Script started on 2026-07-01 12:54:05+00:00 [TERM="screen" TTY="/dev/pts/1" COLUMNS="121" LINES="36"]
[?2004h[01;34m/content/bnnr[00m# [200~python benchmarks/run_grand_benchmark.py --dataset imagewoof --smoke[201~[3mpython benchmarks/run_grand_benchmark.py --dataset imagewoof --smoke[23m 
[C[C[C[C[C[C[C[C[C[C[C[C[C[C[Cpython benchmarks/run_grand_benchmark.py --dataset imagewoof --smoke ----ddeevviiccee  ccppuu

[?2004l
Grand Benchmark — BNNR cross-dataset ablation (paper-quality)
  dataset=imagewoof  regime=scratch  arch=resnet18
  pretrained=False  img_size=64
  num_classes=10  train_per_class=5
  seeds=[42, 43]  conditions=['no_aug', 'bnnr_xai']
  budget=4 epochs  epochs_per_phase=1  N_candidates=3  device=cpu
  ~0.3h wall-clock estimate (2 conditions x 2 seeds, cpu, budget=4)
  results -> /content/bnnr/benchmarks/results_imagewoof_scratch.json

>>> dataset=imagewoof  condition=no_aug  seed=42  regime=scratch

============================================================
  NO_AUG (grand benchmark)
  dataset=imagewoof  policy=base  batch_augs=none
  budget=4 epochs  seed=42  device=cpu
============================================================
  epoch 1/4 — accuracy=0.1018  loss=2.3302
  epoch 2/4 — accuracy=0.1069  loss=2.4741
  epoch 3/4 — accuracy=0.1124  loss=2.7635
  epoch 4/4 — accuracy=0.1069  loss=2.8052
    no_aug seed=42: val_metric(held_out)=0.1079

>>> dataset=imagewoof  condition=bnnr_xai  seed=42  regime=scratch

============================================================
  BNNR XAI (grand benchmark, equal-compute)
  dataset=imagewoof  budget=4  epochs_per_phase=1
  N_CANDIDATES=3  total_gpu_epochs=4
  seed=42  device=cpu
============================================================
  [Phase 0/baseline] 1 epochs (no extra aug)...
  [baseline] epoch 1/1 — accuracy=0.1018  loss=2.3302
  [Phase 0] baseline selection_val accuracy=0.1018
  [XAI cache] Pre-computing saliency maps...

Precomputing XAI cache: 0it [00:00, ?it/s]
Precomputing XAI cache: 1it [00:53, 53.34s/it]
                                              
  [XAI cache] 50 maps cached to /content/bnnr/benchmarks/runs_grand/imagewoof_scratch/20260701_125546_bnnr_xai_s42/xai_cache
  [Phase 1/3 — ICD] 1 epochs...
  [ICD] epoch 1/1 — accuracy=0.1165  loss=2.4627
  [ICD] selection_val accuracy=0.1165
  [Phase 2/3 — AICD] 1 epochs...
  [AICD] epoch 1/1 — accuracy=0.1084  loss=2.5188
  [AICD] selection_val accuracy=0.1084
  [Phase 3/3 — ChurchNoise] 1 epochs...
  [ChurchNoise] epoch 1/1 — accuracy=0.1076  loss=2.4763
  [ChurchNoise] selection_val accuracy=0.1076
  [XAI selection] best=ICD  score=0.1165
  [held_out_test] accuracy=0.1126  elapsed=112.5s
    bnnr_xai seed=42: val_metric(held_out)=0.1126

>>> dataset=imagewoof  condition=no_aug  seed=43  regime=scratch

============================================================
  NO_AUG (grand benchmark)
  dataset=imagewoof  policy=base  batch_augs=none
  budget=4 epochs  seed=43  device=cpu
============================================================
  epoch 1/4 — accuracy=0.0615  loss=2.9494
  epoch 2/4 — accuracy=0.0565  loss=4.5739
  epoch 3/4 — accuracy=0.0675  loss=3.9840
  epoch 4/4 — accuracy=0.0796  loss=3.3354
    no_aug seed=43: val_metric(held_out)=0.0822

>>> dataset=imagewoof  condition=bnnr_xai  seed=43  regime=scratch

============================================================
  BNNR XAI (grand benchmark, equal-compute)
  dataset=imagewoof  budget=4  epochs_per_phase=1
  N_CANDIDATES=3  total_gpu_epochs=4
  seed=43  device=cpu
============================================================
  [Phase 0/baseline] 1 epochs (no extra aug)...
  [baseline] epoch 1/1 — accuracy=0.0615  loss=2.9494
  [Phase 0] baseline selection_val accuracy=0.0615
  [XAI cache] Pre-computing saliency maps...

Precomputing XAI cache: 0it [00:00, ?it/s]
Precomputing XAI cache: 1it [00:50, 50.90s/it]
                                              
  [XAI cache] 50 maps cached to /content/bnnr/benchmarks/runs_grand/imagewoof_scratch/20260701_125932_bnnr_xai_s43/xai_cache
  [Phase 1/3 — ICD] 1 epochs...
  [ICD] epoch 1/1 — accuracy=0.0625  loss=2.4620
  [ICD] selection_val accuracy=0.0625
  [Phase 2/3 — AICD] 1 epochs...
  [AICD] epoch 1/1 — accuracy=0.0781  loss=2.4735
  [AICD] selection_val accuracy=0.0781
  [Phase 3/3 — ChurchNoise] 1 epochs...
  [ChurchNoise] epoch 1/1 — accuracy=0.0625  loss=2.4292
  [ChurchNoise] selection_val accuracy=0.0625
  [XAI selection] best=AICD  score=0.0781
  [held_out_test] accuracy=0.0806  elapsed=108.1s
    bnnr_xai seed=43: val_metric(held_out)=0.0806

Done. 4 valid run records in /content/bnnr/benchmarks/results_imagewoof_scratch.json
Summarize: python benchmarks/summarize_grand.py --results-dir benchmarks/ --datasets imagewoof
[?2004h[01;34m/content/bnnr[00m# nnvviiddiiaa--ssmmii

[?2004l
Wed Jul  1 13:02:16 2026       
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 580.82.07              Driver Version: 580.82.07      CUDA Version: 13.0     |
+-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  Tesla T4                       Off |   00000000:00:04.0 Off |                    0 |
| N/A   55C    P8             14W /   70W |       0MiB /  15360MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI              PID   Type   Process name                        GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|  No running processes found                                                             |
+-----------------------------------------------------------------------------------------+
[?2004h[01;34m/content/bnnr[00m# [200~python benchmarks/run_grand_benchmark.py --dataset eurosat --device cuda --seeds 67 --conditions no_aug,bnnr_xai[201~[3mpython benchmarks/run_grand_benchmark.py --dataset eurosat --device cuda --seeds 67 --conditions no_aug,bn[23m[3mn[23m[3mr_xai[23m
M
[01;34m/content/bnnr[00m# python benchmarks/run_grand_benchmark.py --dataset eurosat --device cuda --seeds 67 --conditions no_aug,bnnr_xai
[?2004l
Grand Benchmark — BNNR cross-dataset ablation (paper-quality)
  dataset=eurosat  regime=scratch  arch=resnet18
  pretrained=False  img_size=64
  num_classes=10  train_per_class=100
  seeds=[67]  conditions=['no_aug', 'bnnr_xai']
  budget=40 epochs  epochs_per_phase=10  N_candidates=3  device=cuda
  ~0.2h wall-clock estimate (2 conditions x 1 seeds, cuda, budget=40)
  results -> /content/bnnr/benchmarks/results_eurosat_scratch.json

>>> dataset=eurosat  condition=no_aug  seed=67  regime=scratch

============================================================
  NO_AUG (grand benchmark)
  dataset=eurosat  policy=base  batch_augs=none
  budget=40 epochs  seed=67  device=cuda
============================================================
  epoch 1/40 — accuracy=0.1088  loss=30.4302
  epoch 2/40 — accuracy=0.1209  loss=57.1866
  epoch 3/40 — accuracy=0.2683  loss=8.9870
  epoch 4/40 — accuracy=0.3843  loss=1.7308
  epoch 5/40 — accuracy=0.4512  loss=1.8884
  epoch 6/40 — accuracy=0.3826  loss=2.1028
  epoch 7/40 — accuracy=0.5658  loss=1.2276
  epoch 8/40 — accuracy=0.3244  loss=3.5604
  epoch 9/40 — accuracy=0.6287  loss=1.0734
  epoch 10/40 — accuracy=0.6177  loss=1.0859
  epoch 11/40 — accuracy=0.6026  loss=1.1420
  epoch 12/40 — accuracy=0.5296  loss=1.4103
  epoch 13/40 — accuracy=0.3932  loss=2.7714
  epoch 14/40 — accuracy=0.4023  loss=1.8475
  epoch 15/40 — accuracy=0.5891  loss=1.3953
  epoch 16/40 — accuracy=0.5658  loss=1.4213
  epoch 17/40 — accuracy=0.6451  loss=1.0508
  epoch 18/40 — accuracy=0.7144  loss=0.8206
  epoch 19/40 — accuracy=0.5610  loss=1.4072
  epoch 20/40 — accuracy=0.7206  loss=0.8106
  epoch 21/40 — accuracy=0.6557  loss=0.9876
  epoch 22/40 — accuracy=0.5895  loss=1.2706
  epoch 23/40 — accuracy=0.6346  loss=1.1078
  epoch 24/40 — accuracy=0.6994  loss=0.9284
  epoch 25/40 — accuracy=0.6438  loss=1.1665
  epoch 26/40 — accuracy=0.7500  loss=0.7290
  epoch 27/40 — accuracy=0.6958  loss=0.8989
  epoch 28/40 — accuracy=0.6965  loss=0.9381
  epoch 29/40 — accuracy=0.6560  loss=1.1084
  epoch 30/40 — accuracy=0.7306  loss=0.8111
  epoch 31/40 — accuracy=0.6985  loss=0.9328
  epoch 32/40 — accuracy=0.7373  loss=0.7981
  epoch 33/40 — accuracy=0.7464  loss=0.7808
  epoch 34/40 — accuracy=0.7535  loss=0.7736
  epoch 35/40 — accuracy=0.7259  loss=0.8701
  epoch 36/40 — accuracy=0.7505  loss=0.7787
  epoch 37/40 — accuracy=0.7583  loss=0.7718
  epoch 38/40 — accuracy=0.7531  loss=0.7870
  epoch 39/40 — accuracy=0.7590  loss=0.7735
  epoch 40/40 — accuracy=0.7619  loss=0.7511
    no_aug seed=67: val_metric(held_out)=0.7658

>>> dataset=eurosat  condition=bnnr_xai  seed=67  regime=scratch

============================================================
  BNNR XAI (grand benchmark, equal-compute)
  dataset=eurosat  budget=40  epochs_per_phase=10
  N_CANDIDATES=3  total_gpu_epochs=40
  seed=67  device=cuda
============================================================
  [Phase 0/baseline] 10 epochs (no extra aug)...
  [baseline] epoch 1/10 — accuracy=0.1088  loss=30.4302
  [baseline] epoch 2/10 — accuracy=0.1783  loss=35.6142
  [baseline] epoch 3/10 — accuracy=0.2931  loss=2.5134
^CTraceback (most recent call last):
  File "/content/bnnr/benchmarks/run_grand_benchmark.py", line 1595, in <module>
    main()
  File "/content/bnnr/benchmarks/run_grand_benchmark.py", line 1558, in main
    entry = run_condition(
            ^^^^^^^^^^^^^^
  File "/content/bnnr/benchmarks/run_grand_benchmark.py", line 1225, in run_condition
    return _run_bnnr_equal_compute(
           ^^^^^^^^^^^^^^^^^^^^^^^^
  File "/content/bnnr/benchmarks/run_grand_benchmark.py", line 1008, in _run_bnnr_equal_compute
    baseline_val, _, _ = _run_epochs_on_loader(
                         ^^^^^^^^^^^^^^^^^^^^^^
  File "/content/bnnr/benchmarks/run_grand_benchmark.py", line 383, in _run_epochs_on_loader
    train_epoch(trainer, train_loader, augmentations=augmentations if augmentations else None)
  File "/content/bnnr/src/bnnr/training/loop.py", line 75, in train_epoch
    metrics = trainer.model.train_step(batch)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/content/bnnr/src/bnnr/adapter.py", line 94, in train_step
    logits = self.model(images)
             ^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1779, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1790, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/content/bnnr/benchmarks/run_grand_benchmark.py", line 293, in forward
    return self.backbone(x)
           ^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1779, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1790, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/torchvision/models/resnet.py", line 285, in forward
    return self._forward_impl(x)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/torchvision/models/resnet.py", line 275, in _forward_impl
    x = self.layer3(x)
        ^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1779, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1790, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/container.py", line 253, in forward
    input = module(input)
            ^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1779, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1790, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/torchvision/models/resnet.py", line 92, in forward
    out = self.conv1(x)
          ^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1779, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1790, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/conv.py", line 553, in forward
    return self._conv_forward(input, self.weight, self.bias)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/conv.py", line 548, in _conv_forward
    return F.conv2d(
           ^^^^^^^^^
KeyboardInterrupt
[?2004h[01;34m/content/bnnr[00m# !!!!

[?2004l
python benchmarks/run_grand_benchmark.py --dataset eurosat --device cuda --seeds 67 --conditions no_aug,bnnr_xai
Grand Benchmark — BNNR cross-dataset ablation (paper-quality)
  dataset=eurosat  regime=scratch  arch=resnet18
  pretrained=False  img_size=64
  num_classes=10  train_per_class=100
  seeds=[67]  conditions=['no_aug', 'bnnr_xai']
  budget=40 epochs  epochs_per_phase=10  N_candidates=3  device=cuda
  ~0.2h wall-clock estimate (2 conditions x 1 seeds, cuda, budget=40)
  results -> /content/bnnr/benchmarks/results_eurosat_scratch.json
SKIP no_aug seed=67 regime=scratch (already done)

>>> dataset=eurosat  condition=bnnr_xai  seed=67  regime=scratch

============================================================
  BNNR XAI (grand benchmark, equal-compute)
  dataset=eurosat  budget=40  epochs_per_phase=10
  N_CANDIDATES=3  total_gpu_epochs=40
  seed=67  device=cuda
============================================================
  [Phase 0/baseline] 10 epochs (no extra aug)...
  [baseline] epoch 1/10 — accuracy=0.1088  loss=30.4302
  [baseline] epoch 2/10 — accuracy=0.1783  loss=35.6142
  [baseline] epoch 3/10 — accuracy=0.2931  loss=2.5134
  [baseline] epoch 4/10 — accuracy=0.4439  loss=1.8685
  [baseline] epoch 5/10 — accuracy=0.4671  loss=1.7961
  [baseline] epoch 6/10 — accuracy=0.5443  loss=1.4473
  [baseline] epoch 7/10 — accuracy=0.6643  loss=0.9327
  [baseline] epoch 8/10 — accuracy=0.5637  loss=1.1730
  [baseline] epoch 9/10 — accuracy=0.6404  loss=0.9844
  [baseline] epoch 10/10 — accuracy=0.6678  loss=0.9302
  [Phase 0] baseline selection_val accuracy=0.6678
  [XAI cache] Pre-computing saliency maps...

Precomputing XAI cache: 0it [00:00, ?it/s]
Precomputing XAI cache: 1it [00:02,  2.74s/it]
Precomputing XAI cache: 65it [00:05, 15.27it/s]
Precomputing XAI cache: 129it [00:07, 20.60it/s]
Precomputing XAI cache: 193it [00:09, 23.20it/s]
Precomputing XAI cache: 257it [00:12, 24.62it/s]
Precomputing XAI cache: 321it [00:14, 25.51it/s]
Precomputing XAI cache: 385it [00:16, 26.06it/s]
Precomputing XAI cache: 449it [00:19, 26.43it/s]
Precomputing XAI cache: 513it [00:21, 26.46it/s]
Precomputing XAI cache: 577it [00:23, 26.72it/s]
Precomputing XAI cache: 641it [00:26, 26.94it/s]
Precomputing XAI cache: 705it [00:28, 27.12it/s]
Precomputing XAI cache: 769it [00:30, 27.28it/s]
Precomputing XAI cache: 833it [00:33, 27.35it/s]
Precomputing XAI cache: 897it [00:35, 27.44it/s]
Precomputing XAI cache: 961it [00:37, 29.80it/s]
                                                
  [XAI cache] 1000 maps cached to /content/bnnr/benchmarks/runs_grand/eurosat_scratch/20260701_130904_bnnr_xai_s67/xai_cache
  [Phase 1/3 — ICD] 10 epochs...
  [ICD] epoch 1/10 — accuracy=0.5619  loss=1.5146
  [ICD] epoch 2/10 — accuracy=0.4409  loss=1.7326
  [ICD] epoch 3/10 — accuracy=0.4819  loss=1.5427
  [ICD] epoch 4/10 — accuracy=0.4077  loss=2.3113
  [ICD] epoch 5/10 — accuracy=0.5923  loss=1.1211
  [ICD] epoch 6/10 — accuracy=0.5938  loss=1.1477
  [ICD] epoch 7/10 — accuracy=0.6316  loss=1.0469
  [ICD] epoch 8/10 — accuracy=0.5296  loss=1.3809
  [ICD] epoch 9/10 — accuracy=0.6742  loss=0.9126
  [ICD] epoch 10/10 — accuracy=0.7118  loss=0.8573
  [ICD] selection_val accuracy=0.7118
  [Phase 2/3 — AICD] 10 epochs...
  [AICD] epoch 1/10 — accuracy=0.5576  loss=1.2434
  [AICD] epoch 2/10 — accuracy=0.4667  loss=1.9567
  [AICD] epoch 3/10 — accuracy=0.5144  loss=1.3389
  [AICD] epoch 4/10 — accuracy=0.4642  loss=1.9913
  [AICD] epoch 5/10 — accuracy=0.5286  loss=1.5109
  [AICD] epoch 6/10 — accuracy=0.4653  loss=1.7970
  [AICD] epoch 7/10 — accuracy=0.6446  loss=1.1291
  [AICD] epoch 8/10 — accuracy=0.6601  loss=0.9431
  [AICD] epoch 9/10 — accuracy=0.6582  loss=1.0295
  [AICD] epoch 10/10 — accuracy=0.7283  loss=0.8155
  [AICD] selection_val accuracy=0.7283
  [Phase 3/3 — ChurchNoise] 10 epochs...
  [ChurchNoise] epoch 1/10 — accuracy=0.5496  loss=1.2906
  [ChurchNoise] epoch 2/10 — accuracy=0.3558  loss=2.9300
  [ChurchNoise] epoch 3/10 — accuracy=0.5703  loss=1.3363
  [ChurchNoise] epoch 4/10 — accuracy=0.3735  loss=2.8747
  [ChurchNoise] epoch 5/10 — accuracy=0.4529  loss=2.7023
  [ChurchNoise] epoch 6/10 — accuracy=0.5767  loss=1.1399
  [ChurchNoise] epoch 7/10 — accuracy=0.6985  loss=0.8242
  [ChurchNoise] epoch 8/10 — accuracy=0.7034  loss=0.8382
  [ChurchNoise] epoch 9/10 — accuracy=0.6760  loss=0.8917
  [ChurchNoise] epoch 10/10 — accuracy=0.7325  loss=0.7518
  [ChurchNoise] selection_val accuracy=0.7325
  [XAI selection] best=ChurchNoise  score=0.7325
  [held_out_test] accuracy=0.7309  elapsed=385.4s
    bnnr_xai seed=67: val_metric(held_out)=0.7309

Done. 2 valid run records in /content/bnnr/benchmarks/results_eurosat_scratch.json
Summarize: python benchmarks/summarize_grand.py --results-dir benchmarks/ --datasets eurosat
[?2004h[01;34m/content/bnnr[00m# eexxiitt

[?2004l
exit

Script done on 2026-07-01 13:15:52+00:00 [COMMAND_EXIT_CODE="0"]
