:[
+      x2:[
+        save_img: True
+        label: val_x2
+        metrics:[
+          l1_latent:[
+            type: L1Loss
+            space: latent
+          ]
+          pixel_psnr_pt:[
+            type: calculate_psnr_pt
+            space: pixel
+            crop_border: 2
+            test_y_channel: False
+          ]
+        ]
+      ]
+      x4:[
+        save_img: True
+        label: val_x4
+        metrics:[
+          l1_latent:[
+            type: L1Loss
+            space: latent
+          ]
+          l2_latent:[
+            type: MSELoss
+            space: latent
+          ]
+          pixel_psnr_pt:[
+            type: calculate_psnr_pt
+            space: pixel
+            crop_border: 2
+            test_y_channel: False
+          ]
+        ]
+      ]
+    ]
+  ]
+  logger:[
+    print_freq: 100
+    save_checkpoint_freq: 5000
+    use_tb_logger: True
+    wandb:[
+      project: Swin2SR-Latent-SR
+      entity: kazanplova-it-more
+      resume_id: None
+      max_val_images: 10
+    ]
+  ]
+  dist_params:[
+    backend: nccl
+    port: 29500
+    dist: True
+  ]
+  load_networks_only: False
+  exp_name: 32
+  name: 32
+  path:[
+    experiments_root: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/32
+    models: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/32/models
+    training_states: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/32/training_states
+    log: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/32
+    visualization: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/32/visualization
+  ]
+  dist: True
+  rank: 0
+  world_size: 3
+  auto_resume: False
+  is_train: True
+  root_path: /data/kazanplova/latent_vae_upscale_train
+
+2025-11-01 09:59:45,536 INFO: Use wandb logger with id=jtopgky0; project=Swin2SR-Latent-SR.
+2025-11-01 09:59:58,701 INFO: Dataset [MultiScaleLatentCacheDataset] - gpt4_nanobana_midjourney_recraft_CropsFluxVAE is built.
+2025-11-01 09:59:58,702 INFO: Training statistics:
+	Number of train images: 4858507
+	Dataset enlarge ratio: 1
+	Batch size per gpu: 256
+	World size (gpu number): 3
+	Steps per epoch: 6327
+	Configured training steps: 125000
+	Approximate epochs to cover: 20.
+2025-11-01 09:59:58,705 INFO: Dataset [MultiScaleLatentCacheDataset] - sdxk_120_1024x1024 is built.
+2025-11-01 09:59:58,706 INFO: Number of val images/folders in sdxk_120_1024x1024: 153
+2025-11-01 09:59:59,178 INFO: Network [SwinIRMultiHead] is created.
+2025-11-01 10:00:00,760 INFO: Network: DistributedDataParallel - SwinIRMultiHead, with parameters: 54,917,584
+2025-11-01 10:00:00,761 INFO: SwinIRMultiHead(
+  (conv_first): Conv2d(16, 360, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+  (patch_embed): PatchEmbed(
+    (norm): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+  )
+  (patch_unembed): PatchUnEmbed()
+  (pos_drop): Dropout(p=0.0, inplace=False)
+  (layers): ModuleList(
+    (0): RSTB(
+      (residual_group): BasicLayer(
+        dim=360, input_resolution=(32, 32), depth=6
+        (blocks): ModuleList(
+          (0): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=0, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): Identity()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (1): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=8, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (2): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=0, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (3): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=8, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (4): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=0, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (5): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=8, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+        )
+      )
+      (conv): Conv2d(360, 360, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+      (patch_embed): PatchEmbed()
+      (patch_unembed): PatchUnEmbed()
+    )
+    (1-5): 5 x RSTB(
+      (residual_group): BasicLayer(
+        dim=360, input_resolution=(32, 32), depth=6
+        (blocks): ModuleList(
+          (0): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=0, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (1): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=8, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (2): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=0, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (3): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=8, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (4): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=0, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+          (5): SwinTransformerBlock(
+            dim=360, input_resolution=(32, 32), num_heads=12, window_size=16, shift_size=8, mlp_ratio=2.0
+            (norm1): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (attn): WindowAttention(
+              dim=360, window_size=(16, 16), num_heads=12
+              (qkv): Linear(in_features=360, out_features=1080, bias=True)
+              (attn_drop): Dropout(p=0.0, inplace=False)
+              (proj): Linear(in_features=360, out_features=360, bias=True)
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (softmax): Softmax(dim=-1)
+            )
+            (drop_path): DropPath()
+            (norm2): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+            (mlp): Mlp(
+              (fc1): Linear(in_features=360, out_features=720, bias=True)
+              (act): GELU(approximate='none')
+              (fc2): Linear(in_features=720, out_features=360, bias=True)
+              (drop): Dropout(p=0.0, inplace=False)
+            )
+          )
+        )
+      )
+      (conv): Conv2d(360, 360, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+      (patch_embed): PatchEmbed()
+      (patch_unembed): PatchUnEmbed()
+    )
+  )
+  (norm): LayerNorm((360,), eps=1e-05, elementwise_affine=True)
+  (conv_after_body): Conv2d(360, 360, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+  (heads): ModuleDict(
+    (x2): _SwinIRPixelShuffleHead(
+      (conv_before): Sequential(
+        (0): Conv2d(360, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        (1): LeakyReLU(negative_slope=0.01, inplace=True)
+      )
+      (upsample): Upsample(
+        (0): Conv2d(256, 1024, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        (1): PixelShuffle(upscale_factor=2)
+      )
+      (conv_last): Conv2d(256, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+    )
+    (x4): _SwinIRPixelShuffleHead(
+      (conv_before): Sequential(
+        (0): Conv2d(360, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        (1): LeakyReLU(negative_slope=0.01, inplace=True)
+      )
+      (upsample): Upsample(
+        (0): Conv2d(256, 1024, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        (1): PixelShuffle(upscale_factor=2)
+        (2): Conv2d(256, 1024, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        (3): PixelShuffle(upscale_factor=2)
+      )
+      (conv_last): Conv2d(256, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+    )
+  )
+)
+2025-11-01 10:00:00,763 INFO: Use EMA with decay: 0.999
+2025-11-01 10:00:01,169 INFO: Network [SwinIRMultiHead] is created.
+2025-11-01 10:00:01,236 INFO: Loss [L1Loss] is created.
+2025-11-01 10:00:01,237 INFO: Initialized l1_latent_x2_opt in latent space (w=0.5).
+2025-11-01 10:00:01,239 INFO: Loss [FFTFrequencyLoss] is created.
+2025-11-01 10:00:01,239 INFO: Initialized fft_latent_x2_opt in latent space (w=0.1).
+2025-11-01 10:00:01,239 INFO: Loss [L1Loss] is created.
+2025-11-01 10:00:01,240 INFO: Initialized l1_latent_x4_opt in latent space (w=0.5).
+2025-11-01 10:00:01,241 INFO: Loss [FFTFrequencyLoss] is created.
+2025-11-01 10:00:01,242 INFO: Initialized fft_latent_x4_opt in latent space (w=0.1).
+2025-11-01 10:00:01,244 INFO: Gradient clipping enabled (G:norm(max_norm=0.4, norm_type=2.0)).
+2025-11-01 10:00:01,244 INFO: Model [SwinIRLatentModelMultiHead] is created.

01_11_2025/32_archived_20251101_161010/basicsr_options.yaml

@@ -0,0 +1,213 @@
+# GENERATE TIME: Sat Nov  1 10:49:33 2025
+# CMD:
+# train_vae.py -opt /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025/basicsr_options.yaml --launcher pytorch --local_rank 0
+
+model_type: SwinIRLatentModelMultiHead
+primary_head: x4
+scale: 4
+num_gpu: auto
+manual_seed: 0
+find_unused_parameters: false
+vae_sources:
+  flux_vae:
+    hf_repo: wolfgangblack/flux_vae
+    vae_kind: kl
+datasets:
+  train:
+    name: gpt4_nanobana_midjourney_recraft_CropsFluxVAE
+    type: MultiScaleLatentCacheDataset
+    scales:
+    - 128
+    - 256
+    - 512
+    cache_dirs:
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/nano_banana_crops/embeddings/flux_vae
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/midjourney_full_dataset_crops_new/embeddings/flux_vae
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/gpt4_crops/embeddings/flux_vae
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/recraft_data_crops/embeddings/flux_vae
+    - /data/kazanplova/datasets/full_latent_upscale_dataset_train/mjv5_crops_fix/embeddings/flux_vae
+    vae_names:
+    - flux_vae
+    phase: train
+    filename_tmpl: '{}'
+    io_backend:
+      type: disk
+    scale: 4
+    mean: null
+    std: null
+    num_worker_per_gpu: 32
+    batch_size_per_gpu: 256
+    pin_memory: true
+    persistent_workers: true
+  val:
+    name: sdxk_120_1024x1024
+    type: MultiScaleLatentCacheDataset
+    scales:
+    - 256
+    - 512
+    - 1024
+    cache_dirs:
+    - /data/kazanplova/datasets/latent_upscale_validation_120_samples/embeddings/flux_vae
+    vae_names:
+    - flux_vae
+    phase: val
+    io_backend:
+      type: disk
+    scale: 4
+    mean: null
+    std: null
+    batch_size_per_gpu: 16
+    num_worker_per_gpu: 4
+    pin_memory: true
+network_g:
+  type: SwinIRMultiHead
+  in_chans: 16
+  img_size: 32
+  window_size: 16
+  img_range: 1.0
+  depths:
+  - 6
+  - 6
+  - 6
+  - 6
+  - 6
+  - 6
+  embed_dim: 360
+  num_heads:
+  - 12
+  - 12
+  - 12
+  - 12
+  - 12
+  - 12
+  mlp_ratio: 2
+  resi_connection: 1conv
+  primary_head: x4
+  head_num_feat: 256
+  heads:
+  - name: x2
+    scale: 2
+    out_chans: 16
+  - name: x4
+    scale: 4
+    out_chans: 16
+    primary: true
+compile:
+  enabled: false
+  mode: max-autotune
+  dynamic: true
+  fullgraph: false
+  backend: null
+train:
+  ema_decay: 0.999
+  optim_g:
+    type: Adam
+    lr: 0.0002
+    weight_decay: 0
+    betas:
+    - 0.9
+    - 0.995
+  grad_clip:
+    enabled: true
+    generator:
+      type: norm
+      max_norm: 0.4
+      norm_type: 2.0
+  scheduler:
+    type: MultiStepLR
+    milestones:
+    - 62500
+    - 93750
+    - 112500
+    gamma: 0.5
+  total_steps: 125000
+  warmup_iter: -1
+  l1_latent_x2_opt:
+    type: L1Loss
+    loss_weight: 1.0
+    reduction: mean
+    space: latent
+    target: x2
+  l1_latent_x4_opt:
+    type: L1Loss
+    loss_weight: 1.0
+    reduction: mean
+    space: latent
+    target: x4
+  fft_latent_x2_opt:
+    type: FFTFrequencyLoss
+    loss_weight: 0.1
+    reduction: mean
+    space: latent
+    target: x2
+    norm: ortho
+    use_log_amplitude: false
+    alpha: 0.0
+    normalize_weight: true
+    eps: 1e-8
+  fft_latent_x4_opt:
+    type: FFTFrequencyLoss
+    loss_weight: 0.1
+    reduction: mean
+    space: latent
+    target: x4
+    norm: ortho
+    use_log_amplitude: false
+    alpha: 0.0
+    normalize_weight: true
+    eps: 1e-8
+val:
+  val_freq: 5000
+  save_img: true
+  head_evals:
+    x2:
+      save_img: true
+      label: val_x2
+      val_sizes:
+        lq: 512
+        gt: 1024
+      metrics:
+        l1_latent:
+          type: L1Loss
+          space: latent
+        pixel_psnr_pt:
+          type: calculate_psnr_pt
+          space: pixel
+          crop_border: 2
+          test_y_channel: false
+    x4:
+      save_img: true
+      label: val_x4
+      val_sizes:
+        lq: 256
+        gt: 1024
+      metrics:
+        l1_latent:
+          type: L1Loss
+          space: latent
+        l2_latent:
+          type: MSELoss
+          space: latent
+        pixel_psnr_pt:
+          type: calculate_psnr_pt
+          space: pixel
+          crop_border: 2
+          test_y_channel: false
+logger:
+  print_freq: 100
+  save_checkpoint_freq: 5000
+  use_tb_logger: true
+  wandb:
+    project: Swin2SR-Latent-SR
+    entity: kazanplova-it-more
+    resume_id: null
+    max_val_images: 10
+dist_params:
+  backend: nccl
+  port: 29500
+  dist: true
+load_networks_only: false
+exp_name: '32'
+name: '32'
+path:
+  experiments_root: /data/kazanplova/latent_vae_upscale_train/runs/01_11_2025