Upload folder using huggingface_hub

fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000/5min/m-eurosat_knn/config.yaml

@@ -0,0 +1,43 @@
+task:
+  id: knn
+  is_multilabel: false
+  metrics:
+  - id: MulticlassAccuracy
+    top_k: 1
+    average: micro
+  backbone_to_features:
+    pooling: knn
+    use_n_blocks: 1
+heads:
+  nb_knn:
+  - 20
+  temperature: 0.07
+  gather_on_cpu: false
+  n_per_class_list:
+  - -1
+  n_tries: 1
+train_dataset:
+  id: geobench.m-eurosat
+  split: train
+  transform:
+  - id: Resize
+    size: 224
+  normalize: false
+test_dataset:
+  id: geobench.m-eurosat
+  split: test
+  transform:
+  - id: Resize
+    size: 224
+  normalize: false
+seed: 42
+optim:
+  dl:
+    batch_size: 200
+    num_workers: 4
+    persistent_workers: true
+_vars:
+  transforms:
+  - id: Resize
+    size: 224
+output_dir: /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000/5min/m-eurosat_knn

fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000/5min/m-eurosat_knn/log

@@ -0,0 +1,360 @@
+I20260214 22:33:54 49291 dinov2 setup.py:34] task:
+  id: knn
+  is_multilabel: false
+  metrics:
+  - id: MulticlassAccuracy
+    top_k: 1
+    average: micro
+  backbone_to_features:
+    pooling: knn
+    use_n_blocks: 1
+heads:
+  nb_knn:
+  - 20
+  temperature: 0.07
+  gather_on_cpu: false
+  n_per_class_list:
+  - -1
+  n_tries: 1
+train_dataset:
+  id: geobench.m-eurosat
+  split: train
+  transform:
+  - id: Resize
+    size: 224
+test_dataset:
+  id: geobench.m-eurosat
+  split: test
+  transform:
+  - id: Resize
+    size: 224
+seed: 42
+optim:
+  dl:
+    batch_size: 200
+    num_workers: 4
+    persistent_workers: true
+_vars:
+  transforms:
+  - id: Resize
+    size: 224
+output_dir: /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000/5min/m-eurosat_knn
+
+I20260214 22:33:54 49291 dinov2 wrapper.py:38] Built model nanochat_fmvit
+I20260214 22:33:55 49291 dinov2 loaders.py:72] Building dataset "geobench.m-eurosat" ...
+I20260214 22:33:55 49291 dinov2 augmentations.py:44] Augmentations in order: ['Resize']
+I20260214 22:33:55 49291 dinov2 loaders.py:161] Built dataset "geobench.m-eurosat" with #samples 2000
+I20260214 22:33:55 49291 dinov2 loaders.py:72] Building dataset "geobench.m-eurosat" ...
+I20260214 22:33:55 49291 dinov2 augmentations.py:44] Augmentations in order: ['Resize']
+I20260214 22:33:55 49291 dinov2 loaders.py:161] Built dataset "geobench.m-eurosat" with #samples 1000
+I20260214 22:33:55 49291 dinov2 knn.py:224] Extracting features for train set...
+I20260214 22:33:55 49291 dinov2 loaders.py:328] Detected non-CombinedDataset. Using SamplerType.EPOCH with bsz=200.
+I20260214 22:33:55 49291 dinov2 loaders.py:234] sampler: epoch
+I20260214 22:33:55 49291 dinov2 loaders.py:238] # of samples / epoch: 2,000
+I20260214 22:33:55 49291 dinov2 loaders.py:344] DataLoader kwargs: num_workers=4, pin_memory=True, drop_last=False, persistent_workers=True
+I20260214 22:33:55 49291 dinov2 loaders.py:356] # of batches: 10
+I20260214 22:35:04 49748 dinov2 setup.py:34] task:
+  id: knn
+  is_multilabel: false
+  metrics:
+  - id: MulticlassAccuracy
+    top_k: 1
+    average: micro
+  backbone_to_features:
+    pooling: knn
+    use_n_blocks: 1
+heads:
+  nb_knn:
+  - 20
+  temperature: 0.07
+  gather_on_cpu: false
+  n_per_class_list:
+  - -1
+  n_tries: 1
+train_dataset:
+  id: geobench.m-eurosat
+  split: train
+  transform:
+  - id: Resize
+    size: 224
+test_dataset:
+  id: geobench.m-eurosat
+  split: test
+  transform:
+  - id: Resize
+    size: 224
+seed: 42
+optim:
+  dl:
+    batch_size: 200
+    num_workers: 4
+    persistent_workers: true
+_vars:
+  transforms:
+  - id: Resize
+    size: 224
+output_dir: /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000/5min/m-eurosat_knn
+
+I20260214 22:35:04 49748 dinov2 wrapper.py:38] Built model nanochat_fmvit
+I20260214 22:35:06 49748 dinov2 loaders.py:72] Building dataset "geobench.m-eurosat" ...
+I20260214 22:35:06 49748 dinov2 augmentations.py:44] Augmentations in order: ['Resize']
+I20260214 22:35:06 49748 dinov2 loaders.py:161] Built dataset "geobench.m-eurosat" with #samples 2000
+I20260214 22:35:06 49748 dinov2 loaders.py:72] Building dataset "geobench.m-eurosat" ...
+I20260214 22:35:06 49748 dinov2 augmentations.py:44] Augmentations in order: ['Resize']
+I20260214 22:35:06 49748 dinov2 loaders.py:161] Built dataset "geobench.m-eurosat" with #samples 1000
+I20260214 22:35:06 49748 dinov2 knn.py:224] Extracting features for train set...
+I20260214 22:35:06 49748 dinov2 loaders.py:328] Detected non-CombinedDataset. Using SamplerType.EPOCH with bsz=200.
+I20260214 22:35:06 49748 dinov2 loaders.py:234] sampler: epoch
+I20260214 22:35:06 49748 dinov2 loaders.py:238] # of samples / epoch: 2,000
+I20260214 22:35:06 49748 dinov2 loaders.py:344] DataLoader kwargs: num_workers=4, pin_memory=True, drop_last=False, persistent_workers=True
+I20260214 22:35:06 49748 dinov2 loaders.py:356] # of batches: 10
+I20260214 22:36:08 50158 dinov2 setup.py:34] task:
+  id: knn
+  is_multilabel: false
+  metrics:
+  - id: MulticlassAccuracy
+    top_k: 1
+    average: micro
+  backbone_to_features:
+    pooling: knn
+    use_n_blocks: 1
+heads:
+  nb_knn:
+  - 20
+  temperature: 0.07
+  gather_on_cpu: false
+  n_per_class_list:
+  - -1
+  n_tries: 1
+train_dataset:
+  id: geobench.m-eurosat
+  split: train
+  transform:
+  - id: Resize
+    size: 224
+test_dataset:
+  id: geobench.m-eurosat
+  split: test
+  transform:
+  - id: Resize
+    size: 224
+seed: 42
+optim:
+  dl:
+    batch_size: 200
+    num_workers: 4
+    persistent_workers: true
+_vars:
+  transforms:
+  - id: Resize
+    size: 224
+output_dir: /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000/5min/m-eurosat_knn
+
+I20260214 22:36:08 50158 dinov2 wrapper.py:38] Built model nanochat_fmvit
+I20260214 22:36:10 50158 dinov2 loaders.py:72] Building dataset "geobench.m-eurosat" ...
+I20260214 22:36:10 50158 dinov2 augmentations.py:44] Augmentations in order: ['Resize']
+I20260214 22:36:10 50158 dinov2 loaders.py:161] Built dataset "geobench.m-eurosat" with #samples 2000
+I20260214 22:36:10 50158 dinov2 loaders.py:72] Building dataset "geobench.m-eurosat" ...
+I20260214 22:36:10 50158 dinov2 augmentations.py:44] Augmentations in order: ['Resize']
+I20260214 22:36:10 50158 dinov2 loaders.py:161] Built dataset "geobench.m-eurosat" with #samples 1000
+I20260214 22:36:10 50158 dinov2 knn.py:224] Extracting features for train set...
+I20260214 22:36:10 50158 dinov2 loaders.py:328] Detected non-CombinedDataset. Using SamplerType.EPOCH with bsz=200.
+I20260214 22:36:10 50158 dinov2 loaders.py:234] sampler: epoch
+I20260214 22:36:10 50158 dinov2 loaders.py:238] # of samples / epoch: 2,000
+I20260214 22:36:10 50158 dinov2 loaders.py:344] DataLoader kwargs: num_workers=4, pin_memory=True, drop_last=False, persistent_workers=True
+I20260214 22:36:10 50158 dinov2 loaders.py:356] # of batches: 10
+I20260214 22:36:19 50158 dinov2 utils.py:115] Storing features into tensor of shape torch.Size([2000, 768])
+I20260214 22:36:19 50158 dinov2 helpers.py:190]   [iter:  0/10,   epoch: 0/1]  eta: 0:01:29    time: 8.931544  data: 7.434315  max mem: 6853
+I20260214 22:36:30 50158 dinov2 helpers.py:190]   [iter:  9/10,   epoch: 0/1]  eta: 0:00:02    time: 2.008143  data: 1.230095  max mem: 6858
+I20260214 22:36:30 50158 dinov2 helpers.py:217] Epoch 0/1 done in 20.08s
+
+I20260214 22:36:30 50158 dinov2 helpers.py:225]  Total time: 0:00:20 (2.008463 s / it)
+
+I20260214 22:36:30 50158 dinov2 utils.py:127] Features shape: (2000, 768)
+I20260214 22:36:30 50158 dinov2 utils.py:128] Labels shape: (2000,)
+I20260214 22:36:31 50158 dinov2 loaders.py:328] Detected non-CombinedDataset. Using SamplerType.EPOCH with bsz=200.
+I20260214 22:36:31 50158 dinov2 loaders.py:234] sampler: epoch
+I20260214 22:36:31 50158 dinov2 loaders.py:238] # of samples / epoch: 1,000
+I20260214 22:36:31 50158 dinov2 loaders.py:344] DataLoader kwargs: num_workers=4, pin_memory=True, drop_last=False, persistent_workers=True
+I20260214 22:36:31 50158 dinov2 loaders.py:356] # of batches: 5
+I20260214 22:36:31 50158 dinov2 knn.py:241] Using knn module: <class 'dinov2.eval.knn.KnnModule'> with num_classes 10
+I20260214 22:36:31 50158 dinov2 knn.py:262] Start the k-NN classification.
+I20260214 22:36:40 50158 dinov2 helpers.py:190] Test:  [iter: 0/5,   epoch: 0/1]  eta: 0:00:45    time: 9.168560  data: 8.075690  max mem: 6858
+I20260214 22:36:46 50158 dinov2 helpers.py:190] Test:  [iter: 4/5,   epoch: 0/1]  eta: 0:00:03    time: 3.081460  data: 2.111583  max mem: 6858
+I20260214 22:36:46 50158 dinov2 helpers.py:217] Epoch 0/1 done in 15.41s
+
+I20260214 22:36:46 50158 dinov2 helpers.py:225] Test: Total time: 0:00:15 (3.081754 s / it)
+
+I20260214 22:36:46 50158 dinov2 utils.py:56] Averaged stats: 
+D20260214 22:36:46 50158 dinov2 utils.py:59] Post compute
+D20260214 22:36:46 50158 dinov2 knn.py:264] Finished KNN classification
+I20260214 22:36:46 50158 dinov2 knn.py:327] All metrics result:
+('full', 20): {acc_top-1_micro: 70.50, }
+I20260214 22:38:04 50775 dinov2 setup.py:34] task:
+  id: knn
+  is_multilabel: false
+  metrics:
+  - id: MulticlassAccuracy
+    top_k: 1
+    average: micro
+  backbone_to_features:
+    pooling: knn
+    use_n_blocks: 1
+heads:
+  nb_knn:
+  - 20
+  temperature: 0.07
+  gather_on_cpu: false
+  n_per_class_list:
+  - -1
+  n_tries: 1
+train_dataset:
+  id: geobench.m-eurosat
+  split: train
+  transform:
+  - id: Resize
+    size: 224
+  normalize: false
+test_dataset:
+  id: geobench.m-eurosat
+  split: test
+  transform:
+  - id: Resize
+    size: 224
+  normalize: false
+seed: 42
+optim:
+  dl:
+    batch_size: 200
+    num_workers: 4
+    persistent_workers: true
+_vars:
+  transforms:
+  - id: Resize
+    size: 224
+output_dir: /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000/5min/m-eurosat_knn
+
+I20260214 22:38:04 50775 dinov2 wrapper.py:38] Built model nanochat_fmvit
+I20260214 22:38:06 50775 dinov2 loaders.py:72] Building dataset "geobench.m-eurosat" ...
+I20260214 22:38:06 50775 dinov2 augmentations.py:44] Augmentations in order: ['Resize']
+I20260214 22:38:06 50775 dinov2 loaders.py:161] Built dataset "geobench.m-eurosat" with #samples 2000
+I20260214 22:38:06 50775 dinov2 loaders.py:72] Building dataset "geobench.m-eurosat" ...
+I20260214 22:38:06 50775 dinov2 augmentations.py:44] Augmentations in order: ['Resize']
+I20260214 22:38:06 50775 dinov2 loaders.py:161] Built dataset "geobench.m-eurosat" with #samples 1000
+I20260214 22:38:06 50775 dinov2 knn.py:224] Extracting features for train set...
+I20260214 22:38:06 50775 dinov2 loaders.py:328] Detected non-CombinedDataset. Using SamplerType.EPOCH with bsz=200.
+I20260214 22:38:06 50775 dinov2 loaders.py:234] sampler: epoch
+I20260214 22:38:06 50775 dinov2 loaders.py:238] # of samples / epoch: 2,000
+I20260214 22:38:06 50775 dinov2 loaders.py:344] DataLoader kwargs: num_workers=4, pin_memory=True, drop_last=False, persistent_workers=True
+I20260214 22:38:06 50775 dinov2 loaders.py:356] # of batches: 10
+I20260214 22:38:15 50775 dinov2 utils.py:115] Storing features into tensor of shape torch.Size([2000, 768])
+I20260214 22:38:15 50775 dinov2 helpers.py:190]   [iter:  0/10,   epoch: 0/1]  eta: 0:01:32    time: 9.213921  data: 7.967088  max mem: 6853
+I20260214 22:38:27 50775 dinov2 helpers.py:190]   [iter:  9/10,   epoch: 0/1]  eta: 0:00:02    time: 2.170295  data: 1.416511  max mem: 6858
+I20260214 22:38:27 50775 dinov2 helpers.py:217] Epoch 0/1 done in 21.70s
+
+I20260214 22:38:27 50775 dinov2 helpers.py:225]  Total time: 0:00:21 (2.170450 s / it)
+
+I20260214 22:38:27 50775 dinov2 utils.py:127] Features shape: (2000, 768)
+I20260214 22:38:27 50775 dinov2 utils.py:128] Labels shape: (2000,)
+I20260214 22:38:28 50775 dinov2 loaders.py:328] Detected non-CombinedDataset. Using SamplerType.EPOCH with bsz=200.
+I20260214 22:38:28 50775 dinov2 loaders.py:234] sampler: epoch
+I20260214 22:38:28 50775 dinov2 loaders.py:238] # of samples / epoch: 1,000
+I20260214 22:38:28 50775 dinov2 loaders.py:344] DataLoader kwargs: num_workers=4, pin_memory=True, drop_last=False, persistent_workers=True
+I20260214 22:38:28 50775 dinov2 loaders.py:356] # of batches: 5
+I20260214 22:38:28 50775 dinov2 knn.py:241] Using knn module: <class 'dinov2.eval.knn.KnnModule'> with num_classes 10
+I20260214 22:38:28 50775 dinov2 knn.py:262] Start the k-NN classification.
+I20260214 22:38:36 50775 dinov2 helpers.py:190] Test:  [iter: 0/5,   epoch: 0/1]  eta: 0:00:36    time: 7.324920  data: 6.243755  max mem: 6858
+I20260214 22:38:41 50775 dinov2 helpers.py:190] Test:  [iter: 4/5,   epoch: 0/1]  eta: 0:00:02    time: 2.581742  data: 1.613907  max mem: 6858
+I20260214 22:38:41 50775 dinov2 helpers.py:217] Epoch 0/1 done in 12.91s
+
+I20260214 22:38:41 50775 dinov2 helpers.py:225] Test: Total time: 0:00:12 (2.582041 s / it)
+
+I20260214 22:38:41 50775 dinov2 utils.py:56] Averaged stats: 
+D20260214 22:38:41 50775 dinov2 utils.py:59] Post compute
+D20260214 22:38:41 50775 dinov2 knn.py:264] Finished KNN classification
+I20260214 22:38:41 50775 dinov2 knn.py:327] All metrics result:
+('full', 20): {acc_top-1_micro: 58.00, }
+I20260214 22:40:26 51715 dinov2 setup.py:34] task:
+  id: knn
+  is_multilabel: false
+  metrics:
+  - id: MulticlassAccuracy
+    top_k: 1
+    average: micro
+  backbone_to_features:
+    pooling: knn
+    use_n_blocks: 1
+heads:
+  nb_knn:
+  - 20
+  temperature: 0.07
+  gather_on_cpu: false
+  n_per_class_list:
+  - -1
+  n_tries: 1
+train_dataset:
+  id: geobench.m-eurosat
+  split: train
+  transform:
+  - id: Resize
+    size: 224
+  normalize: false
+test_dataset:
+  id: geobench.m-eurosat
+  split: test
+  transform:
+  - id: Resize
+    size: 224
+  normalize: false
+seed: 42
+optim:
+  dl:
+    batch_size: 200
+    num_workers: 4
+    persistent_workers: true
+_vars:
+  transforms:
+  - id: Resize
+    size: 224
+output_dir: /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000/5min/m-eurosat_knn
+
+I20260214 22:40:26 51715 dinov2 wrapper.py:39] Built model nanochat_fmvit
+I20260214 22:40:29 51715 dinov2 loaders.py:72] Building dataset "geobench.m-eurosat" ...
+I20260214 22:40:29 51715 dinov2 augmentations.py:44] Augmentations in order: ['Resize']
+I20260214 22:40:29 51715 dinov2 loaders.py:161] Built dataset "geobench.m-eurosat" with #samples 2000
+I20260214 22:40:29 51715 dinov2 loaders.py:72] Building dataset "geobench.m-eurosat" ...
+I20260214 22:40:29 51715 dinov2 augmentations.py:44] Augmentations in order: ['Resize']
+I20260214 22:40:29 51715 dinov2 loaders.py:161] Built dataset "geobench.m-eurosat" with #samples 1000
+I20260214 22:40:29 51715 dinov2 knn.py:224] Extracting features for train set...
+I20260214 22:40:29 51715 dinov2 loaders.py:328] Detected non-CombinedDataset. Using SamplerType.EPOCH with bsz=200.
+I20260214 22:40:29 51715 dinov2 loaders.py:234] sampler: epoch
+I20260214 22:40:29 51715 dinov2 loaders.py:238] # of samples / epoch: 2,000
+I20260214 22:40:29 51715 dinov2 loaders.py:344] DataLoader kwargs: num_workers=4, pin_memory=True, drop_last=False, persistent_workers=True
+I20260214 22:40:29 51715 dinov2 loaders.py:356] # of batches: 10
+I20260214 22:40:37 51715 dinov2 utils.py:115] Storing features into tensor of shape torch.Size([2000, 768])
+I20260214 22:40:37 51715 dinov2 helpers.py:190]   [iter:  0/10,   epoch: 0/1]  eta: 0:01:26    time: 8.666345  data: 7.815904  max mem: 2984
+I20260214 22:40:49 51715 dinov2 helpers.py:190]   [iter:  9/10,   epoch: 0/1]  eta: 0:00:02    time: 2.029455  data: 1.751290  max mem: 2990
+I20260214 22:40:49 51715 dinov2 helpers.py:217] Epoch 0/1 done in 20.30s
+
+I20260214 22:40:49 51715 dinov2 helpers.py:225]  Total time: 0:00:20 (2.029649 s / it)
+
+I20260214 22:40:49 51715 dinov2 utils.py:127] Features shape: (2000, 768)
+I20260214 22:40:49 51715 dinov2 utils.py:128] Labels shape: (2000,)
+I20260214 22:40:49 51715 dinov2 loaders.py:328] Detected non-CombinedDataset. Using SamplerType.EPOCH with bsz=200.
+I20260214 22:40:49 51715 dinov2 loaders.py:234] sampler: epoch
+I20260214 22:40:49 51715 dinov2 loaders.py:238] # of samples / epoch: 1,000
+I20260214 22:40:49 51715 dinov2 loaders.py:344] DataLoader kwargs: num_workers=4, pin_memory=True, drop_last=False, persistent_workers=True
+I20260214 22:40:49 51715 dinov2 loaders.py:356] # of batches: 5
+I20260214 22:40:49 51715 dinov2 knn.py:241] Using knn module: <class 'dinov2.eval.knn.KnnModule'> with num_classes 10
+I20260214 22:40:49 51715 dinov2 knn.py:262] Start the k-NN classification.
+I20260214 22:40:57 51715 dinov2 helpers.py:190] Test:  [iter: 0/5,   epoch: 0/1]  eta: 0:00:35    time: 7.095051  data: 6.543128  max mem: 2990
+I20260214 22:41:02 51715 dinov2 helpers.py:190] Test:  [iter: 4/5,   epoch: 0/1]  eta: 0:00:02    time: 2.526883  data: 2.169243  max mem: 2990
+I20260214 22:41:02 51715 dinov2 helpers.py:217] Epoch 0/1 done in 12.64s
+
+I20260214 22:41:02 51715 dinov2 helpers.py:225] Test: Total time: 0:00:12 (2.527236 s / it)
+
+I20260214 22:41:02 51715 dinov2 utils.py:56] Averaged stats: 
+D20260214 22:41:02 51715 dinov2 utils.py:59] Post compute
+D20260214 22:41:02 51715 dinov2 knn.py:264] Finished KNN classification
+I20260214 22:41:02 51715 dinov2 knn.py:327] All metrics result:
+('full', 20): {acc_top-1_micro: 56.90, }

fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000/5min/m-eurosat_knn/results.csv

@@ -0,0 +1,2 @@
+metric,best_classifier,value
+acc_top-1_micro,"('full', 20)",56.9

fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000/5min/m-eurosat_knn/results_eval_knn.json

@@ -0,0 +1,3 @@
+{"('full', 20)": "{'acc_top-1_micro': 70.5}"}
+{"('full', 20)": "{'acc_top-1_micro': 58.0}"}
+{"('full', 20)": "{'acc_top-1_micro': 56.9}"}

fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000/5min/m-eurosat_knn_nanochat/config.yaml

@@ -0,0 +1,37 @@
+task:
+  id: knn
+  is_multilabel: false
+  metrics:
+  - id: MulticlassAccuracy
+    top_k: 1
+    average: micro
+  backbone_to_features:
+    pooling: knn
+    use_n_blocks: 1
+heads:
+  nb_knn:
+  - 20
+  temperature: 0.07
+  gather_on_cpu: false
+  n_per_class_list:
+  - -1
+  n_tries: 1
+train_dataset:
+  id: geobench.m-eurosat
+  split: train
+  transform: []
+  normalize: false
+test_dataset:
+  id: geobench.m-eurosat
+  split: test
+  transform: []
+  normalize: false
+seed: 42
+optim:
+  dl:
+    batch_size: 200
+    num_workers: 4
+    persistent_workers: true
+_vars:
+  transforms: []
+output_dir: /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000/5min/m-eurosat_knn_nanochat

fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000/5min/m-eurosat_knn_nanochat/log

@@ -0,0 +1,441 @@
+I20260215 07:10:15 1984 dinov2 setup.py:34] task:
+  id: knn
+  is_multilabel: false
+  metrics:
+  - id: MulticlassAccuracy
+    top_k: 1
+    average: micro
+  backbone_to_features:
+    pooling: knn
+    use_n_blocks: 1
+heads:
+  nb_knn:
+  - 20
+  temperature: 0.07
+  gather_on_cpu: false
+  n_per_class_list:
+  - -1
+  n_tries: 1
+train_dataset:
+  id: geobench.m-eurosat
+  split: train
+  transform: []
+  normalize: false
+test_dataset:
+  id: geobench.m-eurosat
+  split: test
+  transform: []
+  normalize: false
+seed: 42
+optim:
+  dl:
+    batch_size: 200
+    num_workers: 4
+    persistent_workers: true
+_vars:
+  transforms: []
+output_dir: /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000/5min/m-eurosat_knn_nanochat
+
+I20260215 07:10:15 1984 dinov2 wrapper.py:39] Built model nanochat_fmvit
+I20260215 07:10:17 1984 dinov2 loaders.py:72] Building dataset "geobench.m-eurosat" ...
+I20260215 07:10:17 1984 dinov2 augmentations.py:44] Augmentations in order: []
+I20260215 07:10:17 1984 dinov2 loaders.py:161] Built dataset "geobench.m-eurosat" with #samples 2000
+I20260215 07:10:17 1984 dinov2 loaders.py:72] Building dataset "geobench.m-eurosat" ...
+I20260215 07:10:17 1984 dinov2 augmentations.py:44] Augmentations in order: []
+I20260215 07:10:17 1984 dinov2 loaders.py:161] Built dataset "geobench.m-eurosat" with #samples 1000
+I20260215 07:10:17 1984 dinov2 knn.py:224] Extracting features for train set...
+I20260215 07:10:17 1984 dinov2 loaders.py:328] Detected non-CombinedDataset. Using SamplerType.EPOCH with bsz=200.
+I20260215 07:10:17 1984 dinov2 loaders.py:234] sampler: epoch
+I20260215 07:10:17 1984 dinov2 loaders.py:238] # of samples / epoch: 2,000
+I20260215 07:10:17 1984 dinov2 loaders.py:344] DataLoader kwargs: num_workers=4, pin_memory=True, drop_last=False, persistent_workers=True
+I20260215 07:10:17 1984 dinov2 loaders.py:356] # of batches: 10
+I20260215 07:10:27 1984 dinov2 utils.py:115] Storing features into tensor of shape torch.Size([2000, 768])
+I20260215 07:10:27 1984 dinov2 helpers.py:190]   [iter:  0/10,   epoch: 0/1]  eta: 0:01:34    time: 9.415029  data: 5.697691  max mem: 21502
+I20260215 07:10:55 1984 dinov2 helpers.py:190]   [iter:  9/10,   epoch: 0/1]  eta: 0:00:03    time: 3.819382  data: 0.570064  max mem: 21508
+I20260215 07:10:55 1984 dinov2 helpers.py:217] Epoch 0/1 done in 38.20s
+
+I20260215 07:10:55 1984 dinov2 helpers.py:225]  Total time: 0:00:38 (3.819566 s / it)
+
+I20260215 07:10:55 1984 dinov2 utils.py:127] Features shape: (2000, 768)
+I20260215 07:10:55 1984 dinov2 utils.py:128] Labels shape: (2000,)
+I20260215 07:10:57 1984 dinov2 loaders.py:328] Detected non-CombinedDataset. Using SamplerType.EPOCH with bsz=200.
+I20260215 07:10:57 1984 dinov2 loaders.py:234] sampler: epoch
+I20260215 07:10:57 1984 dinov2 loaders.py:238] # of samples / epoch: 1,000
+I20260215 07:10:57 1984 dinov2 loaders.py:344] DataLoader kwargs: num_workers=4, pin_memory=True, drop_last=False, persistent_workers=True
+I20260215 07:10:57 1984 dinov2 loaders.py:356] # of batches: 5
+I20260215 07:10:57 1984 dinov2 knn.py:241] Using knn module: <class 'dinov2.eval.knn.KnnModule'> with num_classes 10
+I20260215 07:10:57 1984 dinov2 knn.py:262] Start the k-NN classification.
+I20260215 07:11:07 1984 dinov2 helpers.py:190] Test:  [iter: 0/5,   epoch: 0/1]  eta: 0:00:46    time: 9.314626  data: 5.740987  max mem: 21508
+I20260215 07:11:21 1984 dinov2 helpers.py:190] Test:  [iter: 4/5,   epoch: 0/1]  eta: 0:00:04    time: 4.613917  data: 1.148376  max mem: 21508
+I20260215 07:11:21 1984 dinov2 helpers.py:217] Epoch 0/1 done in 23.07s
+
+I20260215 07:11:21 1984 dinov2 helpers.py:225] Test: Total time: 0:00:23 (4.614269 s / it)
+
+I20260215 07:11:21 1984 dinov2 utils.py:56] Averaged stats: 
+D20260215 07:11:21 1984 dinov2 utils.py:59] Post compute
+D20260215 07:11:21 1984 dinov2 knn.py:264] Finished KNN classification
+I20260215 07:11:21 1984 dinov2 knn.py:327] All metrics result:
+('full', 20): {acc_top-1_micro: 56.30, }
+I20260215 07:13:03 3370 dinov2 setup.py:34] task:
+  id: knn
+  is_multilabel: false
+  metrics:
+  - id: MulticlassAccuracy
+    top_k: 1
+    average: micro
+  backbone_to_features:
+    pooling: knn
+    use_n_blocks: 1
+heads:
+  nb_knn:
+  - 20
+  temperature: 0.07
+  gather_on_cpu: false
+  n_per_class_list:
+  - -1
+  n_tries: 1
+train_dataset:
+  id: geobench.m-eurosat
+  split: train
+  transform: []
+  normalize: false
+test_dataset:
+  id: geobench.m-eurosat
+  split: test
+  transform: []
+  normalize: false
+seed: 42
+optim:
+  dl:
+    batch_size: 200
+    num_workers: 4
+    persistent_workers: true
+_vars:
+  transforms: []
+output_dir: /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000/5min/m-eurosat_knn_nanochat
+
+I20260215 07:13:03 3370 dinov2 wrapper.py:39] Built model nanochat_fmvit
+I20260215 07:13:05 3370 dinov2 loaders.py:72] Building dataset "geobench.m-eurosat" ...
+I20260215 07:13:05 3370 dinov2 augmentations.py:44] Augmentations in order: []
+I20260215 07:13:06 3370 dinov2 loaders.py:161] Built dataset "geobench.m-eurosat" with #samples 2000
+I20260215 07:13:06 3370 dinov2 loaders.py:72] Building dataset "geobench.m-eurosat" ...
+I20260215 07:13:06 3370 dinov2 augmentations.py:44] Augmentations in order: []
+I20260215 07:13:06 3370 dinov2 loaders.py:161] Built dataset "geobench.m-eurosat" with #samples 1000
+I20260215 07:13:06 3370 dinov2 knn.py:224] Extracting features for train set...
+I20260215 07:13:06 3370 dinov2 loaders.py:328] Detected non-CombinedDataset. Using SamplerType.EPOCH with bsz=200.
+I20260215 07:13:06 3370 dinov2 loaders.py:234] sampler: epoch
+I20260215 07:13:06 3370 dinov2 loaders.py:238] # of samples / epoch: 2,000
+I20260215 07:13:06 3370 dinov2 loaders.py:344] DataLoader kwargs: num_workers=4, pin_memory=True, drop_last=False, persistent_workers=True
+I20260215 07:13:06 3370 dinov2 loaders.py:356] # of batches: 10
+I20260215 07:14:03 4072 dinov2 setup.py:34] task:
+  id: knn
+  is_multilabel: false
+  metrics:
+  - id: MulticlassAccuracy
+    top_k: 1
+    average: micro
+  backbone_to_features:
+    pooling: knn
+    use_n_blocks: 1
+heads:
+  nb_knn:
+  - 20
+  temperature: 0.07
+  gather_on_cpu: false
+  n_per_class_list:
+  - -1
+  n_tries: 1
+train_dataset:
+  id: geobench.m-eurosat
+  split: train
+  transform: []
+  normalize: false
+test_dataset:
+  id: geobench.m-eurosat
+  split: test
+  transform: []
+  normalize: false
+seed: 42
+optim:
+  dl:
+    batch_size: 200
+    num_workers: 4
+    persistent_workers: true
+_vars:
+  transforms: []
+output_dir: /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000/5min/m-eurosat_knn_nanochat
+
+I20260215 07:14:03 4072 dinov2 wrapper.py:39] Built model nanochat_fmvit
+I20260215 07:14:06 4072 dinov2 loaders.py:72] Building dataset "geobench.m-eurosat" ...
+I20260215 07:14:06 4072 dinov2 augmentations.py:44] Augmentations in order: []
+I20260215 07:14:06 4072 dinov2 loaders.py:161] Built dataset "geobench.m-eurosat" with #samples 2000
+I20260215 07:14:06 4072 dinov2 loaders.py:72] Building dataset "geobench.m-eurosat" ...
+I20260215 07:14:06 4072 dinov2 augmentations.py:44] Augmentations in order: []
+I20260215 07:14:06 4072 dinov2 loaders.py:161] Built dataset "geobench.m-eurosat" with #samples 1000
+I20260215 07:14:06 4072 dinov2 knn.py:224] Extracting features for train set...
+I20260215 07:14:06 4072 dinov2 loaders.py:328] Detected non-CombinedDataset. Using SamplerType.EPOCH with bsz=200.
+I20260215 07:14:06 4072 dinov2 loaders.py:234] sampler: epoch
+I20260215 07:14:06 4072 dinov2 loaders.py:238] # of samples / epoch: 2,000
+I20260215 07:14:06 4072 dinov2 loaders.py:344] DataLoader kwargs: num_workers=4, pin_memory=True, drop_last=False, persistent_workers=True
+I20260215 07:14:06 4072 dinov2 loaders.py:356] # of batches: 10
+I20260215 07:14:15 4072 dinov2 utils.py:115] Storing features into tensor of shape torch.Size([2000, 768])
+I20260215 07:14:15 4072 dinov2 helpers.py:190]   [iter:  0/10,   epoch: 0/1]  eta: 0:01:31    time: 9.167652  data: 7.663894  max mem: 17023
+I20260215 07:14:26 4072 dinov2 helpers.py:190]   [iter:  9/10,   epoch: 0/1]  eta: 0:00:01    time: 1.980520  data: 1.040272  max mem: 17030
+I20260215 07:14:26 4072 dinov2 helpers.py:217] Epoch 0/1 done in 19.81s
+
+I20260215 07:14:26 4072 dinov2 helpers.py:225]  Total time: 0:00:19 (1.980689 s / it)
+
+I20260215 07:14:26 4072 dinov2 utils.py:127] Features shape: (2000, 768)
+I20260215 07:14:26 4072 dinov2 utils.py:128] Labels shape: (2000,)
+I20260215 07:14:27 4072 dinov2 loaders.py:328] Detected non-CombinedDataset. Using SamplerType.EPOCH with bsz=200.
+I20260215 07:14:27 4072 dinov2 loaders.py:234] sampler: epoch
+I20260215 07:14:27 4072 dinov2 loaders.py:238] # of samples / epoch: 1,000
+I20260215 07:14:27 4072 dinov2 loaders.py:344] DataLoader kwargs: num_workers=4, pin_memory=True, drop_last=False, persistent_workers=True
+I20260215 07:14:27 4072 dinov2 loaders.py:356] # of batches: 5
+I20260215 07:14:27 4072 dinov2 knn.py:241] Using knn module: <class 'dinov2.eval.knn.KnnModule'> with num_classes 10
+I20260215 07:14:27 4072 dinov2 knn.py:262] Start the k-NN classification.
+I20260215 07:14:34 4072 dinov2 helpers.py:190] Test:  [iter: 0/5,   epoch: 0/1]  eta: 0:00:35    time: 7.155606  data: 6.047797  max mem: 17030
+I20260215 07:14:39 4072 dinov2 helpers.py:190] Test:  [iter: 4/5,   epoch: 0/1]  eta: 0:00:02    time: 2.463502  data: 1.390873  max mem: 17030
+I20260215 07:14:39 4072 dinov2 helpers.py:217] Epoch 0/1 done in 12.32s
+
+I20260215 07:14:39 4072 dinov2 helpers.py:225] Test: Total time: 0:00:12 (2.463805 s / it)
+
+I20260215 07:14:39 4072 dinov2 utils.py:56] Averaged stats: 
+D20260215 07:14:39 4072 dinov2 utils.py:59] Post compute
+D20260215 07:14:39 4072 dinov2 knn.py:264] Finished KNN classification
+I20260215 07:14:39 4072 dinov2 knn.py:327] All metrics result:
+('full', 20): {acc_top-1_micro: 28.30, }
+I20260215 07:16:43 5293 dinov2 setup.py:34] task:
+  id: knn
+  is_multilabel: false
+  metrics:
+  - id: MulticlassAccuracy
+    top_k: 1
+    average: micro
+  backbone_to_features:
+    pooling: knn
+    use_n_blocks: 1
+heads:
+  nb_knn:
+  - 20
+  temperature: 0.07
+  gather_on_cpu: false
+  n_per_class_list:
+  - -1
+  n_tries: 1
+train_dataset:
+  id: geobench.m-eurosat
+  split: train
+  transform: []
+  normalize: false
+test_dataset:
+  id: geobench.m-eurosat
+  split: test
+  transform: []
+  normalize: false
+seed: 42
+optim:
+  dl:
+    batch_size: 200
+    num_workers: 4
+    persistent_workers: true
+_vars:
+  transforms: []
+output_dir: /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000/5min/m-eurosat_knn_nanochat
+
+I20260215 07:16:43 5293 dinov2 wrapper.py:39] Built model nanochat_fmvit
+I20260215 07:16:45 5293 dinov2 loaders.py:72] Building dataset "geobench.m-eurosat" ...
+I20260215 07:16:45 5293 dinov2 augmentations.py:44] Augmentations in order: []
+I20260215 07:16:45 5293 dinov2 loaders.py:161] Built dataset "geobench.m-eurosat" with #samples 2000
+I20260215 07:16:45 5293 dinov2 loaders.py:72] Building dataset "geobench.m-eurosat" ...
+I20260215 07:16:45 5293 dinov2 augmentations.py:44] Augmentations in order: []
+I20260215 07:16:45 5293 dinov2 loaders.py:161] Built dataset "geobench.m-eurosat" with #samples 1000
+I20260215 07:16:45 5293 dinov2 knn.py:224] Extracting features for train set...
+I20260215 07:16:45 5293 dinov2 loaders.py:328] Detected non-CombinedDataset. Using SamplerType.EPOCH with bsz=200.
+I20260215 07:16:45 5293 dinov2 loaders.py:234] sampler: epoch
+I20260215 07:16:45 5293 dinov2 loaders.py:238] # of samples / epoch: 2,000
+I20260215 07:16:45 5293 dinov2 loaders.py:344] DataLoader kwargs: num_workers=4, pin_memory=True, drop_last=False, persistent_workers=True
+I20260215 07:16:45 5293 dinov2 loaders.py:356] # of batches: 10
+I20260215 07:16:52 5293 dinov2 utils.py:115] Storing features into tensor of shape torch.Size([2000, 768])
+I20260215 07:16:53 5293 dinov2 helpers.py:190]   [iter:  0/10,   epoch: 0/1]  eta: 0:01:10    time: 7.087961  data: 5.525864  max mem: 17024
+I20260215 07:17:03 5293 dinov2 helpers.py:190]   [iter:  9/10,   epoch: 0/1]  eta: 0:00:01    time: 1.728850  data: 0.788224  max mem: 17031
+I20260215 07:17:03 5293 dinov2 helpers.py:217] Epoch 0/1 done in 17.29s
+
+I20260215 07:17:03 5293 dinov2 helpers.py:225]  Total time: 0:00:17 (1.729026 s / it)
+
+I20260215 07:17:03 5293 dinov2 utils.py:127] Features shape: (2000, 768)
+I20260215 07:17:03 5293 dinov2 utils.py:128] Labels shape: (2000,)
+I20260215 07:17:03 5293 dinov2 loaders.py:328] Detected non-CombinedDataset. Using SamplerType.EPOCH with bsz=200.
+I20260215 07:17:03 5293 dinov2 loaders.py:234] sampler: epoch
+I20260215 07:17:03 5293 dinov2 loaders.py:238] # of samples / epoch: 1,000
+I20260215 07:17:03 5293 dinov2 loaders.py:344] DataLoader kwargs: num_workers=4, pin_memory=True, drop_last=False, persistent_workers=True
+I20260215 07:17:03 5293 dinov2 loaders.py:356] # of batches: 5
+I20260215 07:17:03 5293 dinov2 knn.py:241] Using knn module: <class 'dinov2.eval.knn.KnnModule'> with num_classes 10
+I20260215 07:17:03 5293 dinov2 knn.py:262] Start the k-NN classification.
+I20260215 07:17:11 5293 dinov2 helpers.py:190] Test:  [iter: 0/5,   epoch: 0/1]  eta: 0:00:36    time: 7.231919  data: 6.124722  max mem: 17031
+I20260215 07:17:16 5293 dinov2 helpers.py:190] Test:  [iter: 4/5,   epoch: 0/1]  eta: 0:00:02    time: 2.510810  data: 1.437286  max mem: 17031
+I20260215 07:17:16 5293 dinov2 helpers.py:217] Epoch 0/1 done in 12.56s
+
+I20260215 07:17:16 5293 dinov2 helpers.py:225] Test: Total time: 0:00:12 (2.511179 s / it)
+
+I20260215 07:17:16 5293 dinov2 utils.py:56] Averaged stats: 
+D20260215 07:17:16 5293 dinov2 utils.py:59] Post compute
+D20260215 07:17:16 5293 dinov2 knn.py:264] Finished KNN classification
+I20260215 07:17:16 5293 dinov2 knn.py:327] All metrics result:
+('full', 20): {acc_top-1_micro: 28.30, }
+I20260215 07:23:32 8681 dinov2 setup.py:34] task:
+  id: knn
+  is_multilabel: false
+  metrics:
+  - id: MulticlassAccuracy
+    top_k: 1
+    average: micro
+  backbone_to_features:
+    pooling: knn
+    use_n_blocks: 1
+heads:
+  nb_knn:
+  - 20
+  temperature: 0.07
+  gather_on_cpu: false
+  n_per_class_list:
+  - -1
+  n_tries: 1
+train_dataset:
+  id: geobench.m-eurosat
+  split: train
+  transform: []
+  normalize: false
+test_dataset:
+  id: geobench.m-eurosat
+  split: test
+  transform: []
+  normalize: false
+seed: 42
+optim:
+  dl:
+    batch_size: 200
+    num_workers: 4
+    persistent_workers: true
+_vars:
+  transforms: []
+output_dir: /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000/5min/m-eurosat_knn_nanochat
+
+I20260215 07:23:32 8681 dinov2 wrapper.py:39] Built model nanochat_fmvit
+I20260215 07:23:34 8681 dinov2 loaders.py:72] Building dataset "geobench.m-eurosat" ...
+I20260215 07:23:34 8681 dinov2 augmentations.py:44] Augmentations in order: []
+I20260215 07:23:34 8681 dinov2 loaders.py:161] Built dataset "geobench.m-eurosat" with #samples 2000
+I20260215 07:23:34 8681 dinov2 loaders.py:72] Building dataset "geobench.m-eurosat" ...
+I20260215 07:23:34 8681 dinov2 augmentations.py:44] Augmentations in order: []
+I20260215 07:23:34 8681 dinov2 loaders.py:161] Built dataset "geobench.m-eurosat" with #samples 1000
+I20260215 07:23:34 8681 dinov2 knn.py:224] Extracting features for train set...
+I20260215 07:23:34 8681 dinov2 loaders.py:328] Detected non-CombinedDataset. Using SamplerType.EPOCH with bsz=200.
+I20260215 07:23:34 8681 dinov2 loaders.py:234] sampler: epoch
+I20260215 07:23:34 8681 dinov2 loaders.py:238] # of samples / epoch: 2,000
+I20260215 07:23:34 8681 dinov2 loaders.py:344] DataLoader kwargs: num_workers=4, pin_memory=True, drop_last=False, persistent_workers=True
+I20260215 07:23:34 8681 dinov2 loaders.py:356] # of batches: 10
+I20260215 07:23:42 8681 dinov2 utils.py:115] Storing features into tensor of shape torch.Size([2000, 768])
+I20260215 07:23:42 8681 dinov2 helpers.py:190]   [iter:  0/10,   epoch: 0/1]  eta: 0:01:19    time: 7.954972  data: 6.260513  max mem: 17024
+I20260215 07:23:53 8681 dinov2 helpers.py:190]   [iter:  9/10,   epoch: 0/1]  eta: 0:00:01    time: 1.950771  data: 0.992408  max mem: 17031
+I20260215 07:23:53 8681 dinov2 helpers.py:217] Epoch 0/1 done in 19.51s
+
+I20260215 07:23:53 8681 dinov2 helpers.py:225]  Total time: 0:00:19 (1.950971 s / it)
+
+I20260215 07:23:53 8681 dinov2 utils.py:127] Features shape: (2000, 768)
+I20260215 07:23:53 8681 dinov2 utils.py:128] Labels shape: (2000,)
+I20260215 07:23:54 8681 dinov2 loaders.py:328] Detected non-CombinedDataset. Using SamplerType.EPOCH with bsz=200.
+I20260215 07:23:54 8681 dinov2 loaders.py:234] sampler: epoch
+I20260215 07:23:54 8681 dinov2 loaders.py:238] # of samples / epoch: 1,000
+I20260215 07:23:54 8681 dinov2 loaders.py:344] DataLoader kwargs: num_workers=4, pin_memory=True, drop_last=False, persistent_workers=True
+I20260215 07:23:54 8681 dinov2 loaders.py:356] # of batches: 5
+I20260215 07:23:54 8681 dinov2 knn.py:241] Using knn module: <class 'dinov2.eval.knn.KnnModule'> with num_classes 10
+I20260215 07:23:54 8681 dinov2 knn.py:262] Start the k-NN classification.
+I20260215 07:24:01 8681 dinov2 helpers.py:190] Test:  [iter: 0/5,   epoch: 0/1]  eta: 0:00:35    time: 7.127646  data: 6.022355  max mem: 17031
+I20260215 07:24:06 8681 dinov2 helpers.py:190] Test:  [iter: 4/5,   epoch: 0/1]  eta: 0:00:02    time: 2.476352  data: 1.404564  max mem: 17031
+I20260215 07:24:06 8681 dinov2 helpers.py:217] Epoch 0/1 done in 12.38s
+
+I20260215 07:24:06 8681 dinov2 helpers.py:225] Test: Total time: 0:00:12 (2.476661 s / it)
+
+I20260215 07:24:06 8681 dinov2 utils.py:56] Averaged stats: 
+D20260215 07:24:06 8681 dinov2 utils.py:59] Post compute
+D20260215 07:24:06 8681 dinov2 knn.py:264] Finished KNN classification
+I20260215 07:24:06 8681 dinov2 knn.py:327] All metrics result:
+('full', 20): {acc_top-1_micro: 28.30, }
+I20260215 07:30:08 11131 dinov2 setup.py:34] task:
+  id: knn
+  is_multilabel: false
+  metrics:
+  - id: MulticlassAccuracy
+    top_k: 1
+    average: micro
+  backbone_to_features:
+    pooling: knn
+    use_n_blocks: 1
+heads:
+  nb_knn:
+  - 20
+  temperature: 0.07
+  gather_on_cpu: false
+  n_per_class_list:
+  - -1
+  n_tries: 1
+train_dataset:
+  id: geobench.m-eurosat
+  split: train
+  transform: []
+  normalize: false
+test_dataset:
+  id: geobench.m-eurosat
+  split: test
+  transform: []
+  normalize: false
+seed: 42
+optim:
+  dl:
+    batch_size: 200
+    num_workers: 4
+    persistent_workers: true
+_vars:
+  transforms: []
+output_dir: /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000/5min/m-eurosat_knn_nanochat
+
+I20260215 07:30:08 11131 dinov2 wrapper.py:39] Built model nanochat_fmvit
+I20260215 07:30:10 11131 dinov2 loaders.py:72] Building dataset "geobench.m-eurosat" ...
+I20260215 07:30:10 11131 dinov2 augmentations.py:44] Augmentations in order: []
+I20260215 07:30:10 11131 dinov2 loaders.py:161] Built dataset "geobench.m-eurosat" with #samples 2000
+I20260215 07:30:10 11131 dinov2 loaders.py:72] Building dataset "geobench.m-eurosat" ...
+I20260215 07:30:10 11131 dinov2 augmentations.py:44] Augmentations in order: []
+I20260215 07:30:10 11131 dinov2 loaders.py:161] Built dataset "geobench.m-eurosat" with #samples 1000
+I20260215 07:30:10 11131 dinov2 knn.py:224] Extracting features for train set...
+I20260215 07:30:10 11131 dinov2 loaders.py:328] Detected non-CombinedDataset. Using SamplerType.EPOCH with bsz=200.
+I20260215 07:30:10 11131 dinov2 loaders.py:234] sampler: epoch
+I20260215 07:30:10 11131 dinov2 loaders.py:238] # of samples / epoch: 2,000
+I20260215 07:30:10 11131 dinov2 loaders.py:344] DataLoader kwargs: num_workers=4, pin_memory=True, drop_last=False, persistent_workers=True
+I20260215 07:30:10 11131 dinov2 loaders.py:356] # of batches: 10
+I20260215 07:30:17 11131 dinov2 utils.py:115] Storing features into tensor of shape torch.Size([2000, 768])
+I20260215 07:30:17 11131 dinov2 helpers.py:190]   [iter:  0/10,   epoch: 0/1]  eta: 0:01:14    time: 7.451039  data: 5.838424  max mem: 17024
+I20260215 07:30:28 11131 dinov2 helpers.py:190]   [iter:  9/10,   epoch: 0/1]  eta: 0:00:01    time: 1.806605  data: 0.856597  max mem: 17031
+I20260215 07:30:28 11131 dinov2 helpers.py:217] Epoch 0/1 done in 18.07s
+
+I20260215 07:30:28 11131 dinov2 helpers.py:225]  Total time: 0:00:18 (1.806776 s / it)
+
+I20260215 07:30:28 11131 dinov2 utils.py:127] Features shape: (2000, 768)
+I20260215 07:30:28 11131 dinov2 utils.py:128] Labels shape: (2000,)
+I20260215 07:30:29 11131 dinov2 loaders.py:328] Detected non-CombinedDataset. Using SamplerType.EPOCH with bsz=200.
+I20260215 07:30:29 11131 dinov2 loaders.py:234] sampler: epoch
+I20260215 07:30:29 11131 dinov2 loaders.py:238] # of samples / epoch: 1,000
+I20260215 07:30:29 11131 dinov2 loaders.py:344] DataLoader kwargs: num_workers=4, pin_memory=True, drop_last=False, persistent_workers=True
+I20260215 07:30:29 11131 dinov2 loaders.py:356] # of batches: 5
+I20260215 07:30:29 11131 dinov2 knn.py:241] Using knn module: <class 'dinov2.eval.knn.KnnModule'> with num_classes 10
+I20260215 07:30:29 11131 dinov2 knn.py:262] Start the k-NN classification.
+I20260215 07:30:35 11131 dinov2 helpers.py:190] Test:  [iter: 0/5,   epoch: 0/1]  eta: 0:00:33    time: 6.648186  data: 5.541687  max mem: 17031
+I20260215 07:30:41 11131 dinov2 helpers.py:190] Test:  [iter: 4/5,   epoch: 0/1]  eta: 0:00:02    time: 2.401287  data: 1.329063  max mem: 17031
+I20260215 07:30:41 11131 dinov2 helpers.py:217] Epoch 0/1 done in 12.01s
+
+I20260215 07:30:41 11131 dinov2 helpers.py:225] Test: Total time: 0:00:12 (2.401601 s / it)
+
+I20260215 07:30:41 11131 dinov2 utils.py:56] Averaged stats: 
+D20260215 07:30:41 11131 dinov2 utils.py:59] Post compute
+D20260215 07:30:41 11131 dinov2 knn.py:264] Finished KNN classification
+I20260215 07:30:41 11131 dinov2 knn.py:327] All metrics result:
+('full', 20): {acc_top-1_micro: 65.40, }

fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000/5min/m-eurosat_knn_nanochat/results.csv

@@ -0,0 +1,2 @@
+metric,best_classifier,value
+acc_top-1_micro,"('full', 20)",65.4

fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000/5min/m-eurosat_knn_nanochat/results_eval_knn.json

@@ -0,0 +1,5 @@
+{"('full', 20)": "{'acc_top-1_micro': 56.3}"}
+{"('full', 20)": "{'acc_top-1_micro': 28.3}"}
+{"('full', 20)": "{'acc_top-1_micro': 28.3}"}
+{"('full', 20)": "{'acc_top-1_micro': 28.3}"}
+{"('full', 20)": "{'acc_top-1_micro': 65.4}"}

fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000/5min/results.csv

@@ -0,0 +1,3 @@
+,metric,best_classifier,value,relpath
+0,acc_top-1_micro,"('full', 20)",56.9,m-eurosat_knn
+1,acc_top-1_micro,"('full', 20)",65.4,m-eurosat_knn_nanochat

fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000/5min/results.txt

@@ -0,0 +1,4 @@
+                                       best_classifier  value
+lvl0                   metric                                
+m-eurosat_knn          acc_top-1_micro    ('full', 20)   56.9
+m-eurosat_knn_nanochat acc_top-1_micro    ('full', 20)   65.4

fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000/log

@@ -0,0 +1,181 @@
+I20260214 22:33:54 49291 eval.m000 eval.py:365] ------------------------------------
+I20260214 22:33:54 49291 eval.m000 eval.py:366] 2026-02-14 22:33:54, root
+I20260214 22:33:54 49291 eval.m000 eval.py:367] model-obj:  /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/model_000000.pt
+I20260214 22:33:54 49291 eval.m000 eval.py:369] config-obj: dinov2/configs/eval/5min/m-eurosat_knn.yaml
+I20260214 22:33:54 49291 eval.m000 eval.py:370] output-dir: /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000
+I20260214 22:33:54 49291 eval.m000 eval.py:371] overwrite:  True
+I20260214 22:33:54 49291 eval.m000 eval.py:372] ------------------------------------
+I20260214 22:33:54 49291 eval.m000 eval.py:418] Running  5min/m-eurosat_knn ... (rank 0/1)
+
+
+
+
+I20260214 22:35:04 49748 eval.m000 eval.py:365] ------------------------------------
+I20260214 22:35:04 49748 eval.m000 eval.py:366] 2026-02-14 22:35:04, root
+I20260214 22:35:04 49748 eval.m000 eval.py:367] model-obj:  /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/model_000000.pt
+I20260214 22:35:04 49748 eval.m000 eval.py:369] config-obj: dinov2/configs/eval/5min/m-eurosat_knn.yaml
+I20260214 22:35:04 49748 eval.m000 eval.py:370] output-dir: /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000
+I20260214 22:35:04 49748 eval.m000 eval.py:371] overwrite:  True
+I20260214 22:35:04 49748 eval.m000 eval.py:372] ------------------------------------
+I20260214 22:35:04 49748 eval.m000 eval.py:418] Running  5min/m-eurosat_knn ... (rank 0/1)
+
+
+
+
+I20260214 22:36:08 50158 eval.m000 eval.py:365] ------------------------------------
+I20260214 22:36:08 50158 eval.m000 eval.py:366] 2026-02-14 22:36:08, root
+I20260214 22:36:08 50158 eval.m000 eval.py:367] model-obj:  /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/model_000000.pt
+I20260214 22:36:08 50158 eval.m000 eval.py:369] config-obj: dinov2/configs/eval/5min/m-eurosat_knn.yaml
+I20260214 22:36:08 50158 eval.m000 eval.py:370] output-dir: /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000
+I20260214 22:36:08 50158 eval.m000 eval.py:371] overwrite:  True
+I20260214 22:36:08 50158 eval.m000 eval.py:372] ------------------------------------
+I20260214 22:36:08 50158 eval.m000 eval.py:418] Running  5min/m-eurosat_knn ... (rank 0/1)
+I20260214 22:36:46 50158 eval.m000 eval.py:421] Finished 5min/m-eurosat_knn in 38.22s (rank 0/1)
+I20260214 22:36:46 50158 eval.m000 eval.py:431] All tasks done.
+I20260214 22:36:46 50158 eval.m000 eval.py:432] 
+                                   best_classifier  value
+lvl0 lvl1          metric                                
+5min m-eurosat_knn acc_top-1_micro    ('full', 20)   70.5
+
+
+
+
+I20260214 22:38:04 50775 eval.m000 eval.py:365] ------------------------------------
+I20260214 22:38:04 50775 eval.m000 eval.py:366] 2026-02-14 22:38:04, root
+I20260214 22:38:04 50775 eval.m000 eval.py:367] model-obj:  /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/model_000000.pt
+I20260214 22:38:04 50775 eval.m000 eval.py:369] config-obj: dinov2/configs/eval/5min/m-eurosat_knn.yaml
+I20260214 22:38:04 50775 eval.m000 eval.py:370] output-dir: /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000
+I20260214 22:38:04 50775 eval.m000 eval.py:371] overwrite:  True
+I20260214 22:38:04 50775 eval.m000 eval.py:372] ------------------------------------
+I20260214 22:38:04 50775 eval.m000 eval.py:418] Running  5min/m-eurosat_knn ... (rank 0/1)
+I20260214 22:38:41 50775 eval.m000 eval.py:421] Finished 5min/m-eurosat_knn in 37.31s (rank 0/1)
+I20260214 22:38:41 50775 eval.m000 eval.py:431] All tasks done.
+I20260214 22:38:41 50775 eval.m000 eval.py:432] 
+                                   best_classifier  value
+lvl0 lvl1          metric                                
+5min m-eurosat_knn acc_top-1_micro    ('full', 20)   58.0
+
+
+
+
+I20260214 22:40:26 51715 eval.m000 eval.py:365] ------------------------------------
+I20260214 22:40:26 51715 eval.m000 eval.py:366] 2026-02-14 22:40:26, root
+I20260214 22:40:26 51715 eval.m000 eval.py:367] model-obj:  /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/model_000000.pt
+I20260214 22:40:26 51715 eval.m000 eval.py:369] config-obj: dinov2/configs/eval/5min/m-eurosat_knn.yaml
+I20260214 22:40:26 51715 eval.m000 eval.py:370] output-dir: /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000
+I20260214 22:40:26 51715 eval.m000 eval.py:371] overwrite:  True
+I20260214 22:40:26 51715 eval.m000 eval.py:372] ------------------------------------
+I20260214 22:40:26 51715 eval.m000 eval.py:418] Running  5min/m-eurosat_knn ... (rank 0/1)
+I20260214 22:41:02 51715 eval.m000 eval.py:421] Finished 5min/m-eurosat_knn in 36.08s (rank 0/1)
+I20260214 22:41:02 51715 eval.m000 eval.py:431] All tasks done.
+I20260214 22:41:02 51715 eval.m000 eval.py:432] 
+                                   best_classifier  value
+lvl0 lvl1          metric                                
+5min m-eurosat_knn acc_top-1_micro    ('full', 20)   56.9
+
+
+
+
+I20260215 07:10:15 1984 eval.m000 eval.py:365] ------------------------------------
+I20260215 07:10:15 1984 eval.m000 eval.py:366] 2026-02-15 07:10:15, root
+I20260215 07:10:15 1984 eval.m000 eval.py:367] model-obj:  /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/model_000000.pt
+I20260215 07:10:15 1984 eval.m000 eval.py:369] config-obj: dinov2/configs/eval/5min/m-eurosat_knn_nanochat.yaml
+I20260215 07:10:15 1984 eval.m000 eval.py:370] output-dir: /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000
+I20260215 07:10:15 1984 eval.m000 eval.py:371] overwrite:  True
+I20260215 07:10:15 1984 eval.m000 eval.py:372] ------------------------------------
+I20260215 07:10:15 1984 eval.m000 eval.py:418] Running  5min/m-eurosat_knn_nanochat ... (rank 0/1)
+I20260215 07:11:21 1984 eval.m000 eval.py:421] Finished 5min/m-eurosat_knn_nanochat in 65.61s (rank 0/1)
+I20260215 07:11:21 1984 eval.m000 eval.py:431] All tasks done.
+I20260215 07:11:21 1984 eval.m000 eval.py:432] 
+                                            best_classifier  value
+lvl0 lvl1                   metric                                
+5min m-eurosat_knn          acc_top-1_micro    ('full', 20)   56.9
+     m-eurosat_knn_nanochat acc_top-1_micro    ('full', 20)   56.3
+
+
+
+
+I20260215 07:13:03 3370 eval.m000 eval.py:365] ------------------------------------
+I20260215 07:13:03 3370 eval.m000 eval.py:366] 2026-02-15 07:13:03, root
+I20260215 07:13:03 3370 eval.m000 eval.py:367] model-obj:  /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/model_000000.pt
+I20260215 07:13:03 3370 eval.m000 eval.py:369] config-obj: dinov2/configs/eval/5min/m-eurosat_knn_nanochat.yaml
+I20260215 07:13:03 3370 eval.m000 eval.py:370] output-dir: /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000
+I20260215 07:13:03 3370 eval.m000 eval.py:371] overwrite:  True
+I20260215 07:13:03 3370 eval.m000 eval.py:372] ------------------------------------
+I20260215 07:13:03 3370 eval.m000 eval.py:418] Running  5min/m-eurosat_knn_nanochat ... (rank 0/1)
+
+
+
+
+I20260215 07:14:03 4072 eval.m000 eval.py:365] ------------------------------------
+I20260215 07:14:03 4072 eval.m000 eval.py:366] 2026-02-15 07:14:03, root
+I20260215 07:14:03 4072 eval.m000 eval.py:367] model-obj:  /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/model_000000.pt
+I20260215 07:14:03 4072 eval.m000 eval.py:369] config-obj: dinov2/configs/eval/5min/m-eurosat_knn_nanochat.yaml
+I20260215 07:14:03 4072 eval.m000 eval.py:370] output-dir: /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000
+I20260215 07:14:03 4072 eval.m000 eval.py:371] overwrite:  True
+I20260215 07:14:03 4072 eval.m000 eval.py:372] ------------------------------------
+I20260215 07:14:03 4072 eval.m000 eval.py:418] Running  5min/m-eurosat_knn_nanochat ... (rank 0/1)
+I20260215 07:14:39 4072 eval.m000 eval.py:421] Finished 5min/m-eurosat_knn_nanochat in 35.51s (rank 0/1)
+I20260215 07:14:39 4072 eval.m000 eval.py:431] All tasks done.
+I20260215 07:14:39 4072 eval.m000 eval.py:432] 
+                                            best_classifier  value
+lvl0 lvl1                   metric                                
+5min m-eurosat_knn          acc_top-1_micro    ('full', 20)   56.9
+     m-eurosat_knn_nanochat acc_top-1_micro    ('full', 20)   28.3
+
+
+
+
+I20260215 07:16:43 5293 eval.m000 eval.py:365] ------------------------------------
+I20260215 07:16:43 5293 eval.m000 eval.py:366] 2026-02-15 07:16:43, root
+I20260215 07:16:43 5293 eval.m000 eval.py:367] model-obj:  /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/model_000000.pt
+I20260215 07:16:43 5293 eval.m000 eval.py:369] config-obj: dinov2/configs/eval/5min/m-eurosat_knn_nanochat.yaml
+I20260215 07:16:43 5293 eval.m000 eval.py:370] output-dir: /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000
+I20260215 07:16:43 5293 eval.m000 eval.py:371] overwrite:  True
+I20260215 07:16:43 5293 eval.m000 eval.py:372] ------------------------------------
+I20260215 07:16:43 5293 eval.m000 eval.py:418] Running  5min/m-eurosat_knn_nanochat ... (rank 0/1)
+I20260215 07:17:16 5293 eval.m000 eval.py:421] Finished 5min/m-eurosat_knn_nanochat in 33.25s (rank 0/1)
+I20260215 07:17:16 5293 eval.m000 eval.py:431] All tasks done.
+I20260215 07:17:16 5293 eval.m000 eval.py:432] 
+                                            best_classifier  value
+lvl0 lvl1                   metric                                
+5min m-eurosat_knn          acc_top-1_micro    ('full', 20)   56.9
+     m-eurosat_knn_nanochat acc_top-1_micro    ('full', 20)   28.3
+
+
+
+
+I20260215 07:23:32 8681 eval.m000 eval.py:365] ------------------------------------
+I20260215 07:23:32 8681 eval.m000 eval.py:366] 2026-02-15 07:23:32, root
+I20260215 07:23:32 8681 eval.m000 eval.py:367] model-obj:  /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/model_000000.pt
+I20260215 07:23:32 8681 eval.m000 eval.py:369] config-obj: dinov2/configs/eval/5min/m-eurosat_knn_nanochat.yaml
+I20260215 07:23:32 8681 eval.m000 eval.py:370] output-dir: /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000
+I20260215 07:23:32 8681 eval.m000 eval.py:371] overwrite:  True
+I20260215 07:23:32 8681 eval.m000 eval.py:372] ------------------------------------
+I20260215 07:23:32 8681 eval.m000 eval.py:418] Running  5min/m-eurosat_knn_nanochat ... (rank 0/1)
+I20260215 07:24:06 8681 eval.m000 eval.py:421] Finished 5min/m-eurosat_knn_nanochat in 34.36s (rank 0/1)
+I20260215 07:24:06 8681 eval.m000 eval.py:431] All tasks done.
+I20260215 07:24:06 8681 eval.m000 eval.py:432] 
+                                            best_classifier  value
+lvl0 lvl1                   metric                                
+5min m-eurosat_knn          acc_top-1_micro    ('full', 20)   56.9
+     m-eurosat_knn_nanochat acc_top-1_micro    ('full', 20)   28.3
+
+
+
+
+I20260215 07:30:08 11131 eval.m000 eval.py:365] ------------------------------------
+I20260215 07:30:08 11131 eval.m000 eval.py:366] 2026-02-15 07:30:08, root
+I20260215 07:30:08 11131 eval.m000 eval.py:367] model-obj:  /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/model_000000.pt
+I20260215 07:30:08 11131 eval.m000 eval.py:369] config-obj: dinov2/configs/eval/5min/m-eurosat_knn_nanochat.yaml
+I20260215 07:30:08 11131 eval.m000 eval.py:370] output-dir: /workspace/Spatial/nanochat_artifacts/fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000
+I20260215 07:30:08 11131 eval.m000 eval.py:371] overwrite:  True
+I20260215 07:30:08 11131 eval.m000 eval.py:372] ------------------------------------
+I20260215 07:30:08 11131 eval.m000 eval.py:418] Running  5min/m-eurosat_knn_nanochat ... (rank 0/1)
+I20260215 07:30:41 11131 eval.m000 eval.py:421] Finished 5min/m-eurosat_knn_nanochat in 32.58s (rank 0/1)
+I20260215 07:30:41 11131 eval.m000 eval.py:431] All tasks done.
+I20260215 07:30:41 11131 eval.m000 eval.py:432] 
+                                            best_classifier  value
+lvl0 lvl1                   metric                                
+5min m-eurosat_knn          acc_top-1_micro    ('full', 20)   56.9
+     m-eurosat_knn_nanochat acc_top-1_micro    ('full', 20)   65.4

fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000/results.csv

@@ -0,0 +1,3 @@
+,metric,best_classifier,value,relpath
+0,acc_top-1_micro,"('full', 20)",56.9,5min/m-eurosat_knn
+1,acc_top-1_micro,"('full', 20)",65.4,5min/m-eurosat_knn_nanochat

fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/eval_step000000/results.txt

@@ -0,0 +1,4 @@
+                                            best_classifier  value
+lvl0 lvl1                   metric                                
+5min m-eurosat_knn          acc_top-1_micro    ('full', 20)   56.9
+     m-eurosat_knn_nanochat acc_top-1_micro    ('full', 20)   65.4

fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/meta_000000.json

@@ -0,0 +1,140 @@
+{
+  "step": 0,
+  "model_type": "fm_vit",
+  "model_config": {
+    "embed_dim": 768,
+    "depth": 12,
+    "num_heads": 12,
+    "mlp_ratio": 4.0,
+    "dropout": 0.0,
+    "attn_dropout": 0.0,
+    "patch_size": 16,
+    "modalities": [
+      "sentinel2_l2a",
+      "sentinel1",
+      "landsat"
+    ],
+    "modality_channels": {
+      "sentinel2_l2a": 12,
+      "sentinel1": 2,
+      "landsat": 11
+    }
+  },
+  "user_config": {
+    "h5_dir": "/workspace/olmoearth/h5_shards",
+    "s2_key": "sentinel2_l2a",
+    "s1_key": "sentinel1",
+    "landsat_key": "landsat",
+    "num_timesteps": 12,
+    "timestep_strategy": "uniform",
+    "patch_size": 16,
+    "mask_ratio": 0.75,
+    "s2_value_scale": 0.0001,
+    "s1_value_scale": 1.0,
+    "landsat_value_scale": 0.0001,
+    "norm": "computed",
+    "norm_stats_file": "",
+    "norm_std_multiplier": 2.0,
+    "num_workers": 120,
+    "prefetch": 512,
+    "shuffle_buffer": 10000,
+    "depth": 12,
+    "embed_dim": 768,
+    "num_heads": 12,
+    "mlp_ratio": 4.0,
+    "dropout": 0.0,
+    "attn_dropout": 0.0,
+    "contrastive_weight": 0.1,
+    "contrastive_temp": 0.1,
+    "contrastive_proj_dim": 256,
+    "contrastive_lr": 0.001,
+    "contrastive_queue_size": 4096,
+    "contrastive_pooling": "mean_unmasked",
+    "spatial_aug": "flip_rotate",
+    "drop_s1_p": 0.3,
+    "drop_landsat_p": 0.3,
+    "distill_weight": 1.0,
+    "distill_every": 1,
+    "distill_teacher_patch_size": 4,
+    "distill_teacher_expected_hw": 64,
+    "distill_teacher_timesteps": 1,
+    "distill_olmoearth_src_dir": "/workspace/Spatial/olmoearth_pretrain",
+    "distill_olmoearth_config": "/workspace/Spatial/nanochat_artifacts/teachers/olmoearth_v1_base/config.json",
+    "distill_olmoearth_weights": "/workspace/Spatial/nanochat_artifacts/teachers/olmoearth_v1_base/weights.pth",
+    "run": "fmvit_d12_e768_ps16_normcomputed_contrast_distill_from1200_fixed",
+    "device_type": "",
+    "num_iterations": 3000,
+    "device_batch_size": 2,
+    "grad_accum_steps": 8,
+    "lr": 0.0001,
+    "weight_decay": 0.05,
+    "adam_beta1": 0.9,
+    "adam_beta2": 0.95,
+    "warmup_ratio": 0.01,
+    "warmdown_ratio": 0.4,
+    "final_lr_frac": 0.0,
+    "grad_clip": 1.0,
+    "resume_from_step": 1200,
+    "eval_every": 200,
+    "eval_steps": 25,
+    "save_every": 200,
+    "model_tag": "fmvit_d12_e768_ps16_normcomputed_contrast"
+  },
+  "dataset_meta": {
+    "h5_dir": "/workspace/olmoearth/h5_shards",
+    "s2_key": "sentinel2_l2a",
+    "s1_key": "sentinel1",
+    "landsat_key": "landsat",
+    "num_timesteps": 12,
+    "timestep_strategy": "uniform",
+    "patch_size": 16,
+    "grid_size": 8,
+    "mask_ratio": 0.75,
+    "value_scales": {
+      "sentinel2_l2a": 0.0001,
+      "sentinel1": 1.0,
+      "landsat": 0.0001
+    },
+    "norm": {
+      "strategy": "computed",
+      "stats_file": "/workspace/Spatial/olmoearth_pretrain/olmoearth_pretrain/data/norm_configs/computed.json",
+      "std_multiplier": 2.0
+    },
+    "modalities": [
+      "sentinel2_l2a",
+      "sentinel1",
+      "landsat"
+    ],
+    "modality_channels": {
+      "sentinel2_l2a": 12,
+      "sentinel1": 2,
+      "landsat": 11
+    },
+    "contrastive": {
+      "weight": 0.1,
+      "temp": 0.1,
+      "proj_dim": 256,
+      "queue_size": 4096,
+      "pooling": "mean_unmasked",
+      "spatial_aug": "flip_rotate",
+      "drop_s1_p": 0.3,
+      "drop_landsat_p": 0.3
+    },
+    "distill": {
+      "weight": 1.0,
+      "every": 1,
+      "teacher_patch_size": 4,
+      "teacher_expected_hw": 64,
+      "teacher_timesteps": 1,
+      "olmoearth_config": "/workspace/Spatial/nanochat_artifacts/teachers/olmoearth_v1_base/config.json",
+      "olmoearth_weights": "/workspace/Spatial/nanochat_artifacts/teachers/olmoearth_v1_base/weights.pth"
+    }
+  },
+  "grad_accum_steps": 8,
+  "dataloader_state_dict": null,
+  "loop_state": {
+    "min_val_loss": null,
+    "smooth_train_loss": null,
+    "total_training_time": 0.0
+  }
+}

fm_checkpoints/fmvit_d12_e768_ps16_normcomputed_contrast_panopticon_init/model_000000.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7ffb7058aedc7c1d95f2a31597b4da68bdcc4fe9bf15dfb6bc54d84d278e6187
+size 606637037