delete supervised yamls + comment IJEPA paper settings

configs/supervised_wilds_vith14_ep300-lp.yaml

@@ -41,6 +41,9 @@ optimization:
   lars_eps: 1.0e-8
   lars_exclude_bias_and_norm: true
   ipe_scale: 1.0
+  # Sweep suggestions (manual edits):
+  # - lr: 0.01 | 0.05 | 0.001
+  # - weight_decay: 5.0e-4 | 0.0
   early_stopping:
     enabled: false
     patience: 10

configs/supervised_wilds_vith14_ep50-lp-lars-lr0.001-wd0.0.yaml

@@ -1,56 +0,0 @@
-meta:
-  model_name: vit_huge
-  embed_dim: 1280
-  load_checkpoint: true
-  checkpoint_folder: experiment_logs/vith14.224-bs.128-ep.300/
-  read_checkpoint: jepa-ep300.pth.tar
-  use_bfloat16: true
-  num_classes: 182
-
-data:
-  batch_size: 256
-  root_path: ./wilds_data
-  num_workers: 10
-  pin_mem: true
-  crop_size: 224
-  crop_scale: [1.0, 1.0]
-  use_random_resized_crop: false
-  use_horizontal_flip: false
-  use_color_distortion: false
-  color_jitter_strength: 0.0
-  use_gaussian_blur: false
-
-mask:
-  patch_size: 14
-
-optimization:
-  optimizer: lars # 'adamw', 'sgd', or 'lars'
-  freeze_weights: true # true for linear probing, false for full fine-tuning
-  epochs: 50
-  lr: 0.001
-  weight_decay: 0.0
-  use_cosine_schedule: false
-  lr_schedule: step
-  step_milestones: [15, 30, 45]
-  step_gamma: 0.1
-  start_lr: 0.0
-  final_lr: 0.0
-  warmup: 0
-  momentum: 0.9
-  lars_eta: 0.001
-  lars_eps: 1.0e-8
-  lars_exclude_bias_and_norm: true
-  ipe_scale: 1.0
-  early_stopping:
-    enabled: false
-    patience: 10
-    min_delta: 1.0e-4
-    min_epochs: 50
-    restore_best_weights: true
-
-validation:
-  eval_every: 1
-
-logging:
-  folder: experiment_logs/supervised-vith14.224-bs.256-ep.50-lp-lars-lr0.001-wd0.0/
-  write_tag: linear_probe-lars-lr0.001-wd0.0

configs/supervised_wilds_vith14_ep50-lp-lars-lr0.001-wd0.0005.yaml

@@ -1,56 +0,0 @@
-meta:
-  model_name: vit_huge
-  embed_dim: 1280
-  load_checkpoint: true
-  checkpoint_folder: experiment_logs/vith14.224-bs.128-ep.300/
-  read_checkpoint: jepa-ep300.pth.tar
-  use_bfloat16: true
-  num_classes: 182
-
-data:
-  batch_size: 256
-  root_path: ./wilds_data
-  num_workers: 10
-  pin_mem: true
-  crop_size: 224
-  crop_scale: [1.0, 1.0]
-  use_random_resized_crop: false
-  use_horizontal_flip: false
-  use_color_distortion: false
-  color_jitter_strength: 0.0
-  use_gaussian_blur: false
-
-mask:
-  patch_size: 14
-
-optimization:
-  optimizer: lars # 'adamw', 'sgd', or 'lars'
-  freeze_weights: true # true for linear probing, false for full fine-tuning
-  epochs: 50
-  lr: 0.001
-  weight_decay: 5.0e-4
-  use_cosine_schedule: false
-  lr_schedule: step
-  step_milestones: [15, 30, 45]
-  step_gamma: 0.1
-  start_lr: 0.0
-  final_lr: 0.0
-  warmup: 0
-  momentum: 0.9
-  lars_eta: 0.001
-  lars_eps: 1.0e-8
-  lars_exclude_bias_and_norm: true
-  ipe_scale: 1.0
-  early_stopping:
-    enabled: false
-    patience: 10
-    min_delta: 1.0e-4
-    min_epochs: 50
-    restore_best_weights: true
-
-validation:
-  eval_every: 1
-
-logging:
-  folder: experiment_logs/supervised-vith14.224-bs.256-ep.50-lp-lars-lr0.001-wd0.0005/
-  write_tag: linear_probe-lars-lr0.001-wd0.0005

configs/supervised_wilds_vith14_ep50-lp-lars-lr0.01-wd0.0.yaml

@@ -1,56 +0,0 @@
-meta:
-  model_name: vit_huge
-  embed_dim: 1280
-  load_checkpoint: true
-  checkpoint_folder: experiment_logs/vith14.224-bs.128-ep.300/
-  read_checkpoint: jepa-ep300.pth.tar
-  use_bfloat16: true
-  num_classes: 182
-
-data:
-  batch_size: 256
-  root_path: ./wilds_data
-  num_workers: 10
-  pin_mem: true
-  crop_size: 224
-  crop_scale: [1.0, 1.0]
-  use_random_resized_crop: false
-  use_horizontal_flip: false
-  use_color_distortion: false
-  color_jitter_strength: 0.0
-  use_gaussian_blur: false
-
-mask:
-  patch_size: 14
-
-optimization:
-  optimizer: lars # 'adamw', 'sgd', or 'lars'
-  freeze_weights: true # true for linear probing, false for full fine-tuning
-  epochs: 50
-  lr: 0.01
-  weight_decay: 0.0
-  use_cosine_schedule: false
-  lr_schedule: step
-  step_milestones: [15, 30, 45]
-  step_gamma: 0.1
-  start_lr: 0.0
-  final_lr: 0.0
-  warmup: 0
-  momentum: 0.9
-  lars_eta: 0.001
-  lars_eps: 1.0e-8
-  lars_exclude_bias_and_norm: true
-  ipe_scale: 1.0
-  early_stopping:
-    enabled: false
-    patience: 10
-    min_delta: 1.0e-4
-    min_epochs: 50
-    restore_best_weights: true
-
-validation:
-  eval_every: 1
-
-logging:
-  folder: experiment_logs/supervised-vith14.224-bs.256-ep.50-lp-lars-lr0.01-wd0.0/
-  write_tag: linear_probe-lars-lr0.01-wd0.0

configs/supervised_wilds_vith14_ep50-lp-lars-lr0.05-wd0.0.yaml

@@ -1,56 +0,0 @@
-meta:
-  model_name: vit_huge
-  embed_dim: 1280
-  load_checkpoint: true
-  checkpoint_folder: experiment_logs/vith14.224-bs.128-ep.300/
-  read_checkpoint: jepa-ep300.pth.tar
-  use_bfloat16: true
-  num_classes: 182
-
-data:
-  batch_size: 256
-  root_path: ./wilds_data
-  num_workers: 10
-  pin_mem: true
-  crop_size: 224
-  crop_scale: [1.0, 1.0]
-  use_random_resized_crop: false
-  use_horizontal_flip: false
-  use_color_distortion: false
-  color_jitter_strength: 0.0
-  use_gaussian_blur: false
-
-mask:
-  patch_size: 14
-
-optimization:
-  optimizer: lars # 'adamw', 'sgd', or 'lars'
-  freeze_weights: true # true for linear probing, false for full fine-tuning
-  epochs: 50
-  lr: 0.05
-  weight_decay: 0.0
-  use_cosine_schedule: false
-  lr_schedule: step
-  step_milestones: [15, 30, 45]
-  step_gamma: 0.1
-  start_lr: 0.0
-  final_lr: 0.0
-  warmup: 0
-  momentum: 0.9
-  lars_eta: 0.001
-  lars_eps: 1.0e-8
-  lars_exclude_bias_and_norm: true
-  ipe_scale: 1.0
-  early_stopping:
-    enabled: false
-    patience: 10
-    min_delta: 1.0e-4
-    min_epochs: 50
-    restore_best_weights: true
-
-validation:
-  eval_every: 1
-
-logging:
-  folder: experiment_logs/supervised-vith14.224-bs.256-ep.50-lp-lars-lr0.05-wd0.0/
-  write_tag: linear_probe-lars-lr0.05-wd0.0

configs/supervised_wilds_vith14_ep50-lp-lars-lr0.05-wd0.0005.yaml

@@ -1,56 +0,0 @@
-meta:
-  model_name: vit_huge
-  embed_dim: 1280
-  load_checkpoint: true
-  checkpoint_folder: experiment_logs/vith14.224-bs.128-ep.300/
-  read_checkpoint: jepa-ep300.pth.tar
-  use_bfloat16: true
-  num_classes: 182
-
-data:
-  batch_size: 256
-  root_path: ./wilds_data
-  num_workers: 10
-  pin_mem: true
-  crop_size: 224
-  crop_scale: [1.0, 1.0]
-  use_random_resized_crop: false
-  use_horizontal_flip: false
-  use_color_distortion: false
-  color_jitter_strength: 0.0
-  use_gaussian_blur: false
-
-mask:
-  patch_size: 14
-
-optimization:
-  optimizer: lars # 'adamw', 'sgd', or 'lars'
-  freeze_weights: true # true for linear probing, false for full fine-tuning
-  epochs: 50
-  lr: 0.05
-  weight_decay: 5.0e-4
-  use_cosine_schedule: false
-  lr_schedule: step
-  step_milestones: [15, 30, 45]
-  step_gamma: 0.1
-  start_lr: 0.0
-  final_lr: 0.0
-  warmup: 0
-  momentum: 0.9
-  lars_eta: 0.001
-  lars_eps: 1.0e-8
-  lars_exclude_bias_and_norm: true
-  ipe_scale: 1.0
-  early_stopping:
-    enabled: false
-    patience: 10
-    min_delta: 1.0e-4
-    min_epochs: 50
-    restore_best_weights: true
-
-validation:
-  eval_every: 1
-
-logging:
-  folder: experiment_logs/supervised-vith14.224-bs.256-ep.50-lp-lars-lr0.05-wd0.0005/
-  write_tag: linear_probe-lars-lr0.05-wd0.0005