Starting more_baselines/base_noise_tf.py...
Files already downloaded and verified
Files already downloaded and verified

=== Pretraining External Model (BigTransformer) on 10k Random (Corrupted) Samples ===
/home/ubuntu/refine/more_baselines/base_noise_tf.py:87: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  external_model = torch.load(model_save_path).to(device)
Loaded external model from: ./model_test10/base_noise_tf_0.8.pt
External Model Evaluation: Acc=13.80% | AUC=0.6304 | F1=0.0640 | MinCAcc=0.00%

=== flip_ratio=0.8 | Run 1/5, seed=42 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (Transformer head)...
[LoRA-TF] Ep1/30 loss=2.2823
[LoRA-TF] Ep2/30 loss=2.2185
[LoRA-TF] Ep3/30 loss=2.1782
[LoRA-TF] Ep4/30 loss=2.1655
[LoRA-TF] Ep5/30 loss=2.1614
[LoRA-TF] Ep6/30 loss=2.1495
[LoRA-TF] Ep7/30 loss=2.1498
[LoRA-TF] Ep8/30 loss=2.1474
[LoRA-TF] Ep9/30 loss=2.1464
[LoRA-TF] Ep10/30 loss=2.1438
[LoRA-TF] Ep11/30 loss=2.1409
[LoRA-TF] Ep12/30 loss=2.1462
[LoRA-TF] Ep13/30 loss=2.1494
[LoRA-TF] Ep14/30 loss=2.1394
[LoRA-TF] Ep15/30 loss=2.1422
[LoRA-TF] Ep16/30 loss=2.1476
[LoRA-TF] Ep17/30 loss=2.1405
[LoRA-TF] Ep18/30 loss=2.1387
[LoRA-TF] Ep19/30 loss=2.1407
[LoRA-TF] Ep20/30 loss=2.1452
[LoRA-TF] Ep21/30 loss=2.1419
[LoRA-TF] Ep22/30 loss=2.1424
[LoRA-TF] Ep23/30 loss=2.1391
[LoRA-TF] Ep24/30 loss=2.1378
[LoRA-TF] Ep25/30 loss=2.1356
[LoRA-TF] Ep26/30 loss=2.1391
[LoRA-TF] Ep27/30 loss=2.1451
[LoRA-TF] Ep28/30 loss=2.1426
[LoRA-TF] Ep29/30 loss=2.1346
[LoRA-TF] Ep30/30 loss=2.1374
Training DANN-Gate (Transformer head)...
[DANN-Gate-TF] Ep1/30 λ_grl=0.000 loss=6.7762
[DANN-Gate-TF] Ep2/30 λ_grl=0.171 loss=6.5700
[DANN-Gate-TF] Ep3/30 λ_grl=0.332 loss=6.4284
[DANN-Gate-TF] Ep4/30 λ_grl=0.476 loss=6.4630
[DANN-Gate-TF] Ep5/30 λ_grl=0.598 loss=6.4392
[DANN-Gate-TF] Ep6/30 λ_grl=0.697 loss=6.4210
[DANN-Gate-TF] Ep7/30 λ_grl=0.776 loss=6.4119
[DANN-Gate-TF] Ep8/30 λ_grl=0.836 loss=6.3953
[DANN-Gate-TF] Ep9/30 λ_grl=0.881 loss=6.4069
[DANN-Gate-TF] Ep10/30 λ_grl=0.914 loss=6.4223
[DANN-Gate-TF] Ep11/30 λ_grl=0.938 loss=6.4118
[DANN-Gate-TF] Ep12/30 λ_grl=0.956 loss=6.3967
[DANN-Gate-TF] Ep13/30 λ_grl=0.969 loss=6.4030
[DANN-Gate-TF] Ep14/30 λ_grl=0.978 loss=6.3855
[DANN-Gate-TF] Ep15/30 λ_grl=0.984 loss=6.3773
[DANN-Gate-TF] Ep16/30 λ_grl=0.989 loss=6.3744
[DANN-Gate-TF] Ep17/30 λ_grl=0.992 loss=6.4151
[DANN-Gate-TF] Ep18/30 λ_grl=0.994 loss=6.3922
[DANN-Gate-TF] Ep19/30 λ_grl=0.996 loss=6.4012
[DANN-Gate-TF] Ep20/30 λ_grl=0.997 loss=6.3850
[DANN-Gate-TF] Ep21/30 λ_grl=0.998 loss=6.3592
[DANN-Gate-TF] Ep22/30 λ_grl=0.999 loss=6.3828
[DANN-Gate-TF] Ep23/30 λ_grl=0.999 loss=6.3953
[DANN-Gate-TF] Ep24/30 λ_grl=0.999 loss=6.3737
[DANN-Gate-TF] Ep25/30 λ_grl=0.999 loss=6.3725
[DANN-Gate-TF] Ep26/30 λ_grl=1.000 loss=6.3616
[DANN-Gate-TF] Ep27/30 λ_grl=1.000 loss=6.3634
[DANN-Gate-TF] Ep28/30 λ_grl=1.000 loss=6.3653
[DANN-Gate-TF] Ep29/30 λ_grl=1.000 loss=6.3597
[DANN-Gate-TF] Ep30/30 λ_grl=1.000 loss=6.3484
[Run 1] LoRA Acc=21.18% | DANN-Gate Acc=21.45%

=== flip_ratio=0.8 | Run 2/5, seed=43 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (Transformer head)...
[LoRA-TF] Ep1/30 loss=2.2757
[LoRA-TF] Ep2/30 loss=2.2120
[LoRA-TF] Ep3/30 loss=2.1731
[LoRA-TF] Ep4/30 loss=2.1596
[LoRA-TF] Ep5/30 loss=2.1483
[LoRA-TF] Ep6/30 loss=2.1501
[LoRA-TF] Ep7/30 loss=2.1428
[LoRA-TF] Ep8/30 loss=2.1409
[LoRA-TF] Ep9/30 loss=2.1380
[LoRA-TF] Ep10/30 loss=2.1415
[LoRA-TF] Ep11/30 loss=2.1415
[LoRA-TF] Ep12/30 loss=2.1337
[LoRA-TF] Ep13/30 loss=2.1331
[LoRA-TF] Ep14/30 loss=2.1303
[LoRA-TF] Ep15/30 loss=2.1357
[LoRA-TF] Ep16/30 loss=2.1370
[LoRA-TF] Ep17/30 loss=2.1328
[LoRA-TF] Ep18/30 loss=2.1301
[LoRA-TF] Ep19/30 loss=2.1398
[LoRA-TF] Ep20/30 loss=2.1299
[LoRA-TF] Ep21/30 loss=2.1309
[LoRA-TF] Ep22/30 loss=2.1390
[LoRA-TF] Ep23/30 loss=2.1305
[LoRA-TF] Ep24/30 loss=2.1284
[LoRA-TF] Ep25/30 loss=2.1332
[LoRA-TF] Ep26/30 loss=2.1362
[LoRA-TF] Ep27/30 loss=2.1319
[LoRA-TF] Ep28/30 loss=2.1287
[LoRA-TF] Ep29/30 loss=2.1279
[LoRA-TF] Ep30/30 loss=2.1287
Training DANN-Gate (Transformer head)...
[DANN-Gate-TF] Ep1/30 λ_grl=0.000 loss=6.8628
[DANN-Gate-TF] Ep2/30 λ_grl=0.171 loss=6.5366
[DANN-Gate-TF] Ep3/30 λ_grl=0.332 loss=6.4574
[DANN-Gate-TF] Ep4/30 λ_grl=0.476 loss=6.3902
[DANN-Gate-TF] Ep5/30 λ_grl=0.598 loss=6.4124
[DANN-Gate-TF] Ep6/30 λ_grl=0.697 loss=6.4376
[DANN-Gate-TF] Ep7/30 λ_grl=0.776 loss=6.4153
[DANN-Gate-TF] Ep8/30 λ_grl=0.836 loss=6.3640
[DANN-Gate-TF] Ep9/30 λ_grl=0.881 loss=6.3555
[DANN-Gate-TF] Ep10/30 λ_grl=0.914 loss=6.4040
[DANN-Gate-TF] Ep11/30 λ_grl=0.938 loss=6.4022
[DANN-Gate-TF] Ep12/30 λ_grl=0.956 loss=6.3659
[DANN-Gate-TF] Ep13/30 λ_grl=0.969 loss=6.3501
[DANN-Gate-TF] Ep14/30 λ_grl=0.978 loss=6.3749
[DANN-Gate-TF] Ep15/30 λ_grl=0.984 loss=6.4094
[DANN-Gate-TF] Ep16/30 λ_grl=0.989 loss=6.4011
[DANN-Gate-TF] Ep17/30 λ_grl=0.992 loss=6.3720
[DANN-Gate-TF] Ep18/30 λ_grl=0.994 loss=6.3548
[DANN-Gate-TF] Ep19/30 λ_grl=0.996 loss=6.3509
[DANN-Gate-TF] Ep20/30 λ_grl=0.997 loss=6.3702
[DANN-Gate-TF] Ep21/30 λ_grl=0.998 loss=6.3356
[DANN-Gate-TF] Ep22/30 λ_grl=0.999 loss=6.3412
[DANN-Gate-TF] Ep23/30 λ_grl=0.999 loss=6.3595
[DANN-Gate-TF] Ep24/30 λ_grl=0.999 loss=6.3485
[DANN-Gate-TF] Ep25/30 λ_grl=0.999 loss=6.3534
[DANN-Gate-TF] Ep26/30 λ_grl=1.000 loss=6.3526
[DANN-Gate-TF] Ep27/30 λ_grl=1.000 loss=6.3555
[DANN-Gate-TF] Ep28/30 λ_grl=1.000 loss=6.3918
[DANN-Gate-TF] Ep29/30 λ_grl=1.000 loss=6.3392
[DANN-Gate-TF] Ep30/30 λ_grl=1.000 loss=6.3353
[Run 2] LoRA Acc=21.06% | DANN-Gate Acc=21.04%

=== flip_ratio=0.8 | Run 3/5, seed=44 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (Transformer head)...
[LoRA-TF] Ep1/30 loss=2.2794
[LoRA-TF] Ep2/30 loss=2.2231
[LoRA-TF] Ep3/30 loss=2.1948
[LoRA-TF] Ep4/30 loss=2.1818
[LoRA-TF] Ep5/30 loss=2.1812
[LoRA-TF] Ep6/30 loss=2.1710
[LoRA-TF] Ep7/30 loss=2.1681
[LoRA-TF] Ep8/30 loss=2.1634
[LoRA-TF] Ep9/30 loss=2.1671
[LoRA-TF] Ep10/30 loss=2.1678
[LoRA-TF] Ep11/30 loss=2.1638
[LoRA-TF] Ep12/30 loss=2.1619
[LoRA-TF] Ep13/30 loss=2.1662
[LoRA-TF] Ep14/30 loss=2.1586
[LoRA-TF] Ep15/30 loss=2.1544
[LoRA-TF] Ep16/30 loss=2.1525
[LoRA-TF] Ep17/30 loss=2.1659
[LoRA-TF] Ep18/30 loss=2.1563
[LoRA-TF] Ep19/30 loss=2.1573
[LoRA-TF] Ep20/30 loss=2.1524
[LoRA-TF] Ep21/30 loss=2.1507
[LoRA-TF] Ep22/30 loss=2.1488
[LoRA-TF] Ep23/30 loss=2.1528
[LoRA-TF] Ep24/30 loss=2.1550
[LoRA-TF] Ep25/30 loss=2.1528
[LoRA-TF] Ep26/30 loss=2.1567
[LoRA-TF] Ep27/30 loss=2.1498
[LoRA-TF] Ep28/30 loss=2.1481
[LoRA-TF] Ep29/30 loss=2.1537
[LoRA-TF] Ep30/30 loss=2.1534
Training DANN-Gate (Transformer head)...
[DANN-Gate-TF] Ep1/30 λ_grl=0.000 loss=6.5731
[DANN-Gate-TF] Ep2/30 λ_grl=0.171 loss=6.5308
[DANN-Gate-TF] Ep3/30 λ_grl=0.332 loss=6.4339
[DANN-Gate-TF] Ep4/30 λ_grl=0.476 loss=6.4608
[DANN-Gate-TF] Ep5/30 λ_grl=0.598 loss=6.4339
[DANN-Gate-TF] Ep6/30 λ_grl=0.697 loss=6.4878
[DANN-Gate-TF] Ep7/30 λ_grl=0.776 loss=6.4684
[DANN-Gate-TF] Ep8/30 λ_grl=0.836 loss=6.4824
[DANN-Gate-TF] Ep9/30 λ_grl=0.881 loss=6.4315
[DANN-Gate-TF] Ep10/30 λ_grl=0.914 loss=6.4479
[DANN-Gate-TF] Ep11/30 λ_grl=0.938 loss=6.4168
[DANN-Gate-TF] Ep12/30 λ_grl=0.956 loss=6.4248
[DANN-Gate-TF] Ep13/30 λ_grl=0.969 loss=6.4502
[DANN-Gate-TF] Ep14/30 λ_grl=0.978 loss=6.4017
[DANN-Gate-TF] Ep15/30 λ_grl=0.984 loss=6.4090
[DANN-Gate-TF] Ep16/30 λ_grl=0.989 loss=6.4162
[DANN-Gate-TF] Ep17/30 λ_grl=0.992 loss=6.3778
[DANN-Gate-TF] Ep18/30 λ_grl=0.994 loss=6.4098
[DANN-Gate-TF] Ep19/30 λ_grl=0.996 loss=6.3892
[DANN-Gate-TF] Ep20/30 λ_grl=0.997 loss=6.3957
[DANN-Gate-TF] Ep21/30 λ_grl=0.998 loss=6.3916
[DANN-Gate-TF] Ep22/30 λ_grl=0.999 loss=6.4330
[DANN-Gate-TF] Ep23/30 λ_grl=0.999 loss=6.4096
[DANN-Gate-TF] Ep24/30 λ_grl=0.999 loss=6.4141
[DANN-Gate-TF] Ep25/30 λ_grl=0.999 loss=6.3866
[DANN-Gate-TF] Ep26/30 λ_grl=1.000 loss=6.3920
[DANN-Gate-TF] Ep27/30 λ_grl=1.000 loss=6.4041
[DANN-Gate-TF] Ep28/30 λ_grl=1.000 loss=6.3854
[DANN-Gate-TF] Ep29/30 λ_grl=1.000 loss=6.3920
[DANN-Gate-TF] Ep30/30 λ_grl=1.000 loss=6.4133
[Run 3] LoRA Acc=21.97% | DANN-Gate Acc=21.59%

=== flip_ratio=0.8 | Run 4/5, seed=45 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (Transformer head)...
[LoRA-TF] Ep1/30 loss=2.2764
[LoRA-TF] Ep2/30 loss=2.2037
[LoRA-TF] Ep3/30 loss=2.1647
[LoRA-TF] Ep4/30 loss=2.1482
[LoRA-TF] Ep5/30 loss=2.1487
[LoRA-TF] Ep6/30 loss=2.1417
[LoRA-TF] Ep7/30 loss=2.1458
[LoRA-TF] Ep8/30 loss=2.1360
[LoRA-TF] Ep9/30 loss=2.1500
[LoRA-TF] Ep10/30 loss=2.1372
[LoRA-TF] Ep11/30 loss=2.1424
[LoRA-TF] Ep12/30 loss=2.1367
[LoRA-TF] Ep13/30 loss=2.1410
[LoRA-TF] Ep14/30 loss=2.1388
[LoRA-TF] Ep15/30 loss=2.1390
[LoRA-TF] Ep16/30 loss=2.1283
[LoRA-TF] Ep17/30 loss=2.1311
[LoRA-TF] Ep18/30 loss=2.1283
[LoRA-TF] Ep19/30 loss=2.1333
[LoRA-TF] Ep20/30 loss=2.1352
[LoRA-TF] Ep21/30 loss=2.1309
[LoRA-TF] Ep22/30 loss=2.1342
[LoRA-TF] Ep23/30 loss=2.1288
[LoRA-TF] Ep24/30 loss=2.1329
[LoRA-TF] Ep25/30 loss=2.1208
[LoRA-TF] Ep26/30 loss=2.1290
[LoRA-TF] Ep27/30 loss=2.1242
[LoRA-TF] Ep28/30 loss=2.1240
[LoRA-TF] Ep29/30 loss=2.1213
[LoRA-TF] Ep30/30 loss=2.1190
Training DANN-Gate (Transformer head)...
[DANN-Gate-TF] Ep1/30 λ_grl=0.000 loss=6.5975
[DANN-Gate-TF] Ep2/30 λ_grl=0.171 loss=6.5434
[DANN-Gate-TF] Ep3/30 λ_grl=0.332 loss=6.4255
[DANN-Gate-TF] Ep4/30 λ_grl=0.476 loss=6.4091
[DANN-Gate-TF] Ep5/30 λ_grl=0.598 loss=6.3804
[DANN-Gate-TF] Ep6/30 λ_grl=0.697 loss=6.3903
[DANN-Gate-TF] Ep7/30 λ_grl=0.776 loss=6.4130
[DANN-Gate-TF] Ep8/30 λ_grl=0.836 loss=6.3953
[DANN-Gate-TF] Ep9/30 λ_grl=0.881 loss=6.3752
[DANN-Gate-TF] Ep10/30 λ_grl=0.914 loss=6.3526
[DANN-Gate-TF] Ep11/30 λ_grl=0.938 loss=6.3395
[DANN-Gate-TF] Ep12/30 λ_grl=0.956 loss=6.3817
[DANN-Gate-TF] Ep13/30 λ_grl=0.969 loss=6.3736
[DANN-Gate-TF] Ep14/30 λ_grl=0.978 loss=6.3624
[DANN-Gate-TF] Ep15/30 λ_grl=0.984 loss=6.3844
[DANN-Gate-TF] Ep16/30 λ_grl=0.989 loss=6.3509
[DANN-Gate-TF] Ep17/30 λ_grl=0.992 loss=6.3300
[DANN-Gate-TF] Ep18/30 λ_grl=0.994 loss=6.3456
[DANN-Gate-TF] Ep19/30 λ_grl=0.996 loss=6.3852
[DANN-Gate-TF] Ep20/30 λ_grl=0.997 loss=6.3743
[DANN-Gate-TF] Ep21/30 λ_grl=0.998 loss=6.3057
[DANN-Gate-TF] Ep22/30 λ_grl=0.999 loss=6.3380
[DANN-Gate-TF] Ep23/30 λ_grl=0.999 loss=6.3289
[DANN-Gate-TF] Ep24/30 λ_grl=0.999 loss=6.3500
[DANN-Gate-TF] Ep25/30 λ_grl=0.999 loss=6.3499
[DANN-Gate-TF] Ep26/30 λ_grl=1.000 loss=6.3122
[DANN-Gate-TF] Ep27/30 λ_grl=1.000 loss=6.3157
[DANN-Gate-TF] Ep28/30 λ_grl=1.000 loss=6.3039
[DANN-Gate-TF] Ep29/30 λ_grl=1.000 loss=6.3168
[DANN-Gate-TF] Ep30/30 λ_grl=1.000 loss=6.3148
[Run 4] LoRA Acc=21.97% | DANN-Gate Acc=21.08%

=== flip_ratio=0.8 | Run 5/5, seed=46 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (Transformer head)...
[LoRA-TF] Ep1/30 loss=2.2786
[LoRA-TF] Ep2/30 loss=2.2201
[LoRA-TF] Ep3/30 loss=2.1874
[LoRA-TF] Ep4/30 loss=2.1729
[LoRA-TF] Ep5/30 loss=2.1703
[LoRA-TF] Ep6/30 loss=2.1597
[LoRA-TF] Ep7/30 loss=2.1676
[LoRA-TF] Ep8/30 loss=2.1611
[LoRA-TF] Ep9/30 loss=2.1541
[LoRA-TF] Ep10/30 loss=2.1558
[LoRA-TF] Ep11/30 loss=2.1535
[LoRA-TF] Ep12/30 loss=2.1562
[LoRA-TF] Ep13/30 loss=2.1553
[LoRA-TF] Ep14/30 loss=2.1453
[LoRA-TF] Ep15/30 loss=2.1496
[LoRA-TF] Ep16/30 loss=2.1451
[LoRA-TF] Ep17/30 loss=2.1495
[LoRA-TF] Ep18/30 loss=2.1425
[LoRA-TF] Ep19/30 loss=2.1439
[LoRA-TF] Ep20/30 loss=2.1405
[LoRA-TF] Ep21/30 loss=2.1425
[LoRA-TF] Ep22/30 loss=2.1447
[LoRA-TF] Ep23/30 loss=2.1465
[LoRA-TF] Ep24/30 loss=2.1453
[LoRA-TF] Ep25/30 loss=2.1397
[LoRA-TF] Ep26/30 loss=2.1355
[LoRA-TF] Ep27/30 loss=2.1368
[LoRA-TF] Ep28/30 loss=2.1423
[LoRA-TF] Ep29/30 loss=2.1392
[LoRA-TF] Ep30/30 loss=2.1413
Training DANN-Gate (Transformer head)...
[DANN-Gate-TF] Ep1/30 λ_grl=0.000 loss=6.7777
[DANN-Gate-TF] Ep2/30 λ_grl=0.171 loss=6.5375
[DANN-Gate-TF] Ep3/30 λ_grl=0.332 loss=6.4648
[DANN-Gate-TF] Ep4/30 λ_grl=0.476 loss=6.4482
[DANN-Gate-TF] Ep5/30 λ_grl=0.598 loss=6.4252
[DANN-Gate-TF] Ep6/30 λ_grl=0.697 loss=6.3889
[DANN-Gate-TF] Ep7/30 λ_grl=0.776 loss=6.4034
[DANN-Gate-TF] Ep8/30 λ_grl=0.836 loss=6.3919
[DANN-Gate-TF] Ep9/30 λ_grl=0.881 loss=6.3952
[DANN-Gate-TF] Ep10/30 λ_grl=0.914 loss=6.4144
[DANN-Gate-TF] Ep11/30 λ_grl=0.938 loss=6.4007
[DANN-Gate-TF] Ep12/30 λ_grl=0.956 loss=6.3879
[DANN-Gate-TF] Ep13/30 λ_grl=0.969 loss=6.3962
[DANN-Gate-TF] Ep14/30 λ_grl=0.978 loss=6.3926
[DANN-Gate-TF] Ep15/30 λ_grl=0.984 loss=6.3954
[DANN-Gate-TF] Ep16/30 λ_grl=0.989 loss=6.3976
[DANN-Gate-TF] Ep17/30 λ_grl=0.992 loss=6.3911
[DANN-Gate-TF] Ep18/30 λ_grl=0.994 loss=6.3704
[DANN-Gate-TF] Ep19/30 λ_grl=0.996 loss=6.3878
[DANN-Gate-TF] Ep20/30 λ_grl=0.997 loss=6.4138
[DANN-Gate-TF] Ep21/30 λ_grl=0.998 loss=6.3859
[DANN-Gate-TF] Ep22/30 λ_grl=0.999 loss=6.3858
[DANN-Gate-TF] Ep23/30 λ_grl=0.999 loss=6.3932
[DANN-Gate-TF] Ep24/30 λ_grl=0.999 loss=6.3966
[DANN-Gate-TF] Ep25/30 λ_grl=0.999 loss=6.3847
[DANN-Gate-TF] Ep26/30 λ_grl=1.000 loss=6.3799
[DANN-Gate-TF] Ep27/30 λ_grl=1.000 loss=6.3641
[DANN-Gate-TF] Ep28/30 λ_grl=1.000 loss=6.3630
[DANN-Gate-TF] Ep29/30 λ_grl=1.000 loss=6.3577
[DANN-Gate-TF] Ep30/30 λ_grl=1.000 loss=6.3624
[Run 5] LoRA Acc=22.29% | DANN-Gate Acc=21.70%

All done. Final mean/std results saved to: ./results_test10_base/noise_tf_0.8.json
Files already downloaded and verified
Files already downloaded and verified

=== Pretraining External Model (BigTransformer) on 10k Random (Corrupted) Samples ===
[Epoch 1] Loss: 2.0995
[Epoch 2] Loss: 1.9893
[Epoch 3] Loss: 1.8960
[Epoch 4] Loss: 1.8194
[Epoch 5] Loss: 1.7581
[Epoch 6] Loss: 1.7031
[Epoch 7] Loss: 1.6364
[Epoch 8] Loss: 1.5815
[Epoch 9] Loss: 1.5404
[Epoch 10] Loss: 1.5022
[Epoch 11] Loss: 1.4451
[Epoch 12] Loss: 1.3877
[Epoch 13] Loss: 1.3509
[Epoch 14] Loss: 1.3075
[Epoch 15] Loss: 1.2777
[Epoch 16] Loss: 1.2235
[Epoch 17] Loss: 1.1953
[Epoch 18] Loss: 1.1555
[Epoch 19] Loss: 1.1201
[Epoch 20] Loss: 1.0608
[Epoch 21] Loss: 1.0749
[Epoch 22] Loss: 1.0174
[Epoch 23] Loss: 0.9758
[Epoch 24] Loss: 0.9592
[Epoch 25] Loss: 0.9226
[Epoch 26] Loss: 0.8648
[Epoch 27] Loss: 0.8300
[Epoch 28] Loss: 0.8071
[Epoch 29] Loss: 0.7801
[Epoch 30] Loss: 0.7564
[Epoch 31] Loss: 0.7209
[Epoch 32] Loss: 0.6963
[Epoch 33] Loss: 0.6589
[Epoch 34] Loss: 0.6353
[Epoch 35] Loss: 0.6107
[Epoch 36] Loss: 0.5700
[Epoch 37] Loss: 0.5462
[Epoch 38] Loss: 0.5275
[Epoch 39] Loss: 0.5284
[Epoch 40] Loss: 0.5063
[Epoch 41] Loss: 0.4747
[Epoch 42] Loss: 0.4211
[Epoch 43] Loss: 0.4233
[Epoch 44] Loss: 0.3855
[Epoch 45] Loss: 0.3768
[Epoch 46] Loss: 0.3909
[Epoch 47] Loss: 0.3725
[Epoch 48] Loss: 0.3169
[Epoch 49] Loss: 0.3494
[Epoch 50] Loss: 0.3135
[Epoch 51] Loss: 0.3153
[Epoch 52] Loss: 0.3411
[Epoch 53] Loss: 0.2733
[Epoch 54] Loss: 0.2897
[Epoch 55] Loss: 0.2746
[Epoch 56] Loss: 0.2427
[Epoch 57] Loss: 0.2554
[Epoch 58] Loss: 0.2472
[Epoch 59] Loss: 0.2332
[Epoch 60] Loss: 0.2325
Trained and saved external model to: ./model_test10/base_noise_tf_0.0.pt
External Model Evaluation: Acc=57.27% | AUC=0.9082 | F1=0.5711 | MinCAcc=37.70%

=== flip_ratio=0.0 | Run 1/5, seed=42 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (Transformer head)...
[LoRA-TF] Ep1/30 loss=1.9228
[LoRA-TF] Ep2/30 loss=1.6667
[LoRA-TF] Ep3/30 loss=1.5209
[LoRA-TF] Ep4/30 loss=1.3839
[LoRA-TF] Ep5/30 loss=1.3076
[LoRA-TF] Ep6/30 loss=1.2736
[LoRA-TF] Ep7/30 loss=1.2066
[LoRA-TF] Ep8/30 loss=1.1727
[LoRA-TF] Ep9/30 loss=1.1431
[LoRA-TF] Ep10/30 loss=1.1114
[LoRA-TF] Ep11/30 loss=1.0938
[LoRA-TF] Ep12/30 loss=1.0587
[LoRA-TF] Ep13/30 loss=1.0528
[LoRA-TF] Ep14/30 loss=1.0232
[LoRA-TF] Ep15/30 loss=1.0060
[LoRA-TF] Ep16/30 loss=0.9906
[LoRA-TF] Ep17/30 loss=0.9707
[LoRA-TF] Ep18/30 loss=0.9550
[LoRA-TF] Ep19/30 loss=0.9404
[LoRA-TF] Ep20/30 loss=0.9235
[LoRA-TF] Ep21/30 loss=0.9056
[LoRA-TF] Ep22/30 loss=0.9257
[LoRA-TF] Ep23/30 loss=0.9235
[LoRA-TF] Ep24/30 loss=0.8733
[LoRA-TF] Ep25/30 loss=0.8841
[LoRA-TF] Ep26/30 loss=0.8670
[LoRA-TF] Ep27/30 loss=0.8603
[LoRA-TF] Ep28/30 loss=0.8529
[LoRA-TF] Ep29/30 loss=0.8513
[LoRA-TF] Ep30/30 loss=0.8341
Training DANN-Gate (Transformer head)...
[DANN-Gate-TF] Ep1/30 λ_grl=0.000 loss=5.8833
[DANN-Gate-TF] Ep2/30 λ_grl=0.171 loss=5.4702
[DANN-Gate-TF] Ep3/30 λ_grl=0.332 loss=5.2001
[DANN-Gate-TF] Ep4/30 λ_grl=0.476 loss=4.8860
[DANN-Gate-TF] Ep5/30 λ_grl=0.598 loss=4.7138
[DANN-Gate-TF] Ep6/30 λ_grl=0.697 loss=4.6579
[DANN-Gate-TF] Ep7/30 λ_grl=0.776 loss=4.5681
[DANN-Gate-TF] Ep8/30 λ_grl=0.836 loss=4.4492
[DANN-Gate-TF] Ep9/30 λ_grl=0.881 loss=4.3787
[DANN-Gate-TF] Ep10/30 λ_grl=0.914 loss=4.3630
[DANN-Gate-TF] Ep11/30 λ_grl=0.938 loss=4.2611
[DANN-Gate-TF] Ep12/30 λ_grl=0.956 loss=4.2074
[DANN-Gate-TF] Ep13/30 λ_grl=0.969 loss=4.1135
[DANN-Gate-TF] Ep14/30 λ_grl=0.978 loss=4.1209
[DANN-Gate-TF] Ep15/30 λ_grl=0.984 loss=4.0999
[DANN-Gate-TF] Ep16/30 λ_grl=0.989 loss=4.0410
[DANN-Gate-TF] Ep17/30 λ_grl=0.992 loss=4.0228
[DANN-Gate-TF] Ep18/30 λ_grl=0.994 loss=4.0191
[DANN-Gate-TF] Ep19/30 λ_grl=0.996 loss=3.9489
[DANN-Gate-TF] Ep20/30 λ_grl=0.997 loss=3.9137
[DANN-Gate-TF] Ep21/30 λ_grl=0.998 loss=3.9231
[DANN-Gate-TF] Ep22/30 λ_grl=0.999 loss=3.8548
[DANN-Gate-TF] Ep23/30 λ_grl=0.999 loss=3.8391
[DANN-Gate-TF] Ep24/30 λ_grl=0.999 loss=3.8236
[DANN-Gate-TF] Ep25/30 λ_grl=0.999 loss=3.7937
[DANN-Gate-TF] Ep26/30 λ_grl=1.000 loss=3.8478
[DANN-Gate-TF] Ep27/30 λ_grl=1.000 loss=3.7826
[DANN-Gate-TF] Ep28/30 λ_grl=1.000 loss=3.7532
[DANN-Gate-TF] Ep29/30 λ_grl=1.000 loss=3.7670
[DANN-Gate-TF] Ep30/30 λ_grl=1.000 loss=3.7334
[Run 1] LoRA Acc=58.18% | DANN-Gate Acc=58.79%

=== flip_ratio=0.0 | Run 2/5, seed=43 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (Transformer head)...
[LoRA-TF] Ep1/30 loss=1.5732
[LoRA-TF] Ep2/30 loss=1.4205
[LoRA-TF] Ep3/30 loss=1.3193
[LoRA-TF] Ep4/30 loss=1.2472
[LoRA-TF] Ep5/30 loss=1.1766
[LoRA-TF] Ep6/30 loss=1.1179
[LoRA-TF] Ep7/30 loss=1.0876
[LoRA-TF] Ep8/30 loss=1.0552
[LoRA-TF] Ep9/30 loss=0.9869
[LoRA-TF] Ep10/30 loss=0.9689
[LoRA-TF] Ep11/30 loss=0.9547
[LoRA-TF] Ep12/30 loss=0.9215
[LoRA-TF] Ep13/30 loss=0.8959
[LoRA-TF] Ep14/30 loss=0.8865
[LoRA-TF] Ep15/30 loss=0.8754
[LoRA-TF] Ep16/30 loss=0.8326
[LoRA-TF] Ep17/30 loss=0.8194
[LoRA-TF] Ep18/30 loss=0.8294
[LoRA-TF] Ep19/30 loss=0.8014
[LoRA-TF] Ep20/30 loss=0.7969
[LoRA-TF] Ep21/30 loss=0.7747
[LoRA-TF] Ep22/30 loss=0.7571
[LoRA-TF] Ep23/30 loss=0.7836
[LoRA-TF] Ep24/30 loss=0.7520
[LoRA-TF] Ep25/30 loss=0.7498
[LoRA-TF] Ep26/30 loss=0.7370
[LoRA-TF] Ep27/30 loss=0.7317
[LoRA-TF] Ep28/30 loss=0.7156
[LoRA-TF] Ep29/30 loss=0.7309
[LoRA-TF] Ep30/30 loss=0.7093
Training DANN-Gate (Transformer head)...
[DANN-Gate-TF] Ep1/30 λ_grl=0.000 loss=6.1833
[DANN-Gate-TF] Ep2/30 λ_grl=0.171 loss=5.0211
[DANN-Gate-TF] Ep3/30 λ_grl=0.332 loss=4.8410
[DANN-Gate-TF] Ep4/30 λ_grl=0.476 loss=4.6993
[DANN-Gate-TF] Ep5/30 λ_grl=0.598 loss=4.5215
[DANN-Gate-TF] Ep6/30 λ_grl=0.697 loss=4.3823
[DANN-Gate-TF] Ep7/30 λ_grl=0.776 loss=4.2543
[DANN-Gate-TF] Ep8/30 λ_grl=0.836 loss=4.1997
[DANN-Gate-TF] Ep9/30 λ_grl=0.881 loss=4.1708
[DANN-Gate-TF] Ep10/30 λ_grl=0.914 loss=4.0615
[DANN-Gate-TF] Ep11/30 λ_grl=0.938 loss=3.9897
[DANN-Gate-TF] Ep12/30 λ_grl=0.956 loss=3.9477
[DANN-Gate-TF] Ep13/30 λ_grl=0.969 loss=3.9006
[DANN-Gate-TF] Ep14/30 λ_grl=0.978 loss=3.8991
[DANN-Gate-TF] Ep15/30 λ_grl=0.984 loss=3.8340
[DANN-Gate-TF] Ep16/30 λ_grl=0.989 loss=3.8138
[DANN-Gate-TF] Ep17/30 λ_grl=0.992 loss=3.7706
[DANN-Gate-TF] Ep18/30 λ_grl=0.994 loss=3.7710
[DANN-Gate-TF] Ep19/30 λ_grl=0.996 loss=3.7138
[DANN-Gate-TF] Ep20/30 λ_grl=0.997 loss=3.6911
[DANN-Gate-TF] Ep21/30 λ_grl=0.998 loss=3.6418
[DANN-Gate-TF] Ep22/30 λ_grl=0.999 loss=3.6462
[DANN-Gate-TF] Ep23/30 λ_grl=0.999 loss=3.6517
[DANN-Gate-TF] Ep24/30 λ_grl=0.999 loss=3.6282
[DANN-Gate-TF] Ep25/30 λ_grl=0.999 loss=3.5963
[DANN-Gate-TF] Ep26/30 λ_grl=1.000 loss=3.5753
[DANN-Gate-TF] Ep27/30 λ_grl=1.000 loss=3.5759
[DANN-Gate-TF] Ep28/30 λ_grl=1.000 loss=3.5420
[DANN-Gate-TF] Ep29/30 λ_grl=1.000 loss=3.5492
[DANN-Gate-TF] Ep30/30 λ_grl=1.000 loss=3.5090
[Run 2] LoRA Acc=58.56% | DANN-Gate Acc=58.68%

=== flip_ratio=0.0 | Run 3/5, seed=44 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (Transformer head)...
[LoRA-TF] Ep1/30 loss=1.7231
[LoRA-TF] Ep2/30 loss=1.5211
[LoRA-TF] Ep3/30 loss=1.4120
[LoRA-TF] Ep4/30 loss=1.2901
[LoRA-TF] Ep5/30 loss=1.2401
[LoRA-TF] Ep6/30 loss=1.1754
[LoRA-TF] Ep7/30 loss=1.1160
[LoRA-TF] Ep8/30 loss=1.0693
[LoRA-TF] Ep9/30 loss=1.0227
[LoRA-TF] Ep10/30 loss=0.9897
[LoRA-TF] Ep11/30 loss=0.9880
[LoRA-TF] Ep12/30 loss=0.9801
[LoRA-TF] Ep13/30 loss=0.9476
[LoRA-TF] Ep14/30 loss=0.9335
[LoRA-TF] Ep15/30 loss=0.9052
[LoRA-TF] Ep16/30 loss=0.9082
[LoRA-TF] Ep17/30 loss=0.8727
[LoRA-TF] Ep18/30 loss=0.8938
[LoRA-TF] Ep19/30 loss=0.8681
[LoRA-TF] Ep20/30 loss=0.8497
[LoRA-TF] Ep21/30 loss=0.8419
[LoRA-TF] Ep22/30 loss=0.8507
[LoRA-TF] Ep23/30 loss=0.8133
[LoRA-TF] Ep24/30 loss=0.8101
[LoRA-TF] Ep25/30 loss=0.8083
[LoRA-TF] Ep26/30 loss=0.7821
[LoRA-TF] Ep27/30 loss=0.7697
[LoRA-TF] Ep28/30 loss=0.7808
[LoRA-TF] Ep29/30 loss=0.7757
[LoRA-TF] Ep30/30 loss=0.7689
Training DANN-Gate (Transformer head)...
[DANN-Gate-TF] Ep1/30 λ_grl=0.000 loss=6.3266
[DANN-Gate-TF] Ep2/30 λ_grl=0.171 loss=5.3627
[DANN-Gate-TF] Ep3/30 λ_grl=0.332 loss=5.0282
[DANN-Gate-TF] Ep4/30 λ_grl=0.476 loss=4.8133
[DANN-Gate-TF] Ep5/30 λ_grl=0.598 loss=4.6291
[DANN-Gate-TF] Ep6/30 λ_grl=0.697 loss=4.5008
[DANN-Gate-TF] Ep7/30 λ_grl=0.776 loss=4.4108
[DANN-Gate-TF] Ep8/30 λ_grl=0.836 loss=4.3171
[DANN-Gate-TF] Ep9/30 λ_grl=0.881 loss=4.2666
[DANN-Gate-TF] Ep10/30 λ_grl=0.914 loss=4.1858
[DANN-Gate-TF] Ep11/30 λ_grl=0.938 loss=4.1359
[DANN-Gate-TF] Ep12/30 λ_grl=0.956 loss=4.0939
[DANN-Gate-TF] Ep13/30 λ_grl=0.969 loss=4.0487
[DANN-Gate-TF] Ep14/30 λ_grl=0.978 loss=4.0621
[DANN-Gate-TF] Ep15/30 λ_grl=0.984 loss=3.9942
[DANN-Gate-TF] Ep16/30 λ_grl=0.989 loss=3.9511
[DANN-Gate-TF] Ep17/30 λ_grl=0.992 loss=3.8961
[DANN-Gate-TF] Ep18/30 λ_grl=0.994 loss=3.8650
[DANN-Gate-TF] Ep19/30 λ_grl=0.996 loss=3.8278
[DANN-Gate-TF] Ep20/30 λ_grl=0.997 loss=3.8012
[DANN-Gate-TF] Ep21/30 λ_grl=0.998 loss=3.8355
[DANN-Gate-TF] Ep22/30 λ_grl=0.999 loss=3.7425
[DANN-Gate-TF] Ep23/30 λ_grl=0.999 loss=3.7795
[DANN-Gate-TF] Ep24/30 λ_grl=0.999 loss=3.7144
[DANN-Gate-TF] Ep25/30 λ_grl=0.999 loss=3.7433
[DANN-Gate-TF] Ep26/30 λ_grl=1.000 loss=3.7348
[DANN-Gate-TF] Ep27/30 λ_grl=1.000 loss=3.6743
[DANN-Gate-TF] Ep28/30 λ_grl=1.000 loss=3.6758
[DANN-Gate-TF] Ep29/30 λ_grl=1.000 loss=3.6704
[DANN-Gate-TF] Ep30/30 λ_grl=1.000 loss=3.6682
[Run 3] LoRA Acc=58.81% | DANN-Gate Acc=59.02%

=== flip_ratio=0.0 | Run 4/5, seed=45 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (Transformer head)...
[LoRA-TF] Ep1/30 loss=1.6276
[LoRA-TF] Ep2/30 loss=1.4286
[LoRA-TF] Ep3/30 loss=1.3135
[LoRA-TF] Ep4/30 loss=1.2258
[LoRA-TF] Ep5/30 loss=1.1771
[LoRA-TF] Ep6/30 loss=1.1273
[LoRA-TF] Ep7/30 loss=1.0628
[LoRA-TF] Ep8/30 loss=1.0376
[LoRA-TF] Ep9/30 loss=1.0107
[LoRA-TF] Ep10/30 loss=0.9840
[LoRA-TF] Ep11/30 loss=0.9568
[LoRA-TF] Ep12/30 loss=0.9203
[LoRA-TF] Ep13/30 loss=0.9225
[LoRA-TF] Ep14/30 loss=0.9032
[LoRA-TF] Ep15/30 loss=0.8655
[LoRA-TF] Ep16/30 loss=0.8573
[LoRA-TF] Ep17/30 loss=0.8517
[LoRA-TF] Ep18/30 loss=0.8391
[LoRA-TF] Ep19/30 loss=0.8263
[LoRA-TF] Ep20/30 loss=0.8099
[LoRA-TF] Ep21/30 loss=0.8066
[LoRA-TF] Ep22/30 loss=0.7777
[LoRA-TF] Ep23/30 loss=0.7715
[LoRA-TF] Ep24/30 loss=0.7807
[LoRA-TF] Ep25/30 loss=0.7601
[LoRA-TF] Ep26/30 loss=0.7586
[LoRA-TF] Ep27/30 loss=0.7328
[LoRA-TF] Ep28/30 loss=0.7454
[LoRA-TF] Ep29/30 loss=0.7387
[LoRA-TF] Ep30/30 loss=0.7091
Training DANN-Gate (Transformer head)...
[DANN-Gate-TF] Ep1/30 λ_grl=0.000 loss=5.8786
[DANN-Gate-TF] Ep2/30 λ_grl=0.171 loss=5.2979
[DANN-Gate-TF] Ep3/30 λ_grl=0.332 loss=4.9519
[DANN-Gate-TF] Ep4/30 λ_grl=0.476 loss=4.6896
[DANN-Gate-TF] Ep5/30 λ_grl=0.598 loss=4.5222
[DANN-Gate-TF] Ep6/30 λ_grl=0.697 loss=4.4161
[DANN-Gate-TF] Ep7/30 λ_grl=0.776 loss=4.3058
[DANN-Gate-TF] Ep8/30 λ_grl=0.836 loss=4.2244
[DANN-Gate-TF] Ep9/30 λ_grl=0.881 loss=4.1530
[DANN-Gate-TF] Ep10/30 λ_grl=0.914 loss=4.1035
[DANN-Gate-TF] Ep11/30 λ_grl=0.938 loss=4.0596
[DANN-Gate-TF] Ep12/30 λ_grl=0.956 loss=3.9479
[DANN-Gate-TF] Ep13/30 λ_grl=0.969 loss=3.9600
[DANN-Gate-TF] Ep14/30 λ_grl=0.978 loss=3.9086
[DANN-Gate-TF] Ep15/30 λ_grl=0.984 loss=3.8610
[DANN-Gate-TF] Ep16/30 λ_grl=0.989 loss=3.7929
[DANN-Gate-TF] Ep17/30 λ_grl=0.992 loss=3.7879
[DANN-Gate-TF] Ep18/30 λ_grl=0.994 loss=3.7148
[DANN-Gate-TF] Ep19/30 λ_grl=0.996 loss=3.7329
[DANN-Gate-TF] Ep20/30 λ_grl=0.997 loss=3.7621
[DANN-Gate-TF] Ep21/30 λ_grl=0.998 loss=3.6970
[DANN-Gate-TF] Ep22/30 λ_grl=0.999 loss=3.6997
[DANN-Gate-TF] Ep23/30 λ_grl=0.999 loss=3.6484
[DANN-Gate-TF] Ep24/30 λ_grl=0.999 loss=3.6635
[DANN-Gate-TF] Ep25/30 λ_grl=0.999 loss=3.5718
[DANN-Gate-TF] Ep26/30 λ_grl=1.000 loss=3.6134
[DANN-Gate-TF] Ep27/30 λ_grl=1.000 loss=3.5788
[DANN-Gate-TF] Ep28/30 λ_grl=1.000 loss=3.5632
[DANN-Gate-TF] Ep29/30 λ_grl=1.000 loss=3.5709
[DANN-Gate-TF] Ep30/30 λ_grl=1.000 loss=3.5559
[Run 4] LoRA Acc=59.53% | DANN-Gate Acc=58.95%

=== flip_ratio=0.0 | Run 5/5, seed=46 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (Transformer head)...
[LoRA-TF] Ep1/30 loss=1.6415
[LoRA-TF] Ep2/30 loss=1.4141
[LoRA-TF] Ep3/30 loss=1.2739
[LoRA-TF] Ep4/30 loss=1.1877
[LoRA-TF] Ep5/30 loss=1.1233
[LoRA-TF] Ep6/30 loss=1.0783
[LoRA-TF] Ep7/30 loss=1.0435
[LoRA-TF] Ep8/30 loss=1.0333
[LoRA-TF] Ep9/30 loss=0.9919
[LoRA-TF] Ep10/30 loss=0.9638
[LoRA-TF] Ep11/30 loss=0.9445
[LoRA-TF] Ep12/30 loss=0.9289
[LoRA-TF] Ep13/30 loss=0.9160
[LoRA-TF] Ep14/30 loss=0.8712
[LoRA-TF] Ep15/30 loss=0.8645
[LoRA-TF] Ep16/30 loss=0.8478
[LoRA-TF] Ep17/30 loss=0.8447
[LoRA-TF] Ep18/30 loss=0.8261
[LoRA-TF] Ep19/30 loss=0.8048
[LoRA-TF] Ep20/30 loss=0.8185
[LoRA-TF] Ep21/30 loss=0.7845
[LoRA-TF] Ep22/30 loss=0.7804
[LoRA-TF] Ep23/30 loss=0.7782
[LoRA-TF] Ep24/30 loss=0.7677
[LoRA-TF] Ep25/30 loss=0.7617
[LoRA-TF] Ep26/30 loss=0.7489
[LoRA-TF] Ep27/30 loss=0.7557
[LoRA-TF] Ep28/30 loss=0.7333
[LoRA-TF] Ep29/30 loss=0.7331
[LoRA-TF] Ep30/30 loss=0.7259
Training DANN-Gate (Transformer head)...
[DANN-Gate-TF] Ep1/30 λ_grl=0.000 loss=5.4896
[DANN-Gate-TF] Ep2/30 λ_grl=0.171 loss=5.0883
[DANN-Gate-TF] Ep3/30 λ_grl=0.332 loss=4.8518
[DANN-Gate-TF] Ep4/30 λ_grl=0.476 loss=4.6450
[DANN-Gate-TF] Ep5/30 λ_grl=0.598 loss=4.5671
[DANN-Gate-TF] Ep6/30 λ_grl=0.697 loss=4.4106
[DANN-Gate-TF] Ep7/30 λ_grl=0.776 loss=4.3357
[DANN-Gate-TF] Ep8/30 λ_grl=0.836 loss=4.1865
[DANN-Gate-TF] Ep9/30 λ_grl=0.881 loss=4.1572
[DANN-Gate-TF] Ep10/30 λ_grl=0.914 loss=4.0717
[DANN-Gate-TF] Ep11/30 λ_grl=0.938 loss=4.0575
[DANN-Gate-TF] Ep12/30 λ_grl=0.956 loss=3.9934
[DANN-Gate-TF] Ep13/30 λ_grl=0.969 loss=3.9366
[DANN-Gate-TF] Ep14/30 λ_grl=0.978 loss=3.8863
[DANN-Gate-TF] Ep15/30 λ_grl=0.984 loss=3.8610
[DANN-Gate-TF] Ep16/30 λ_grl=0.989 loss=3.8194
[DANN-Gate-TF] Ep17/30 λ_grl=0.992 loss=3.7983
[DANN-Gate-TF] Ep18/30 λ_grl=0.994 loss=3.7512
[DANN-Gate-TF] Ep19/30 λ_grl=0.996 loss=3.7577
[DANN-Gate-TF] Ep20/30 λ_grl=0.997 loss=3.7014
[DANN-Gate-TF] Ep21/30 λ_grl=0.998 loss=3.7095
[DANN-Gate-TF] Ep22/30 λ_grl=0.999 loss=3.6433
[DANN-Gate-TF] Ep23/30 λ_grl=0.999 loss=3.6593
[DANN-Gate-TF] Ep24/30 λ_grl=0.999 loss=3.6091
[DANN-Gate-TF] Ep25/30 λ_grl=0.999 loss=3.5762
[DANN-Gate-TF] Ep26/30 λ_grl=1.000 loss=3.5967
[DANN-Gate-TF] Ep27/30 λ_grl=1.000 loss=3.5656
[DANN-Gate-TF] Ep28/30 λ_grl=1.000 loss=3.5796
[DANN-Gate-TF] Ep29/30 λ_grl=1.000 loss=3.5856
[DANN-Gate-TF] Ep30/30 λ_grl=1.000 loss=3.5434
[Run 5] LoRA Acc=59.11% | DANN-Gate Acc=59.38%

All done. Final mean/std results saved to: ./results_test10_base/noise_tf_0.0.json
more_baselines/base_noise_tf.py completed successfully.
Starting more_baselines/base_adversarial_tf.py...
Files already downloaded and verified
Files already downloaded and verified

=== Pretraining External Model (BigTransformer) on Adversarial Augment Set ===
Files already downloaded and verified
Files already downloaded and verified
[Epoch 1] Loss: 2.1760
[Epoch 2] Loss: 2.0653
[Epoch 3] Loss: 2.0164
[Epoch 4] Loss: 1.9653
[Epoch 5] Loss: 1.9401
[Epoch 6] Loss: 1.9016
[Epoch 7] Loss: 1.8998
[Epoch 8] Loss: 1.8657
[Epoch 9] Loss: 1.8450
[Epoch 10] Loss: 1.8335
[Epoch 11] Loss: 1.8134
[Epoch 12] Loss: 1.7878
[Epoch 13] Loss: 1.7810
[Epoch 14] Loss: 1.7573
[Epoch 15] Loss: 1.7440
[Epoch 16] Loss: 1.7217
[Epoch 17] Loss: 1.6949
[Epoch 18] Loss: 1.6690
[Epoch 19] Loss: 1.6461
[Epoch 20] Loss: 1.6185
[Epoch 21] Loss: 1.5878
[Epoch 22] Loss: 1.5817
[Epoch 23] Loss: 1.5646
[Epoch 24] Loss: 1.5525
[Epoch 25] Loss: 1.5250
[Epoch 26] Loss: 1.5133
[Epoch 27] Loss: 1.4987
[Epoch 28] Loss: 1.4873
[Epoch 29] Loss: 1.4751
[Epoch 30] Loss: 1.4553
[Epoch 31] Loss: 1.4535
[Epoch 32] Loss: 1.4277
[Epoch 33] Loss: 1.4308
[Epoch 34] Loss: 1.4155
[Epoch 35] Loss: 1.3932
[Epoch 36] Loss: 1.3849
[Epoch 37] Loss: 1.3691
[Epoch 38] Loss: 1.3596
[Epoch 39] Loss: 1.3728
[Epoch 40] Loss: 1.3463
[Epoch 41] Loss: 1.3395
[Epoch 42] Loss: 1.3300
[Epoch 43] Loss: 1.3178
[Epoch 44] Loss: 1.3031
[Epoch 45] Loss: 1.2904
[Epoch 46] Loss: 1.2917
[Epoch 47] Loss: 1.2789
[Epoch 48] Loss: 1.2686
[Epoch 49] Loss: 1.2719
[Epoch 50] Loss: 1.2538
[Epoch 51] Loss: 1.2401
[Epoch 52] Loss: 1.2399
[Epoch 53] Loss: 1.2167
[Epoch 54] Loss: 1.2184
[Epoch 55] Loss: 1.1952
[Epoch 56] Loss: 1.1922
[Epoch 57] Loss: 1.2021
[Epoch 58] Loss: 1.1828
[Epoch 59] Loss: 1.1868
[Epoch 60] Loss: 1.1694
Trained and saved external model to: ./model_test10/base_adversarial_tf.pt
External Transformer Eval: Acc=32.13% | AUC=0.8651 | F1=0.2827 | MinCAcc=1.00%

=== Run 1/5, seed=42 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (Transformer head) ...
[LoRA-TF] Ep1/30 loss=1.5460
[LoRA-TF] Ep2/30 loss=1.4447
[LoRA-TF] Ep3/30 loss=1.3977
[LoRA-TF] Ep4/30 loss=1.3805
[LoRA-TF] Ep5/30 loss=1.3554
[LoRA-TF] Ep6/30 loss=1.3448
[LoRA-TF] Ep7/30 loss=1.3279
[LoRA-TF] Ep8/30 loss=1.3269
[LoRA-TF] Ep9/30 loss=1.3182
[LoRA-TF] Ep10/30 loss=1.3003
[LoRA-TF] Ep11/30 loss=1.3016
[LoRA-TF] Ep12/30 loss=1.2972
[LoRA-TF] Ep13/30 loss=1.2780
[LoRA-TF] Ep14/30 loss=1.2732
[LoRA-TF] Ep15/30 loss=1.2720
[LoRA-TF] Ep16/30 loss=1.2685
[LoRA-TF] Ep17/30 loss=1.2551
[LoRA-TF] Ep18/30 loss=1.2536
[LoRA-TF] Ep19/30 loss=1.2480
[LoRA-TF] Ep20/30 loss=1.2401
[LoRA-TF] Ep21/30 loss=1.2368
[LoRA-TF] Ep22/30 loss=1.2363
[LoRA-TF] Ep23/30 loss=1.2315
[LoRA-TF] Ep24/30 loss=1.2191
[LoRA-TF] Ep25/30 loss=1.2236
[LoRA-TF] Ep26/30 loss=1.2198
[LoRA-TF] Ep27/30 loss=1.2060
[LoRA-TF] Ep28/30 loss=1.2062
[LoRA-TF] Ep29/30 loss=1.2079
[LoRA-TF] Ep30/30 loss=1.1959
Training DANN-Gate (Transformer head) ...
[DANN-Gate-TF] Ep1/30 λ_grl=0.000 loss=5.2768
[DANN-Gate-TF] Ep2/30 λ_grl=0.171 loss=5.1047
[DANN-Gate-TF] Ep3/30 λ_grl=0.332 loss=4.9505
[DANN-Gate-TF] Ep4/30 λ_grl=0.476 loss=4.9230
[DANN-Gate-TF] Ep5/30 λ_grl=0.598 loss=4.8508
[DANN-Gate-TF] Ep6/30 λ_grl=0.697 loss=4.8456
[DANN-Gate-TF] Ep7/30 λ_grl=0.776 loss=4.7841
[DANN-Gate-TF] Ep8/30 λ_grl=0.836 loss=4.7916
[DANN-Gate-TF] Ep9/30 λ_grl=0.881 loss=4.7488
[DANN-Gate-TF] Ep10/30 λ_grl=0.914 loss=4.7230
[DANN-Gate-TF] Ep11/30 λ_grl=0.938 loss=4.7004
[DANN-Gate-TF] Ep12/30 λ_grl=0.956 loss=4.7187
[DANN-Gate-TF] Ep13/30 λ_grl=0.969 loss=4.6863
[DANN-Gate-TF] Ep14/30 λ_grl=0.978 loss=4.6556
[DANN-Gate-TF] Ep15/30 λ_grl=0.984 loss=4.6263
[DANN-Gate-TF] Ep16/30 λ_grl=0.989 loss=4.6354
[DANN-Gate-TF] Ep17/30 λ_grl=0.992 loss=4.6183
[DANN-Gate-TF] Ep18/30 λ_grl=0.994 loss=4.6091
[DANN-Gate-TF] Ep19/30 λ_grl=0.996 loss=4.5887
[DANN-Gate-TF] Ep20/30 λ_grl=0.997 loss=4.5910
[DANN-Gate-TF] Ep21/30 λ_grl=0.998 loss=4.5587
[DANN-Gate-TF] Ep22/30 λ_grl=0.999 loss=4.5584
[DANN-Gate-TF] Ep23/30 λ_grl=0.999 loss=4.5641
[DANN-Gate-TF] Ep24/30 λ_grl=0.999 loss=4.5780
[DANN-Gate-TF] Ep25/30 λ_grl=0.999 loss=4.5438
[DANN-Gate-TF] Ep26/30 λ_grl=1.000 loss=4.5524
[DANN-Gate-TF] Ep27/30 λ_grl=1.000 loss=4.5407
[DANN-Gate-TF] Ep28/30 λ_grl=1.000 loss=4.5278
[DANN-Gate-TF] Ep29/30 λ_grl=1.000 loss=4.5449
[DANN-Gate-TF] Ep30/30 λ_grl=1.000 loss=4.5109
[Run 1] LoRA Acc=52.35% | DANN-Gate Acc=51.86%

=== Run 2/5, seed=43 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (Transformer head) ...
[LoRA-TF] Ep1/30 loss=1.4794
[LoRA-TF] Ep2/30 loss=1.4014
[LoRA-TF] Ep3/30 loss=1.3638
[LoRA-TF] Ep4/30 loss=1.3551
[LoRA-TF] Ep5/30 loss=1.3315
[LoRA-TF] Ep6/30 loss=1.3126
[LoRA-TF] Ep7/30 loss=1.3005
[LoRA-TF] Ep8/30 loss=1.2855
[LoRA-TF] Ep9/30 loss=1.2768
[LoRA-TF] Ep10/30 loss=1.2583
[LoRA-TF] Ep11/30 loss=1.2634
[LoRA-TF] Ep12/30 loss=1.2473
[LoRA-TF] Ep13/30 loss=1.2409
[LoRA-TF] Ep14/30 loss=1.2307
[LoRA-TF] Ep15/30 loss=1.2406
[LoRA-TF] Ep16/30 loss=1.2246
[LoRA-TF] Ep17/30 loss=1.2187
[LoRA-TF] Ep18/30 loss=1.2193
[LoRA-TF] Ep19/30 loss=1.1973
[LoRA-TF] Ep20/30 loss=1.1990
[LoRA-TF] Ep21/30 loss=1.1940
[LoRA-TF] Ep22/30 loss=1.1906
[LoRA-TF] Ep23/30 loss=1.1837
[LoRA-TF] Ep24/30 loss=1.1806
[LoRA-TF] Ep25/30 loss=1.1846
[LoRA-TF] Ep26/30 loss=1.1806
[LoRA-TF] Ep27/30 loss=1.1745
[LoRA-TF] Ep28/30 loss=1.1635
[LoRA-TF] Ep29/30 loss=1.1582
[LoRA-TF] Ep30/30 loss=1.1534
Training DANN-Gate (Transformer head) ...
[DANN-Gate-TF] Ep1/30 λ_grl=0.000 loss=5.2384
[DANN-Gate-TF] Ep2/30 λ_grl=0.171 loss=4.9681
[DANN-Gate-TF] Ep3/30 λ_grl=0.332 loss=4.8566
[DANN-Gate-TF] Ep4/30 λ_grl=0.476 loss=4.8103
[DANN-Gate-TF] Ep5/30 λ_grl=0.598 loss=4.7643
[DANN-Gate-TF] Ep6/30 λ_grl=0.697 loss=4.7295
[DANN-Gate-TF] Ep7/30 λ_grl=0.776 loss=4.7360
[DANN-Gate-TF] Ep8/30 λ_grl=0.836 loss=4.6548
[DANN-Gate-TF] Ep9/30 λ_grl=0.881 loss=4.6169
[DANN-Gate-TF] Ep10/30 λ_grl=0.914 loss=4.6076
[DANN-Gate-TF] Ep11/30 λ_grl=0.938 loss=4.6049
[DANN-Gate-TF] Ep12/30 λ_grl=0.956 loss=4.6148
[DANN-Gate-TF] Ep13/30 λ_grl=0.969 loss=4.5856
[DANN-Gate-TF] Ep14/30 λ_grl=0.978 loss=4.5501
[DANN-Gate-TF] Ep15/30 λ_grl=0.984 loss=4.5403
[DANN-Gate-TF] Ep16/30 λ_grl=0.989 loss=4.5458
[DANN-Gate-TF] Ep17/30 λ_grl=0.992 loss=4.5368
[DANN-Gate-TF] Ep18/30 λ_grl=0.994 loss=4.5154
[DANN-Gate-TF] Ep19/30 λ_grl=0.996 loss=4.5076
[DANN-Gate-TF] Ep20/30 λ_grl=0.997 loss=4.4900
[DANN-Gate-TF] Ep21/30 λ_grl=0.998 loss=4.4743
[DANN-Gate-TF] Ep22/30 λ_grl=0.999 loss=4.4800
[DANN-Gate-TF] Ep23/30 λ_grl=0.999 loss=4.4537
[DANN-Gate-TF] Ep24/30 λ_grl=0.999 loss=4.4417
[DANN-Gate-TF] Ep25/30 λ_grl=0.999 loss=4.4437
[DANN-Gate-TF] Ep26/30 λ_grl=1.000 loss=4.4421
[DANN-Gate-TF] Ep27/30 λ_grl=1.000 loss=4.4163
[DANN-Gate-TF] Ep28/30 λ_grl=1.000 loss=4.4096
[DANN-Gate-TF] Ep29/30 λ_grl=1.000 loss=4.3956
[DANN-Gate-TF] Ep30/30 λ_grl=1.000 loss=4.4077
[Run 2] LoRA Acc=51.95% | DANN-Gate Acc=52.08%

=== Run 3/5, seed=44 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (Transformer head) ...
[LoRA-TF] Ep1/30 loss=1.5006
[LoRA-TF] Ep2/30 loss=1.4086
[LoRA-TF] Ep3/30 loss=1.3787
[LoRA-TF] Ep4/30 loss=1.3680
[LoRA-TF] Ep5/30 loss=1.3439
[LoRA-TF] Ep6/30 loss=1.3298
[LoRA-TF] Ep7/30 loss=1.3187
[LoRA-TF] Ep8/30 loss=1.3094
[LoRA-TF] Ep9/30 loss=1.2929
[LoRA-TF] Ep10/30 loss=1.2865
[LoRA-TF] Ep11/30 loss=1.2721
[LoRA-TF] Ep12/30 loss=1.2744
[LoRA-TF] Ep13/30 loss=1.2746
[LoRA-TF] Ep14/30 loss=1.2650
[LoRA-TF] Ep15/30 loss=1.2470
[LoRA-TF] Ep16/30 loss=1.2504
[LoRA-TF] Ep17/30 loss=1.2419
[LoRA-TF] Ep18/30 loss=1.2408
[LoRA-TF] Ep19/30 loss=1.2385
[LoRA-TF] Ep20/30 loss=1.2246
[LoRA-TF] Ep21/30 loss=1.2164
[LoRA-TF] Ep22/30 loss=1.2093
[LoRA-TF] Ep23/30 loss=1.2230
[LoRA-TF] Ep24/30 loss=1.2042
[LoRA-TF] Ep25/30 loss=1.2090
[LoRA-TF] Ep26/30 loss=1.2065
[LoRA-TF] Ep27/30 loss=1.1953
[LoRA-TF] Ep28/30 loss=1.1989
[LoRA-TF] Ep29/30 loss=1.1842
[LoRA-TF] Ep30/30 loss=1.1821
Training DANN-Gate (Transformer head) ...
[DANN-Gate-TF] Ep1/30 λ_grl=0.000 loss=5.0986
[DANN-Gate-TF] Ep2/30 λ_grl=0.171 loss=4.9565
[DANN-Gate-TF] Ep3/30 λ_grl=0.332 loss=4.8617
[DANN-Gate-TF] Ep4/30 λ_grl=0.476 loss=4.8456
[DANN-Gate-TF] Ep5/30 λ_grl=0.598 loss=4.8173
[DANN-Gate-TF] Ep6/30 λ_grl=0.697 loss=4.7825
[DANN-Gate-TF] Ep7/30 λ_grl=0.776 loss=4.7414
[DANN-Gate-TF] Ep8/30 λ_grl=0.836 loss=4.7110
[DANN-Gate-TF] Ep9/30 λ_grl=0.881 loss=4.6992
[DANN-Gate-TF] Ep10/30 λ_grl=0.914 loss=4.6826
[DANN-Gate-TF] Ep11/30 λ_grl=0.938 loss=4.6815
[DANN-Gate-TF] Ep12/30 λ_grl=0.956 loss=4.6411
[DANN-Gate-TF] Ep13/30 λ_grl=0.969 loss=4.6368
[DANN-Gate-TF] Ep14/30 λ_grl=0.978 loss=4.6108
[DANN-Gate-TF] Ep15/30 λ_grl=0.984 loss=4.6122
[DANN-Gate-TF] Ep16/30 λ_grl=0.989 loss=4.5987
[DANN-Gate-TF] Ep17/30 λ_grl=0.992 loss=4.5772
[DANN-Gate-TF] Ep18/30 λ_grl=0.994 loss=4.5788
[DANN-Gate-TF] Ep19/30 λ_grl=0.996 loss=4.5522
[DANN-Gate-TF] Ep20/30 λ_grl=0.997 loss=4.5333
[DANN-Gate-TF] Ep21/30 λ_grl=0.998 loss=4.5247
[DANN-Gate-TF] Ep22/30 λ_grl=0.999 loss=4.5236
[DANN-Gate-TF] Ep23/30 λ_grl=0.999 loss=4.5098
[DANN-Gate-TF] Ep24/30 λ_grl=0.999 loss=4.5219
[DANN-Gate-TF] Ep25/30 λ_grl=0.999 loss=4.5153
[DANN-Gate-TF] Ep26/30 λ_grl=1.000 loss=4.5082
[DANN-Gate-TF] Ep27/30 λ_grl=1.000 loss=4.4711
[DANN-Gate-TF] Ep28/30 λ_grl=1.000 loss=4.4870
[DANN-Gate-TF] Ep29/30 λ_grl=1.000 loss=4.4672
[DANN-Gate-TF] Ep30/30 λ_grl=1.000 loss=4.4532
[Run 3] LoRA Acc=52.36% | DANN-Gate Acc=51.70%

=== Run 4/5, seed=45 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (Transformer head) ...
[LoRA-TF] Ep1/30 loss=1.4973
[LoRA-TF] Ep2/30 loss=1.4055
[LoRA-TF] Ep3/30 loss=1.3734
[LoRA-TF] Ep4/30 loss=1.3471
[LoRA-TF] Ep5/30 loss=1.3167
[LoRA-TF] Ep6/30 loss=1.3081
[LoRA-TF] Ep7/30 loss=1.2957
[LoRA-TF] Ep8/30 loss=1.2812
[LoRA-TF] Ep9/30 loss=1.2704
[LoRA-TF] Ep10/30 loss=1.2598
[LoRA-TF] Ep11/30 loss=1.2588
[LoRA-TF] Ep12/30 loss=1.2387
[LoRA-TF] Ep13/30 loss=1.2408
[LoRA-TF] Ep14/30 loss=1.2330
[LoRA-TF] Ep15/30 loss=1.2238
[LoRA-TF] Ep16/30 loss=1.2201
[LoRA-TF] Ep17/30 loss=1.2118
[LoRA-TF] Ep18/30 loss=1.2092
[LoRA-TF] Ep19/30 loss=1.1985
[LoRA-TF] Ep20/30 loss=1.1988
[LoRA-TF] Ep21/30 loss=1.1961
[LoRA-TF] Ep22/30 loss=1.1857
[LoRA-TF] Ep23/30 loss=1.1820
[LoRA-TF] Ep24/30 loss=1.1840
[LoRA-TF] Ep25/30 loss=1.1763
[LoRA-TF] Ep26/30 loss=1.1835
[LoRA-TF] Ep27/30 loss=1.1749
[LoRA-TF] Ep28/30 loss=1.1620
[LoRA-TF] Ep29/30 loss=1.1646
[LoRA-TF] Ep30/30 loss=1.1615
Training DANN-Gate (Transformer head) ...
[DANN-Gate-TF] Ep1/30 λ_grl=0.000 loss=5.2283
[DANN-Gate-TF] Ep2/30 λ_grl=0.171 loss=4.9120
[DANN-Gate-TF] Ep3/30 λ_grl=0.332 loss=4.8197
[DANN-Gate-TF] Ep4/30 λ_grl=0.476 loss=4.7786
[DANN-Gate-TF] Ep5/30 λ_grl=0.598 loss=4.7321
[DANN-Gate-TF] Ep6/30 λ_grl=0.697 loss=4.7069
[DANN-Gate-TF] Ep7/30 λ_grl=0.776 loss=4.6918
[DANN-Gate-TF] Ep8/30 λ_grl=0.836 loss=4.6569
[DANN-Gate-TF] Ep9/30 λ_grl=0.881 loss=4.6504
[DANN-Gate-TF] Ep10/30 λ_grl=0.914 loss=4.6012
[DANN-Gate-TF] Ep11/30 λ_grl=0.938 loss=4.6034
[DANN-Gate-TF] Ep12/30 λ_grl=0.956 loss=4.5560
[DANN-Gate-TF] Ep13/30 λ_grl=0.969 loss=4.5440
[DANN-Gate-TF] Ep14/30 λ_grl=0.978 loss=4.5369
[DANN-Gate-TF] Ep15/30 λ_grl=0.984 loss=4.5188
[DANN-Gate-TF] Ep16/30 λ_grl=0.989 loss=4.5302
[DANN-Gate-TF] Ep17/30 λ_grl=0.992 loss=4.5188
[DANN-Gate-TF] Ep18/30 λ_grl=0.994 loss=4.4976
[DANN-Gate-TF] Ep19/30 λ_grl=0.996 loss=4.5004
[DANN-Gate-TF] Ep20/30 λ_grl=0.997 loss=4.4830
[DANN-Gate-TF] Ep21/30 λ_grl=0.998 loss=4.4785
[DANN-Gate-TF] Ep22/30 λ_grl=0.999 loss=4.4612
[DANN-Gate-TF] Ep23/30 λ_grl=0.999 loss=4.4394
[DANN-Gate-TF] Ep24/30 λ_grl=0.999 loss=4.4333
[DANN-Gate-TF] Ep25/30 λ_grl=0.999 loss=4.4537
[DANN-Gate-TF] Ep26/30 λ_grl=1.000 loss=4.4116
[DANN-Gate-TF] Ep27/30 λ_grl=1.000 loss=4.4329
[DANN-Gate-TF] Ep28/30 λ_grl=1.000 loss=4.4391
[DANN-Gate-TF] Ep29/30 λ_grl=1.000 loss=4.4354
[DANN-Gate-TF] Ep30/30 λ_grl=1.000 loss=4.4014
[Run 4] LoRA Acc=51.98% | DANN-Gate Acc=52.69%

=== Run 5/5, seed=46 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (Transformer head) ...
[LoRA-TF] Ep1/30 loss=1.4938
[LoRA-TF] Ep2/30 loss=1.4100
[LoRA-TF] Ep3/30 loss=1.3798
[LoRA-TF] Ep4/30 loss=1.3607
[LoRA-TF] Ep5/30 loss=1.3425
[LoRA-TF] Ep6/30 loss=1.3237
[LoRA-TF] Ep7/30 loss=1.3064
[LoRA-TF] Ep8/30 loss=1.2909
[LoRA-TF] Ep9/30 loss=1.2784
[LoRA-TF] Ep10/30 loss=1.2772
[LoRA-TF] Ep11/30 loss=1.2604
[LoRA-TF] Ep12/30 loss=1.2501
[LoRA-TF] Ep13/30 loss=1.2448
[LoRA-TF] Ep14/30 loss=1.2413
[LoRA-TF] Ep15/30 loss=1.2282
[LoRA-TF] Ep16/30 loss=1.2199
[LoRA-TF] Ep17/30 loss=1.2183
[LoRA-TF] Ep18/30 loss=1.2170
[LoRA-TF] Ep19/30 loss=1.2077
[LoRA-TF] Ep20/30 loss=1.1937
[LoRA-TF] Ep21/30 loss=1.1980
[LoRA-TF] Ep22/30 loss=1.1958
[LoRA-TF] Ep23/30 loss=1.1834
[LoRA-TF] Ep24/30 loss=1.1810
[LoRA-TF] Ep25/30 loss=1.1790
[LoRA-TF] Ep26/30 loss=1.1710
[LoRA-TF] Ep27/30 loss=1.1728
[LoRA-TF] Ep28/30 loss=1.1717
[LoRA-TF] Ep29/30 loss=1.1633
[LoRA-TF] Ep30/30 loss=1.1613
Training DANN-Gate (Transformer head) ...
[DANN-Gate-TF] Ep1/30 λ_grl=0.000 loss=5.1188
[DANN-Gate-TF] Ep2/30 λ_grl=0.171 loss=4.9645
[DANN-Gate-TF] Ep3/30 λ_grl=0.332 loss=4.9159
[DANN-Gate-TF] Ep4/30 λ_grl=0.476 loss=4.8890
[DANN-Gate-TF] Ep5/30 λ_grl=0.598 loss=4.7792
[DANN-Gate-TF] Ep6/30 λ_grl=0.697 loss=4.7028
[DANN-Gate-TF] Ep7/30 λ_grl=0.776 loss=4.7059
[DANN-Gate-TF] Ep8/30 λ_grl=0.836 loss=4.7157
[DANN-Gate-TF] Ep9/30 λ_grl=0.881 loss=4.6803
[DANN-Gate-TF] Ep10/30 λ_grl=0.914 loss=4.6427
[DANN-Gate-TF] Ep11/30 λ_grl=0.938 loss=4.6102
[DANN-Gate-TF] Ep12/30 λ_grl=0.956 loss=4.6008
[DANN-Gate-TF] Ep13/30 λ_grl=0.969 loss=4.6011
[DANN-Gate-TF] Ep14/30 λ_grl=0.978 loss=4.5874
[DANN-Gate-TF] Ep15/30 λ_grl=0.984 loss=4.5371
[DANN-Gate-TF] Ep16/30 λ_grl=0.989 loss=4.5578
[DANN-Gate-TF] Ep17/30 λ_grl=0.992 loss=4.5365
[DANN-Gate-TF] Ep18/30 λ_grl=0.994 loss=4.5243
[DANN-Gate-TF] Ep19/30 λ_grl=0.996 loss=4.5358
[DANN-Gate-TF] Ep20/30 λ_grl=0.997 loss=4.4885
[DANN-Gate-TF] Ep21/30 λ_grl=0.998 loss=4.4867
[DANN-Gate-TF] Ep22/30 λ_grl=0.999 loss=4.4707
[DANN-Gate-TF] Ep23/30 λ_grl=0.999 loss=4.4524
[DANN-Gate-TF] Ep24/30 λ_grl=0.999 loss=4.4452
[DANN-Gate-TF] Ep25/30 λ_grl=0.999 loss=4.4663
[DANN-Gate-TF] Ep26/30 λ_grl=1.000 loss=4.4210
[DANN-Gate-TF] Ep27/30 λ_grl=1.000 loss=4.4369
[DANN-Gate-TF] Ep28/30 λ_grl=1.000 loss=4.4318
[DANN-Gate-TF] Ep29/30 λ_grl=1.000 loss=4.4090
[DANN-Gate-TF] Ep30/30 λ_grl=1.000 loss=4.4159
[Run 5] LoRA Acc=53.11% | DANN-Gate Acc=52.34%

All done. Final mean/std results saved to: ./results_test10_base/adversarial_tf.json
more_baselines/base_adversarial_tf.py completed successfully.
Starting more_baselines/base_imb_tf.py...
Files already downloaded and verified
Files already downloaded and verified

=== Pretraining External Model (BigTransformer) on 10k Imbalanced Samples ===
[Epoch 1] Loss: 1.5805
[Epoch 2] Loss: 1.4327
[Epoch 3] Loss: 1.3284
[Epoch 4] Loss: 1.2680
[Epoch 5] Loss: 1.1703
[Epoch 6] Loss: 1.1068
[Epoch 7] Loss: 1.0491
[Epoch 8] Loss: 0.9984
[Epoch 9] Loss: 0.9543
[Epoch 10] Loss: 0.9066
[Epoch 11] Loss: 0.8745
[Epoch 12] Loss: 0.8310
[Epoch 13] Loss: 0.8024
[Epoch 14] Loss: 0.7582
[Epoch 15] Loss: 0.7200
[Epoch 16] Loss: 0.6861
[Epoch 17] Loss: 0.6722
[Epoch 18] Loss: 0.6319
[Epoch 19] Loss: 0.6254
[Epoch 20] Loss: 0.5857
[Epoch 21] Loss: 0.5485
[Epoch 22] Loss: 0.5322
[Epoch 23] Loss: 0.5369
[Epoch 24] Loss: 0.4881
[Epoch 25] Loss: 0.4896
[Epoch 26] Loss: 0.4475
[Epoch 27] Loss: 0.4335
[Epoch 28] Loss: 0.4166
[Epoch 29] Loss: 0.4127
[Epoch 30] Loss: 0.3947
[Epoch 31] Loss: 0.3682
[Epoch 32] Loss: 0.3396
[Epoch 33] Loss: 0.3549
[Epoch 34] Loss: 0.3116
[Epoch 35] Loss: 0.2967
[Epoch 36] Loss: 0.2945
[Epoch 37] Loss: 0.2806
[Epoch 38] Loss: 0.3010
[Epoch 39] Loss: 0.2456
[Epoch 40] Loss: 0.2299
[Epoch 41] Loss: 0.2307
[Epoch 42] Loss: 0.2645
[Epoch 43] Loss: 0.2005
[Epoch 44] Loss: 0.1716
[Epoch 45] Loss: 0.1911
[Epoch 46] Loss: 0.1833
[Epoch 47] Loss: 0.1930
[Epoch 48] Loss: 0.1834
[Epoch 49] Loss: 0.1567
[Epoch 50] Loss: 0.1517
[Epoch 51] Loss: 0.1491
[Epoch 52] Loss: 0.1546
[Epoch 53] Loss: 0.1241
[Epoch 54] Loss: 0.1242
[Epoch 55] Loss: 0.1227
[Epoch 56] Loss: 0.1158
[Epoch 57] Loss: 0.1054
[Epoch 58] Loss: 0.1341
[Epoch 59] Loss: 0.0961
[Epoch 60] Loss: 0.1438
Trained and saved external model to: ./model_test10/base_imb_tf.pt
External Transformer Eval: Acc=42.93% | AUC=0.8326 | F1=0.4185 | MinCAcc=11.20%

=== Run 1/5, raw-set seed=42 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (Transformer head; imbalanced pretrain)...
[LoRA-TF] Ep1/30 loss=2.8072
[LoRA-TF] Ep2/30 loss=2.3550
[LoRA-TF] Ep3/30 loss=2.1365
[LoRA-TF] Ep4/30 loss=1.9649
[LoRA-TF] Ep5/30 loss=1.8353
[LoRA-TF] Ep6/30 loss=1.7155
[LoRA-TF] Ep7/30 loss=1.6437
[LoRA-TF] Ep8/30 loss=1.5956
[LoRA-TF] Ep9/30 loss=1.5364
[LoRA-TF] Ep10/30 loss=1.4697
[LoRA-TF] Ep11/30 loss=1.4567
[LoRA-TF] Ep12/30 loss=1.3987
[LoRA-TF] Ep13/30 loss=1.3801
[LoRA-TF] Ep14/30 loss=1.3710
[LoRA-TF] Ep15/30 loss=1.3288
[LoRA-TF] Ep16/30 loss=1.3017
[LoRA-TF] Ep17/30 loss=1.2880
[LoRA-TF] Ep18/30 loss=1.2693
[LoRA-TF] Ep19/30 loss=1.2671
[LoRA-TF] Ep20/30 loss=1.2452
[LoRA-TF] Ep21/30 loss=1.2400
[LoRA-TF] Ep22/30 loss=1.2234
[LoRA-TF] Ep23/30 loss=1.2139
[LoRA-TF] Ep24/30 loss=1.2064
[LoRA-TF] Ep25/30 loss=1.2177
[LoRA-TF] Ep26/30 loss=1.1833
[LoRA-TF] Ep27/30 loss=1.1781
[LoRA-TF] Ep28/30 loss=1.1702
[LoRA-TF] Ep29/30 loss=1.1642
[LoRA-TF] Ep30/30 loss=1.1470
Training DANN-Gate (Transformer head; imbalanced pretrain)...
[DANN-Gate-TF] Ep1/30 λ_grl=0.000 loss=8.3897
[DANN-Gate-TF] Ep2/30 λ_grl=0.171 loss=6.9334
[DANN-Gate-TF] Ep3/30 λ_grl=0.332 loss=6.3718
[DANN-Gate-TF] Ep4/30 λ_grl=0.476 loss=6.0239
[DANN-Gate-TF] Ep5/30 λ_grl=0.598 loss=5.7903
[DANN-Gate-TF] Ep6/30 λ_grl=0.697 loss=5.6397
[DANN-Gate-TF] Ep7/30 λ_grl=0.776 loss=5.4735
[DANN-Gate-TF] Ep8/30 λ_grl=0.836 loss=5.2771
[DANN-Gate-TF] Ep9/30 λ_grl=0.881 loss=5.2207
[DANN-Gate-TF] Ep10/30 λ_grl=0.914 loss=5.0475
[DANN-Gate-TF] Ep11/30 λ_grl=0.938 loss=5.0145
[DANN-Gate-TF] Ep12/30 λ_grl=0.956 loss=4.9462
[DANN-Gate-TF] Ep13/30 λ_grl=0.969 loss=4.8873
[DANN-Gate-TF] Ep14/30 λ_grl=0.978 loss=4.7900
[DANN-Gate-TF] Ep15/30 λ_grl=0.984 loss=4.7392
[DANN-Gate-TF] Ep16/30 λ_grl=0.989 loss=4.7247
[DANN-Gate-TF] Ep17/30 λ_grl=0.992 loss=4.6696
[DANN-Gate-TF] Ep18/30 λ_grl=0.994 loss=4.6502
[DANN-Gate-TF] Ep19/30 λ_grl=0.996 loss=4.6103
[DANN-Gate-TF] Ep20/30 λ_grl=0.997 loss=4.5570
[DANN-Gate-TF] Ep21/30 λ_grl=0.998 loss=4.5101
[DANN-Gate-TF] Ep22/30 λ_grl=0.999 loss=4.5075
[DANN-Gate-TF] Ep23/30 λ_grl=0.999 loss=4.4718
[DANN-Gate-TF] Ep24/30 λ_grl=0.999 loss=4.4238
[DANN-Gate-TF] Ep25/30 λ_grl=0.999 loss=4.4145
[DANN-Gate-TF] Ep26/30 λ_grl=1.000 loss=4.3857
[DANN-Gate-TF] Ep27/30 λ_grl=1.000 loss=4.3801
[DANN-Gate-TF] Ep28/30 λ_grl=1.000 loss=4.4081
[DANN-Gate-TF] Ep29/30 λ_grl=1.000 loss=4.3006
[DANN-Gate-TF] Ep30/30 λ_grl=1.000 loss=4.3114
[Run 1] LoRA Acc=49.02% | DANN-Gate Acc=48.97%

=== Run 2/5, raw-set seed=43 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (Transformer head; imbalanced pretrain)...
[LoRA-TF] Ep1/30 loss=2.5055
[LoRA-TF] Ep2/30 loss=2.2271
[LoRA-TF] Ep3/30 loss=2.0934
[LoRA-TF] Ep4/30 loss=1.9358
[LoRA-TF] Ep5/30 loss=1.8544
[LoRA-TF] Ep6/30 loss=1.7599
[LoRA-TF] Ep7/30 loss=1.6792
[LoRA-TF] Ep8/30 loss=1.6153
[LoRA-TF] Ep9/30 loss=1.5191
[LoRA-TF] Ep10/30 loss=1.4518
[LoRA-TF] Ep11/30 loss=1.3877
[LoRA-TF] Ep12/30 loss=1.3580
[LoRA-TF] Ep13/30 loss=1.3359
[LoRA-TF] Ep14/30 loss=1.3169
[LoRA-TF] Ep15/30 loss=1.2939
[LoRA-TF] Ep16/30 loss=1.2676
[LoRA-TF] Ep17/30 loss=1.2703
[LoRA-TF] Ep18/30 loss=1.2479
[LoRA-TF] Ep19/30 loss=1.2289
[LoRA-TF] Ep20/30 loss=1.2057
[LoRA-TF] Ep21/30 loss=1.2013
[LoRA-TF] Ep22/30 loss=1.1993
[LoRA-TF] Ep23/30 loss=1.1757
[LoRA-TF] Ep24/30 loss=1.1674
[LoRA-TF] Ep25/30 loss=1.1536
[LoRA-TF] Ep26/30 loss=1.1456
[LoRA-TF] Ep27/30 loss=1.1307
[LoRA-TF] Ep28/30 loss=1.1026
[LoRA-TF] Ep29/30 loss=1.0936
[LoRA-TF] Ep30/30 loss=1.1052
Training DANN-Gate (Transformer head; imbalanced pretrain)...
[DANN-Gate-TF] Ep1/30 λ_grl=0.000 loss=7.6199
[DANN-Gate-TF] Ep2/30 λ_grl=0.171 loss=6.8868
[DANN-Gate-TF] Ep3/30 λ_grl=0.332 loss=6.3503
[DANN-Gate-TF] Ep4/30 λ_grl=0.476 loss=6.1210
[DANN-Gate-TF] Ep5/30 λ_grl=0.598 loss=5.8149
[DANN-Gate-TF] Ep6/30 λ_grl=0.697 loss=5.5769
[DANN-Gate-TF] Ep7/30 λ_grl=0.776 loss=5.3901
[DANN-Gate-TF] Ep8/30 λ_grl=0.836 loss=5.2482
[DANN-Gate-TF] Ep9/30 λ_grl=0.881 loss=5.1210
[DANN-Gate-TF] Ep10/30 λ_grl=0.914 loss=5.0238
[DANN-Gate-TF] Ep11/30 λ_grl=0.938 loss=4.8972
[DANN-Gate-TF] Ep12/30 λ_grl=0.956 loss=4.8415
[DANN-Gate-TF] Ep13/30 λ_grl=0.969 loss=4.7195
[DANN-Gate-TF] Ep14/30 λ_grl=0.978 loss=4.6618
[DANN-Gate-TF] Ep15/30 λ_grl=0.984 loss=4.6093
[DANN-Gate-TF] Ep16/30 λ_grl=0.989 loss=4.5559
[DANN-Gate-TF] Ep17/30 λ_grl=0.992 loss=4.5306
[DANN-Gate-TF] Ep18/30 λ_grl=0.994 loss=4.4662
[DANN-Gate-TF] Ep19/30 λ_grl=0.996 loss=4.4188
[DANN-Gate-TF] Ep20/30 λ_grl=0.997 loss=4.3612
[DANN-Gate-TF] Ep21/30 λ_grl=0.998 loss=4.3579
[DANN-Gate-TF] Ep22/30 λ_grl=0.999 loss=4.3435
[DANN-Gate-TF] Ep23/30 λ_grl=0.999 loss=4.3226
[DANN-Gate-TF] Ep24/30 λ_grl=0.999 loss=4.3052
[DANN-Gate-TF] Ep25/30 λ_grl=0.999 loss=4.2760
[DANN-Gate-TF] Ep26/30 λ_grl=1.000 loss=4.2277
[DANN-Gate-TF] Ep27/30 λ_grl=1.000 loss=4.2234
[DANN-Gate-TF] Ep28/30 λ_grl=1.000 loss=4.2147
[DANN-Gate-TF] Ep29/30 λ_grl=1.000 loss=4.1641
[DANN-Gate-TF] Ep30/30 λ_grl=1.000 loss=4.1657
[Run 2] LoRA Acc=49.43% | DANN-Gate Acc=49.23%

=== Run 3/5, raw-set seed=44 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (Transformer head; imbalanced pretrain)...
[LoRA-TF] Ep1/30 loss=2.5966
[LoRA-TF] Ep2/30 loss=2.1756
[LoRA-TF] Ep3/30 loss=1.9889
[LoRA-TF] Ep4/30 loss=1.7816
[LoRA-TF] Ep5/30 loss=1.6904
[LoRA-TF] Ep6/30 loss=1.6094
[LoRA-TF] Ep7/30 loss=1.5439
[LoRA-TF] Ep8/30 loss=1.4656
[LoRA-TF] Ep9/30 loss=1.4149
[LoRA-TF] Ep10/30 loss=1.3742
[LoRA-TF] Ep11/30 loss=1.3415
[LoRA-TF] Ep12/30 loss=1.3050
[LoRA-TF] Ep13/30 loss=1.2685
[LoRA-TF] Ep14/30 loss=1.2388
[LoRA-TF] Ep15/30 loss=1.2186
[LoRA-TF] Ep16/30 loss=1.1735
[LoRA-TF] Ep17/30 loss=1.1872
[LoRA-TF] Ep18/30 loss=1.1619
[LoRA-TF] Ep19/30 loss=1.1369
[LoRA-TF] Ep20/30 loss=1.1327
[LoRA-TF] Ep21/30 loss=1.1035
[LoRA-TF] Ep22/30 loss=1.1041
[LoRA-TF] Ep23/30 loss=1.0974
[LoRA-TF] Ep24/30 loss=1.0919
[LoRA-TF] Ep25/30 loss=1.0946
[LoRA-TF] Ep26/30 loss=1.0524
[LoRA-TF] Ep27/30 loss=1.0609
[LoRA-TF] Ep28/30 loss=1.0638
[LoRA-TF] Ep29/30 loss=1.0474
[LoRA-TF] Ep30/30 loss=1.0467
Training DANN-Gate (Transformer head; imbalanced pretrain)...
[DANN-Gate-TF] Ep1/30 λ_grl=0.000 loss=7.3555
[DANN-Gate-TF] Ep2/30 λ_grl=0.171 loss=6.5303
[DANN-Gate-TF] Ep3/30 λ_grl=0.332 loss=6.1471
[DANN-Gate-TF] Ep4/30 λ_grl=0.476 loss=5.7851
[DANN-Gate-TF] Ep5/30 λ_grl=0.598 loss=5.4742
[DANN-Gate-TF] Ep6/30 λ_grl=0.697 loss=5.3774
[DANN-Gate-TF] Ep7/30 λ_grl=0.776 loss=5.2054
[DANN-Gate-TF] Ep8/30 λ_grl=0.836 loss=5.1054
[DANN-Gate-TF] Ep9/30 λ_grl=0.881 loss=4.9704
[DANN-Gate-TF] Ep10/30 λ_grl=0.914 loss=4.8890
[DANN-Gate-TF] Ep11/30 λ_grl=0.938 loss=4.7922
[DANN-Gate-TF] Ep12/30 λ_grl=0.956 loss=4.7313
[DANN-Gate-TF] Ep13/30 λ_grl=0.969 loss=4.6931
[DANN-Gate-TF] Ep14/30 λ_grl=0.978 loss=4.6362
[DANN-Gate-TF] Ep15/30 λ_grl=0.984 loss=4.5249
[DANN-Gate-TF] Ep16/30 λ_grl=0.989 loss=4.5267
[DANN-Gate-TF] Ep17/30 λ_grl=0.992 loss=4.5002
[DANN-Gate-TF] Ep18/30 λ_grl=0.994 loss=4.4477
[DANN-Gate-TF] Ep19/30 λ_grl=0.996 loss=4.4004
[DANN-Gate-TF] Ep20/30 λ_grl=0.997 loss=4.4038
[DANN-Gate-TF] Ep21/30 λ_grl=0.998 loss=4.3223
[DANN-Gate-TF] Ep22/30 λ_grl=0.999 loss=4.3594
[DANN-Gate-TF] Ep23/30 λ_grl=0.999 loss=4.2609
[DANN-Gate-TF] Ep24/30 λ_grl=0.999 loss=4.3045
[DANN-Gate-TF] Ep25/30 λ_grl=0.999 loss=4.2900
[DANN-Gate-TF] Ep26/30 λ_grl=1.000 loss=4.2552
[DANN-Gate-TF] Ep27/30 λ_grl=1.000 loss=4.2386
[DANN-Gate-TF] Ep28/30 λ_grl=1.000 loss=4.1668
[DANN-Gate-TF] Ep29/30 λ_grl=1.000 loss=4.2063
[DANN-Gate-TF] Ep30/30 λ_grl=1.000 loss=4.1842
[Run 3] LoRA Acc=48.91% | DANN-Gate Acc=49.51%

=== Run 4/5, raw-set seed=45 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (Transformer head; imbalanced pretrain)...
[LoRA-TF] Ep1/30 loss=2.5210
[LoRA-TF] Ep2/30 loss=2.2170
[LoRA-TF] Ep3/30 loss=1.9896
[LoRA-TF] Ep4/30 loss=1.7824
[LoRA-TF] Ep5/30 loss=1.6650
[LoRA-TF] Ep6/30 loss=1.5887
[LoRA-TF] Ep7/30 loss=1.5117
[LoRA-TF] Ep8/30 loss=1.4648
[LoRA-TF] Ep9/30 loss=1.4259
[LoRA-TF] Ep10/30 loss=1.3927
[LoRA-TF] Ep11/30 loss=1.3373
[LoRA-TF] Ep12/30 loss=1.3102
[LoRA-TF] Ep13/30 loss=1.2878
[LoRA-TF] Ep14/30 loss=1.2707
[LoRA-TF] Ep15/30 loss=1.2463
[LoRA-TF] Ep16/30 loss=1.2212
[LoRA-TF] Ep17/30 loss=1.2163
[LoRA-TF] Ep18/30 loss=1.1866
[LoRA-TF] Ep19/30 loss=1.1578
[LoRA-TF] Ep20/30 loss=1.1712
[LoRA-TF] Ep21/30 loss=1.1492
[LoRA-TF] Ep22/30 loss=1.1410
[LoRA-TF] Ep23/30 loss=1.1137
[LoRA-TF] Ep24/30 loss=1.1050
[LoRA-TF] Ep25/30 loss=1.1097
[LoRA-TF] Ep26/30 loss=1.0870
[LoRA-TF] Ep27/30 loss=1.0875
[LoRA-TF] Ep28/30 loss=1.0605
[LoRA-TF] Ep29/30 loss=1.0516
[LoRA-TF] Ep30/30 loss=1.0608
Training DANN-Gate (Transformer head; imbalanced pretrain)...
[DANN-Gate-TF] Ep1/30 λ_grl=0.000 loss=7.8437
[DANN-Gate-TF] Ep2/30 λ_grl=0.171 loss=6.5620
[DANN-Gate-TF] Ep3/30 λ_grl=0.332 loss=6.0168
[DANN-Gate-TF] Ep4/30 λ_grl=0.476 loss=5.7337
[DANN-Gate-TF] Ep5/30 λ_grl=0.598 loss=5.4349
[DANN-Gate-TF] Ep6/30 λ_grl=0.697 loss=5.3747
[DANN-Gate-TF] Ep7/30 λ_grl=0.776 loss=5.1599
[DANN-Gate-TF] Ep8/30 λ_grl=0.836 loss=5.0599
[DANN-Gate-TF] Ep9/30 λ_grl=0.881 loss=4.9333
[DANN-Gate-TF] Ep10/30 λ_grl=0.914 loss=4.8033
[DANN-Gate-TF] Ep11/30 λ_grl=0.938 loss=4.7998
[DANN-Gate-TF] Ep12/30 λ_grl=0.956 loss=4.6338
[DANN-Gate-TF] Ep13/30 λ_grl=0.969 loss=4.5993
[DANN-Gate-TF] Ep14/30 λ_grl=0.978 loss=4.6154
[DANN-Gate-TF] Ep15/30 λ_grl=0.984 loss=4.5766
[DANN-Gate-TF] Ep16/30 λ_grl=0.989 loss=4.5067
[DANN-Gate-TF] Ep17/30 λ_grl=0.992 loss=4.4477
[DANN-Gate-TF] Ep18/30 λ_grl=0.994 loss=4.4114
[DANN-Gate-TF] Ep19/30 λ_grl=0.996 loss=4.3982
[DANN-Gate-TF] Ep20/30 λ_grl=0.997 loss=4.3698
[DANN-Gate-TF] Ep21/30 λ_grl=0.998 loss=4.3636
[DANN-Gate-TF] Ep22/30 λ_grl=0.999 loss=4.2856
[DANN-Gate-TF] Ep23/30 λ_grl=0.999 loss=4.2832
[DANN-Gate-TF] Ep24/30 λ_grl=0.999 loss=4.2776
[DANN-Gate-TF] Ep25/30 λ_grl=0.999 loss=4.2627
[DANN-Gate-TF] Ep26/30 λ_grl=1.000 loss=4.1767
[DANN-Gate-TF] Ep27/30 λ_grl=1.000 loss=4.2236
[DANN-Gate-TF] Ep28/30 λ_grl=1.000 loss=4.1776
[DANN-Gate-TF] Ep29/30 λ_grl=1.000 loss=4.1214
[DANN-Gate-TF] Ep30/30 λ_grl=1.000 loss=4.1606
[Run 4] LoRA Acc=48.50% | DANN-Gate Acc=48.41%

=== Run 5/5, raw-set seed=46 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (Transformer head; imbalanced pretrain)...
[LoRA-TF] Ep1/30 loss=2.6061
[LoRA-TF] Ep2/30 loss=2.2415
[LoRA-TF] Ep3/30 loss=2.0206
[LoRA-TF] Ep4/30 loss=1.8582
[LoRA-TF] Ep5/30 loss=1.7056
[LoRA-TF] Ep6/30 loss=1.6201
[LoRA-TF] Ep7/30 loss=1.5498
[LoRA-TF] Ep8/30 loss=1.4682
[LoRA-TF] Ep9/30 loss=1.4288
[LoRA-TF] Ep10/30 loss=1.3946
[LoRA-TF] Ep11/30 loss=1.3394
[LoRA-TF] Ep12/30 loss=1.3131
[LoRA-TF] Ep13/30 loss=1.2791
[LoRA-TF] Ep14/30 loss=1.2546
[LoRA-TF] Ep15/30 loss=1.2302
[LoRA-TF] Ep16/30 loss=1.2317
[LoRA-TF] Ep17/30 loss=1.1982
[LoRA-TF] Ep18/30 loss=1.1923
[LoRA-TF] Ep19/30 loss=1.1794
[LoRA-TF] Ep20/30 loss=1.1695
[LoRA-TF] Ep21/30 loss=1.1406
[LoRA-TF] Ep22/30 loss=1.1444
[LoRA-TF] Ep23/30 loss=1.1106
[LoRA-TF] Ep24/30 loss=1.1111
[LoRA-TF] Ep25/30 loss=1.1081
[LoRA-TF] Ep26/30 loss=1.1126
[LoRA-TF] Ep27/30 loss=1.0754
[LoRA-TF] Ep28/30 loss=1.0958
[LoRA-TF] Ep29/30 loss=1.0828
[LoRA-TF] Ep30/30 loss=1.0835
Training DANN-Gate (Transformer head; imbalanced pretrain)...
[DANN-Gate-TF] Ep1/30 λ_grl=0.000 loss=7.4677
[DANN-Gate-TF] Ep2/30 λ_grl=0.171 loss=6.8921
[DANN-Gate-TF] Ep3/30 λ_grl=0.332 loss=6.4453
[DANN-Gate-TF] Ep4/30 λ_grl=0.476 loss=5.8939
[DANN-Gate-TF] Ep5/30 λ_grl=0.598 loss=5.5492
[DANN-Gate-TF] Ep6/30 λ_grl=0.697 loss=5.2845
[DANN-Gate-TF] Ep7/30 λ_grl=0.776 loss=5.1558
[DANN-Gate-TF] Ep8/30 λ_grl=0.836 loss=4.9990
[DANN-Gate-TF] Ep9/30 λ_grl=0.881 loss=4.8951
[DANN-Gate-TF] Ep10/30 λ_grl=0.914 loss=4.8217
[DANN-Gate-TF] Ep11/30 λ_grl=0.938 loss=4.7054
[DANN-Gate-TF] Ep12/30 λ_grl=0.956 loss=4.7147
[DANN-Gate-TF] Ep13/30 λ_grl=0.969 loss=4.6466
[DANN-Gate-TF] Ep14/30 λ_grl=0.978 loss=4.5501
[DANN-Gate-TF] Ep15/30 λ_grl=0.984 loss=4.5421
[DANN-Gate-TF] Ep16/30 λ_grl=0.989 loss=4.5022
[DANN-Gate-TF] Ep17/30 λ_grl=0.992 loss=4.4469
[DANN-Gate-TF] Ep18/30 λ_grl=0.994 loss=4.4209
[DANN-Gate-TF] Ep19/30 λ_grl=0.996 loss=4.3662
[DANN-Gate-TF] Ep20/30 λ_grl=0.997 loss=4.3845
[DANN-Gate-TF] Ep21/30 λ_grl=0.998 loss=4.3056
[DANN-Gate-TF] Ep22/30 λ_grl=0.999 loss=4.2744
[DANN-Gate-TF] Ep23/30 λ_grl=0.999 loss=4.2642
[DANN-Gate-TF] Ep24/30 λ_grl=0.999 loss=4.2588
[DANN-Gate-TF] Ep25/30 λ_grl=0.999 loss=4.2550
[DANN-Gate-TF] Ep26/30 λ_grl=1.000 loss=4.2262
[DANN-Gate-TF] Ep27/30 λ_grl=1.000 loss=4.1699
[DANN-Gate-TF] Ep28/30 λ_grl=1.000 loss=4.1913
[DANN-Gate-TF] Ep29/30 λ_grl=1.000 loss=4.1683
[DANN-Gate-TF] Ep30/30 λ_grl=1.000 loss=4.1927
[Run 5] LoRA Acc=49.08% | DANN-Gate Acc=48.57%

All done. Final mean/std results saved to: ./results_test10_base/imb_tf.json
more_baselines/base_imb_tf.py completed successfully.
Starting more_baselines/base_mismatch_tf.py...

=== Pretraining Teacher (BigTransformer-100) on CIFAR-100 subset ===
Files already downloaded and verified
Files already downloaded and verified
[Epoch 1] Loss: 4.3771
[Epoch 2] Loss: 4.1335
[Epoch 3] Loss: 3.9701
[Epoch 4] Loss: 3.8520
[Epoch 5] Loss: 3.7278
[Epoch 6] Loss: 3.6337
[Epoch 7] Loss: 3.5387
[Epoch 8] Loss: 3.4465
[Epoch 9] Loss: 3.3576
[Epoch 10] Loss: 3.2543
[Epoch 11] Loss: 3.1834
[Epoch 12] Loss: 3.0875
[Epoch 13] Loss: 2.9916
[Epoch 14] Loss: 2.9175
[Epoch 15] Loss: 2.8027
[Epoch 16] Loss: 2.7218
[Epoch 17] Loss: 2.6128
[Epoch 18] Loss: 2.5320
[Epoch 19] Loss: 2.4408
[Epoch 20] Loss: 2.3792
[Epoch 21] Loss: 2.2715
[Epoch 22] Loss: 2.1885
[Epoch 23] Loss: 2.0782
[Epoch 24] Loss: 1.9955
[Epoch 25] Loss: 1.8953
[Epoch 26] Loss: 1.7989
[Epoch 27] Loss: 1.6942
[Epoch 28] Loss: 1.5905
[Epoch 29] Loss: 1.4933
[Epoch 30] Loss: 1.4119
[Epoch 31] Loss: 1.3557
[Epoch 32] Loss: 1.2277
[Epoch 33] Loss: 1.1590
[Epoch 34] Loss: 1.0795
[Epoch 35] Loss: 1.0021
[Epoch 36] Loss: 0.9397
[Epoch 37] Loss: 0.8786
[Epoch 38] Loss: 0.7975
[Epoch 39] Loss: 0.7544
[Epoch 40] Loss: 0.7114
[Epoch 41] Loss: 0.6057
[Epoch 42] Loss: 0.6559
[Epoch 43] Loss: 0.5933
[Epoch 44] Loss: 0.5413
[Epoch 45] Loss: 0.5620
[Epoch 46] Loss: 0.4842
[Epoch 47] Loss: 0.4608
[Epoch 48] Loss: 0.4387
[Epoch 49] Loss: 0.3736
[Epoch 50] Loss: 0.3945
[Epoch 51] Loss: 0.3546
[Epoch 52] Loss: 0.3172
[Epoch 53] Loss: 0.3482
[Epoch 54] Loss: 0.3565
[Epoch 55] Loss: 0.3076
[Epoch 56] Loss: 0.2974
[Epoch 57] Loss: 0.2555
[Epoch 58] Loss: 0.2603
[Epoch 59] Loss: 0.2496
[Epoch 60] Loss: 0.2244
Trained and saved teacher transformer to: ./model_test10/base_mismatch_tf_teacher_cifar100.pt
Files already downloaded and verified
Files already downloaded and verified

=== Run 1/5, seed=42 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (teacher -> 10-class head + LoRA) ...
[LoRA-TF] Ep1/30 loss=1.9937
[LoRA-TF] Ep2/30 loss=1.5777
[LoRA-TF] Ep3/30 loss=1.4613
[LoRA-TF] Ep4/30 loss=1.3943
[LoRA-TF] Ep5/30 loss=1.3675
[LoRA-TF] Ep6/30 loss=1.3231
[LoRA-TF] Ep7/30 loss=1.2680
[LoRA-TF] Ep8/30 loss=1.2491
[LoRA-TF] Ep9/30 loss=1.2157
[LoRA-TF] Ep10/30 loss=1.1767
[LoRA-TF] Ep11/30 loss=1.1650
[LoRA-TF] Ep12/30 loss=1.1393
[LoRA-TF] Ep13/30 loss=1.1309
[LoRA-TF] Ep14/30 loss=1.1111
[LoRA-TF] Ep15/30 loss=1.0782
[LoRA-TF] Ep16/30 loss=1.0803
[LoRA-TF] Ep17/30 loss=1.0896
[LoRA-TF] Ep18/30 loss=1.0563
[LoRA-TF] Ep19/30 loss=1.0413
[LoRA-TF] Ep20/30 loss=1.0237
[LoRA-TF] Ep21/30 loss=1.0338
[LoRA-TF] Ep22/30 loss=1.0308
[LoRA-TF] Ep23/30 loss=1.0055
[LoRA-TF] Ep24/30 loss=0.9961
[LoRA-TF] Ep25/30 loss=0.9825
[LoRA-TF] Ep26/30 loss=0.9779
[LoRA-TF] Ep27/30 loss=1.0003
[LoRA-TF] Ep28/30 loss=0.9827
[LoRA-TF] Ep29/30 loss=0.9673
[LoRA-TF] Ep30/30 loss=0.9557
Training DANN-Gate (teacher -> 10-class head) ...
[DANN-Gate-TF] Ep1/30 λ_grl=0.000 loss=6.2276
[DANN-Gate-TF] Ep2/30 λ_grl=0.171 loss=5.3153
[DANN-Gate-TF] Ep3/30 λ_grl=0.332 loss=5.0287
[DANN-Gate-TF] Ep4/30 λ_grl=0.476 loss=4.9146
[DANN-Gate-TF] Ep5/30 λ_grl=0.598 loss=4.8227
[DANN-Gate-TF] Ep6/30 λ_grl=0.697 loss=4.7163
[DANN-Gate-TF] Ep7/30 λ_grl=0.776 loss=4.6480
[DANN-Gate-TF] Ep8/30 λ_grl=0.836 loss=4.5641
[DANN-Gate-TF] Ep9/30 λ_grl=0.881 loss=4.5642
[DANN-Gate-TF] Ep10/30 λ_grl=0.914 loss=4.4649
[DANN-Gate-TF] Ep11/30 λ_grl=0.938 loss=4.4397
[DANN-Gate-TF] Ep12/30 λ_grl=0.956 loss=4.4161
[DANN-Gate-TF] Ep13/30 λ_grl=0.969 loss=4.3612
[DANN-Gate-TF] Ep14/30 λ_grl=0.978 loss=4.3610
[DANN-Gate-TF] Ep15/30 λ_grl=0.984 loss=4.2589
[DANN-Gate-TF] Ep16/30 λ_grl=0.989 loss=4.2817
[DANN-Gate-TF] Ep17/30 λ_grl=0.992 loss=4.2527
[DANN-Gate-TF] Ep18/30 λ_grl=0.994 loss=4.1853
[DANN-Gate-TF] Ep19/30 λ_grl=0.996 loss=4.1949
[DANN-Gate-TF] Ep20/30 λ_grl=0.997 loss=4.1535
[DANN-Gate-TF] Ep21/30 λ_grl=0.998 loss=4.1487
[DANN-Gate-TF] Ep22/30 λ_grl=0.999 loss=4.1411
[DANN-Gate-TF] Ep23/30 λ_grl=0.999 loss=4.1499
[DANN-Gate-TF] Ep24/30 λ_grl=0.999 loss=4.0894
[DANN-Gate-TF] Ep25/30 λ_grl=0.999 loss=4.0857
[DANN-Gate-TF] Ep26/30 λ_grl=1.000 loss=4.0483
[DANN-Gate-TF] Ep27/30 λ_grl=1.000 loss=4.0322
[DANN-Gate-TF] Ep28/30 λ_grl=1.000 loss=4.0555
[DANN-Gate-TF] Ep29/30 λ_grl=1.000 loss=4.0395
[DANN-Gate-TF] Ep30/30 λ_grl=1.000 loss=4.0242
[Run 1] LoRA Acc=47.42% | DANN-Gate Acc=48.03%

=== Run 2/5, seed=43 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (teacher -> 10-class head + LoRA) ...
[LoRA-TF] Ep1/30 loss=2.0369
[LoRA-TF] Ep2/30 loss=1.5846
[LoRA-TF] Ep3/30 loss=1.4681
[LoRA-TF] Ep4/30 loss=1.4154
[LoRA-TF] Ep5/30 loss=1.3544
[LoRA-TF] Ep6/30 loss=1.3163
[LoRA-TF] Ep7/30 loss=1.2655
[LoRA-TF] Ep8/30 loss=1.2329
[LoRA-TF] Ep9/30 loss=1.2164
[LoRA-TF] Ep10/30 loss=1.1916
[LoRA-TF] Ep11/30 loss=1.1444
[LoRA-TF] Ep12/30 loss=1.1235
[LoRA-TF] Ep13/30 loss=1.1027
[LoRA-TF] Ep14/30 loss=1.0755
[LoRA-TF] Ep15/30 loss=1.0817
[LoRA-TF] Ep16/30 loss=1.0521
[LoRA-TF] Ep17/30 loss=1.0630
[LoRA-TF] Ep18/30 loss=1.0363
[LoRA-TF] Ep19/30 loss=1.0424
[LoRA-TF] Ep20/30 loss=1.0303
[LoRA-TF] Ep21/30 loss=1.0331
[LoRA-TF] Ep22/30 loss=0.9896
[LoRA-TF] Ep23/30 loss=0.9943
[LoRA-TF] Ep24/30 loss=0.9865
[LoRA-TF] Ep25/30 loss=0.9885
[LoRA-TF] Ep26/30 loss=0.9894
[LoRA-TF] Ep27/30 loss=0.9818
[LoRA-TF] Ep28/30 loss=0.9805
[LoRA-TF] Ep29/30 loss=0.9736
[LoRA-TF] Ep30/30 loss=0.9894
Training DANN-Gate (teacher -> 10-class head) ...
[DANN-Gate-TF] Ep1/30 λ_grl=0.000 loss=5.9586
[DANN-Gate-TF] Ep2/30 λ_grl=0.171 loss=5.2228
[DANN-Gate-TF] Ep3/30 λ_grl=0.332 loss=5.0696
[DANN-Gate-TF] Ep4/30 λ_grl=0.476 loss=4.8734
[DANN-Gate-TF] Ep5/30 λ_grl=0.598 loss=4.8412
[DANN-Gate-TF] Ep6/30 λ_grl=0.697 loss=4.7437
[DANN-Gate-TF] Ep7/30 λ_grl=0.776 loss=4.7004
[DANN-Gate-TF] Ep8/30 λ_grl=0.836 loss=4.5853
[DANN-Gate-TF] Ep9/30 λ_grl=0.881 loss=4.5488
[DANN-Gate-TF] Ep10/30 λ_grl=0.914 loss=4.4782
[DANN-Gate-TF] Ep11/30 λ_grl=0.938 loss=4.4462
[DANN-Gate-TF] Ep12/30 λ_grl=0.956 loss=4.4179
[DANN-Gate-TF] Ep13/30 λ_grl=0.969 loss=4.3843
[DANN-Gate-TF] Ep14/30 λ_grl=0.978 loss=4.3472
[DANN-Gate-TF] Ep15/30 λ_grl=0.984 loss=4.2505
[DANN-Gate-TF] Ep16/30 λ_grl=0.989 loss=4.2321
[DANN-Gate-TF] Ep17/30 λ_grl=0.992 loss=4.2022
[DANN-Gate-TF] Ep18/30 λ_grl=0.994 loss=4.2115
[DANN-Gate-TF] Ep19/30 λ_grl=0.996 loss=4.1321
[DANN-Gate-TF] Ep20/30 λ_grl=0.997 loss=4.1532
[DANN-Gate-TF] Ep21/30 λ_grl=0.998 loss=4.0952
[DANN-Gate-TF] Ep22/30 λ_grl=0.999 loss=4.1262
[DANN-Gate-TF] Ep23/30 λ_grl=0.999 loss=4.1463
[DANN-Gate-TF] Ep24/30 λ_grl=0.999 loss=4.1045
[DANN-Gate-TF] Ep25/30 λ_grl=0.999 loss=4.1056
[DANN-Gate-TF] Ep26/30 λ_grl=1.000 loss=4.1002
[DANN-Gate-TF] Ep27/30 λ_grl=1.000 loss=4.0805
[DANN-Gate-TF] Ep28/30 λ_grl=1.000 loss=4.0714
[DANN-Gate-TF] Ep29/30 λ_grl=1.000 loss=4.0384
[DANN-Gate-TF] Ep30/30 λ_grl=1.000 loss=4.0277
[Run 2] LoRA Acc=48.47% | DANN-Gate Acc=48.41%

=== Run 3/5, seed=44 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (teacher -> 10-class head + LoRA) ...
[LoRA-TF] Ep1/30 loss=2.0387
[LoRA-TF] Ep2/30 loss=1.6397
[LoRA-TF] Ep3/30 loss=1.5029
[LoRA-TF] Ep4/30 loss=1.4333
[LoRA-TF] Ep5/30 loss=1.3735
[LoRA-TF] Ep6/30 loss=1.3394
[LoRA-TF] Ep7/30 loss=1.2884
[LoRA-TF] Ep8/30 loss=1.2655
[LoRA-TF] Ep9/30 loss=1.2352
[LoRA-TF] Ep10/30 loss=1.1932
[LoRA-TF] Ep11/30 loss=1.1656
[LoRA-TF] Ep12/30 loss=1.1371
[LoRA-TF] Ep13/30 loss=1.1243
[LoRA-TF] Ep14/30 loss=1.1090
[LoRA-TF] Ep15/30 loss=1.0997
[LoRA-TF] Ep16/30 loss=1.0765
[LoRA-TF] Ep17/30 loss=1.0533
[LoRA-TF] Ep18/30 loss=1.0550
[LoRA-TF] Ep19/30 loss=1.0343
[LoRA-TF] Ep20/30 loss=1.0429
[LoRA-TF] Ep21/30 loss=1.0306
[LoRA-TF] Ep22/30 loss=1.0281
[LoRA-TF] Ep23/30 loss=0.9913
[LoRA-TF] Ep24/30 loss=1.0123
[LoRA-TF] Ep25/30 loss=1.0155
[LoRA-TF] Ep26/30 loss=1.0022
[LoRA-TF] Ep27/30 loss=0.9929
[LoRA-TF] Ep28/30 loss=0.9687
[LoRA-TF] Ep29/30 loss=0.9868
[LoRA-TF] Ep30/30 loss=0.9841
Training DANN-Gate (teacher -> 10-class head) ...
[DANN-Gate-TF] Ep1/30 λ_grl=0.000 loss=5.9097
[DANN-Gate-TF] Ep2/30 λ_grl=0.171 loss=5.3968
[DANN-Gate-TF] Ep3/30 λ_grl=0.332 loss=5.1356
[DANN-Gate-TF] Ep4/30 λ_grl=0.476 loss=4.9866
[DANN-Gate-TF] Ep5/30 λ_grl=0.598 loss=4.8994
[DANN-Gate-TF] Ep6/30 λ_grl=0.697 loss=4.7989
[DANN-Gate-TF] Ep7/30 λ_grl=0.776 loss=4.7321
[DANN-Gate-TF] Ep8/30 λ_grl=0.836 loss=4.6534
[DANN-Gate-TF] Ep9/30 λ_grl=0.881 loss=4.5929
[DANN-Gate-TF] Ep10/30 λ_grl=0.914 loss=4.5302
[DANN-Gate-TF] Ep11/30 λ_grl=0.938 loss=4.4665
[DANN-Gate-TF] Ep12/30 λ_grl=0.956 loss=4.4392
[DANN-Gate-TF] Ep13/30 λ_grl=0.969 loss=4.3976
[DANN-Gate-TF] Ep14/30 λ_grl=0.978 loss=4.3644
[DANN-Gate-TF] Ep15/30 λ_grl=0.984 loss=4.2981
[DANN-Gate-TF] Ep16/30 λ_grl=0.989 loss=4.2887
[DANN-Gate-TF] Ep17/30 λ_grl=0.992 loss=4.3000
[DANN-Gate-TF] Ep18/30 λ_grl=0.994 loss=4.2564
[DANN-Gate-TF] Ep19/30 λ_grl=0.996 loss=4.2515
[DANN-Gate-TF] Ep20/30 λ_grl=0.997 loss=4.1809
[DANN-Gate-TF] Ep21/30 λ_grl=0.998 loss=4.1805
[DANN-Gate-TF] Ep22/30 λ_grl=0.999 loss=4.2044
[DANN-Gate-TF] Ep23/30 λ_grl=0.999 loss=4.1763
[DANN-Gate-TF] Ep24/30 λ_grl=0.999 loss=4.0939
[DANN-Gate-TF] Ep25/30 λ_grl=0.999 loss=4.0853
[DANN-Gate-TF] Ep26/30 λ_grl=1.000 loss=4.1518
[DANN-Gate-TF] Ep27/30 λ_grl=1.000 loss=4.1031
[DANN-Gate-TF] Ep28/30 λ_grl=1.000 loss=4.0811
[DANN-Gate-TF] Ep29/30 λ_grl=1.000 loss=4.0670
[DANN-Gate-TF] Ep30/30 λ_grl=1.000 loss=4.0673
[Run 3] LoRA Acc=47.46% | DANN-Gate Acc=47.39%

=== Run 4/5, seed=45 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (teacher -> 10-class head + LoRA) ...
[LoRA-TF] Ep1/30 loss=2.0712
[LoRA-TF] Ep2/30 loss=1.6453
[LoRA-TF] Ep3/30 loss=1.5115
[LoRA-TF] Ep4/30 loss=1.4334
[LoRA-TF] Ep5/30 loss=1.3825
[LoRA-TF] Ep6/30 loss=1.3407
[LoRA-TF] Ep7/30 loss=1.3166
[LoRA-TF] Ep8/30 loss=1.2763
[LoRA-TF] Ep9/30 loss=1.2482
[LoRA-TF] Ep10/30 loss=1.2154
[LoRA-TF] Ep11/30 loss=1.1839
[LoRA-TF] Ep12/30 loss=1.1749
[LoRA-TF] Ep13/30 loss=1.1518
[LoRA-TF] Ep14/30 loss=1.1353
[LoRA-TF] Ep15/30 loss=1.1350
[LoRA-TF] Ep16/30 loss=1.1059
[LoRA-TF] Ep17/30 loss=1.0791
[LoRA-TF] Ep18/30 loss=1.0596
[LoRA-TF] Ep19/30 loss=1.0738
[LoRA-TF] Ep20/30 loss=1.0598
[LoRA-TF] Ep21/30 loss=1.0304
[LoRA-TF] Ep22/30 loss=1.0460
[LoRA-TF] Ep23/30 loss=1.0333
[LoRA-TF] Ep24/30 loss=1.0100
[LoRA-TF] Ep25/30 loss=1.0283
[LoRA-TF] Ep26/30 loss=0.9971
[LoRA-TF] Ep27/30 loss=1.0196
[LoRA-TF] Ep28/30 loss=0.9867
[LoRA-TF] Ep29/30 loss=0.9900
[LoRA-TF] Ep30/30 loss=1.0021
Training DANN-Gate (teacher -> 10-class head) ...
[DANN-Gate-TF] Ep1/30 λ_grl=0.000 loss=6.1783
[DANN-Gate-TF] Ep2/30 λ_grl=0.171 loss=5.3686
[DANN-Gate-TF] Ep3/30 λ_grl=0.332 loss=5.1235
[DANN-Gate-TF] Ep4/30 λ_grl=0.476 loss=4.9586
[DANN-Gate-TF] Ep5/30 λ_grl=0.598 loss=4.8910
[DANN-Gate-TF] Ep6/30 λ_grl=0.697 loss=4.7851
[DANN-Gate-TF] Ep7/30 λ_grl=0.776 loss=4.6902
[DANN-Gate-TF] Ep8/30 λ_grl=0.836 loss=4.6465
[DANN-Gate-TF] Ep9/30 λ_grl=0.881 loss=4.5801
[DANN-Gate-TF] Ep10/30 λ_grl=0.914 loss=4.4768
[DANN-Gate-TF] Ep11/30 λ_grl=0.938 loss=4.4531
[DANN-Gate-TF] Ep12/30 λ_grl=0.956 loss=4.4090
[DANN-Gate-TF] Ep13/30 λ_grl=0.969 loss=4.3815
[DANN-Gate-TF] Ep14/30 λ_grl=0.978 loss=4.3069
[DANN-Gate-TF] Ep15/30 λ_grl=0.984 loss=4.3200
[DANN-Gate-TF] Ep16/30 λ_grl=0.989 loss=4.2785
[DANN-Gate-TF] Ep17/30 λ_grl=0.992 loss=4.2105
[DANN-Gate-TF] Ep18/30 λ_grl=0.994 loss=4.2352
[DANN-Gate-TF] Ep19/30 λ_grl=0.996 loss=4.1764
[DANN-Gate-TF] Ep20/30 λ_grl=0.997 loss=4.1676
[DANN-Gate-TF] Ep21/30 λ_grl=0.998 loss=4.1774
[DANN-Gate-TF] Ep22/30 λ_grl=0.999 loss=4.1706
[DANN-Gate-TF] Ep23/30 λ_grl=0.999 loss=4.1616
[DANN-Gate-TF] Ep24/30 λ_grl=0.999 loss=4.1591
[DANN-Gate-TF] Ep25/30 λ_grl=0.999 loss=4.1092
[DANN-Gate-TF] Ep26/30 λ_grl=1.000 loss=4.0972
[DANN-Gate-TF] Ep27/30 λ_grl=1.000 loss=4.0946
[DANN-Gate-TF] Ep28/30 λ_grl=1.000 loss=4.0396
[DANN-Gate-TF] Ep29/30 λ_grl=1.000 loss=4.0522
[DANN-Gate-TF] Ep30/30 λ_grl=1.000 loss=4.1426
[Run 4] LoRA Acc=47.67% | DANN-Gate Acc=47.34%

=== Run 5/5, seed=46 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (teacher -> 10-class head + LoRA) ...
[LoRA-TF] Ep1/30 loss=2.0299
[LoRA-TF] Ep2/30 loss=1.5983
[LoRA-TF] Ep3/30 loss=1.4768
[LoRA-TF] Ep4/30 loss=1.4074
[LoRA-TF] Ep5/30 loss=1.3755
[LoRA-TF] Ep6/30 loss=1.3284
[LoRA-TF] Ep7/30 loss=1.2914
[LoRA-TF] Ep8/30 loss=1.2695
[LoRA-TF] Ep9/30 loss=1.2343
[LoRA-TF] Ep10/30 loss=1.2117
[LoRA-TF] Ep11/30 loss=1.1794
[LoRA-TF] Ep12/30 loss=1.1557
[LoRA-TF] Ep13/30 loss=1.1331
[LoRA-TF] Ep14/30 loss=1.1074
[LoRA-TF] Ep15/30 loss=1.1072
[LoRA-TF] Ep16/30 loss=1.0999
[LoRA-TF] Ep17/30 loss=1.0749
[LoRA-TF] Ep18/30 loss=1.0683
[LoRA-TF] Ep19/30 loss=1.0631
[LoRA-TF] Ep20/30 loss=1.0397
[LoRA-TF] Ep21/30 loss=1.0599
[LoRA-TF] Ep22/30 loss=1.0399
[LoRA-TF] Ep23/30 loss=1.0349
[LoRA-TF] Ep24/30 loss=1.0311
[LoRA-TF] Ep25/30 loss=1.0098
[LoRA-TF] Ep26/30 loss=1.0277
[LoRA-TF] Ep27/30 loss=1.0042
[LoRA-TF] Ep28/30 loss=0.9972
[LoRA-TF] Ep29/30 loss=1.0128
[LoRA-TF] Ep30/30 loss=0.9983
Training DANN-Gate (teacher -> 10-class head) ...
[DANN-Gate-TF] Ep1/30 λ_grl=0.000 loss=6.0121
[DANN-Gate-TF] Ep2/30 λ_grl=0.171 loss=5.3571
[DANN-Gate-TF] Ep3/30 λ_grl=0.332 loss=5.1269
[DANN-Gate-TF] Ep4/30 λ_grl=0.476 loss=4.9797
[DANN-Gate-TF] Ep5/30 λ_grl=0.598 loss=4.8661
[DANN-Gate-TF] Ep6/30 λ_grl=0.697 loss=4.8338
[DANN-Gate-TF] Ep7/30 λ_grl=0.776 loss=4.6886
[DANN-Gate-TF] Ep8/30 λ_grl=0.836 loss=4.6480
[DANN-Gate-TF] Ep9/30 λ_grl=0.881 loss=4.6104
[DANN-Gate-TF] Ep10/30 λ_grl=0.914 loss=4.5505
[DANN-Gate-TF] Ep11/30 λ_grl=0.938 loss=4.4658
[DANN-Gate-TF] Ep12/30 λ_grl=0.956 loss=4.4628
[DANN-Gate-TF] Ep13/30 λ_grl=0.969 loss=4.4094
[DANN-Gate-TF] Ep14/30 λ_grl=0.978 loss=4.4108
[DANN-Gate-TF] Ep15/30 λ_grl=0.984 loss=4.3003
[DANN-Gate-TF] Ep16/30 λ_grl=0.989 loss=4.2981
[DANN-Gate-TF] Ep17/30 λ_grl=0.992 loss=4.2655
[DANN-Gate-TF] Ep18/30 λ_grl=0.994 loss=4.2990
[DANN-Gate-TF] Ep19/30 λ_grl=0.996 loss=4.2739
[DANN-Gate-TF] Ep20/30 λ_grl=0.997 loss=4.2122
[DANN-Gate-TF] Ep21/30 λ_grl=0.998 loss=4.1889
[DANN-Gate-TF] Ep22/30 λ_grl=0.999 loss=4.1995
[DANN-Gate-TF] Ep23/30 λ_grl=0.999 loss=4.1941
[DANN-Gate-TF] Ep24/30 λ_grl=0.999 loss=4.1472
[DANN-Gate-TF] Ep25/30 λ_grl=0.999 loss=4.1661
[DANN-Gate-TF] Ep26/30 λ_grl=1.000 loss=4.1175
[DANN-Gate-TF] Ep27/30 λ_grl=1.000 loss=4.1457
[DANN-Gate-TF] Ep28/30 λ_grl=1.000 loss=4.1195
[DANN-Gate-TF] Ep29/30 λ_grl=1.000 loss=4.1296
[DANN-Gate-TF] Ep30/30 λ_grl=1.000 loss=4.0984
[Run 5] LoRA Acc=47.69% | DANN-Gate Acc=47.80%

All done. Final mean/std results saved to: ./results_test10_base/mismatch_tf.json
more_baselines/base_mismatch_tf.py completed successfully.
Starting more_baselines/base_adversarial_tf_test100.py...
Files already downloaded and verified
Files already downloaded and verified

=== Pretraining BigTransformer on adversarial CIFAR-100 (confusion pairs + noise) ===
Files already downloaded and verified
Files already downloaded and verified
[Epoch 1] Loss: 4.5463
[Epoch 2] Loss: 4.4618
[Epoch 3] Loss: 4.4273
[Epoch 4] Loss: 4.4069
[Epoch 5] Loss: 4.4009
[Epoch 6] Loss: 4.3897
[Epoch 7] Loss: 4.3708
[Epoch 8] Loss: 4.3529
[Epoch 9] Loss: 4.3389
[Epoch 10] Loss: 4.3225
[Epoch 11] Loss: 4.3225
[Epoch 12] Loss: 4.2998
[Epoch 13] Loss: 4.2878
[Epoch 14] Loss: 4.2912
[Epoch 15] Loss: 4.2743
[Epoch 16] Loss: 4.2637
[Epoch 17] Loss: 4.2497
[Epoch 18] Loss: 4.2317
[Epoch 19] Loss: 4.2454
[Epoch 20] Loss: 4.2383
[Epoch 21] Loss: 4.2245
[Epoch 22] Loss: 4.2160
[Epoch 23] Loss: 4.2064
[Epoch 24] Loss: 4.1919
[Epoch 25] Loss: 4.1900
[Epoch 26] Loss: 4.1789
[Epoch 27] Loss: 4.1534
[Epoch 28] Loss: 4.1561
[Epoch 29] Loss: 4.1376
[Epoch 30] Loss: 4.1226
[Epoch 31] Loss: 4.1148
[Epoch 32] Loss: 4.1063
[Epoch 33] Loss: 4.1099
[Epoch 34] Loss: 4.0942
[Epoch 35] Loss: 4.0880
[Epoch 36] Loss: 4.0731
[Epoch 37] Loss: 4.0757
[Epoch 38] Loss: 4.0611
[Epoch 39] Loss: 4.0559
[Epoch 40] Loss: 4.0496
[Epoch 41] Loss: 4.0272
[Epoch 42] Loss: 4.0251
[Epoch 43] Loss: 4.0363
[Epoch 44] Loss: 4.0220
[Epoch 45] Loss: 4.0174
[Epoch 46] Loss: 4.0103
[Epoch 47] Loss: 4.0012
[Epoch 48] Loss: 3.9904
[Epoch 49] Loss: 3.9919
[Epoch 50] Loss: 3.9869
[Epoch 51] Loss: 3.9741
[Epoch 52] Loss: 3.9828
[Epoch 53] Loss: 3.9735
[Epoch 54] Loss: 3.9586
[Epoch 55] Loss: 3.9601
[Epoch 56] Loss: 3.9388
[Epoch 57] Loss: 3.9328
[Epoch 58] Loss: 3.9291
[Epoch 59] Loss: 3.9348
[Epoch 60] Loss: 3.9261
Saved pretrained transformer to: ./model_test100/base_adversarial_tf_cifar100.pt
External Transformer Eval: Acc=4.90% | AUC=0.7276 | F1=0.0356 | MinCAcc=0.00%

--- Run 1/5, seed=42 ---
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (Transformer head)...
[LoRA-TF-100] Ep1/30 loss=4.2883
[LoRA-TF-100] Ep2/30 loss=4.0893
[LoRA-TF-100] Ep3/30 loss=3.9933
[LoRA-TF-100] Ep4/30 loss=3.9234
[LoRA-TF-100] Ep5/30 loss=3.8828
[LoRA-TF-100] Ep6/30 loss=3.8552
[LoRA-TF-100] Ep7/30 loss=3.8334
[LoRA-TF-100] Ep8/30 loss=3.7966
[LoRA-TF-100] Ep9/30 loss=3.7798
[LoRA-TF-100] Ep10/30 loss=3.7587
[LoRA-TF-100] Ep11/30 loss=3.7355
[LoRA-TF-100] Ep12/30 loss=3.7269
[LoRA-TF-100] Ep13/30 loss=3.7178
[LoRA-TF-100] Ep14/30 loss=3.7026
[LoRA-TF-100] Ep15/30 loss=3.6912
[LoRA-TF-100] Ep16/30 loss=3.6717
[LoRA-TF-100] Ep17/30 loss=3.6744
[LoRA-TF-100] Ep18/30 loss=3.6687
[LoRA-TF-100] Ep19/30 loss=3.6662
[LoRA-TF-100] Ep20/30 loss=3.6606
[LoRA-TF-100] Ep21/30 loss=3.6414
[LoRA-TF-100] Ep22/30 loss=3.6328
[LoRA-TF-100] Ep23/30 loss=3.6417
[LoRA-TF-100] Ep24/30 loss=3.6279
[LoRA-TF-100] Ep25/30 loss=3.6250
[LoRA-TF-100] Ep26/30 loss=3.6152
[LoRA-TF-100] Ep27/30 loss=3.6251
[LoRA-TF-100] Ep28/30 loss=3.6153
[LoRA-TF-100] Ep29/30 loss=3.6133
[LoRA-TF-100] Ep30/30 loss=3.5994
Training DANN-Gate (Transformer head)...
[DANN-Gate-TF-100] Ep1/30 λ_grl=0.000 loss=10.9912
[DANN-Gate-TF-100] Ep2/30 λ_grl=0.171 loss=10.3591
[DANN-Gate-TF-100] Ep3/30 λ_grl=0.332 loss=10.2124
[DANN-Gate-TF-100] Ep4/30 λ_grl=0.476 loss=10.0536
[DANN-Gate-TF-100] Ep5/30 λ_grl=0.598 loss=9.8941
[DANN-Gate-TF-100] Ep6/30 λ_grl=0.697 loss=9.8328
[DANN-Gate-TF-100] Ep7/30 λ_grl=0.776 loss=9.8071
[DANN-Gate-TF-100] Ep8/30 λ_grl=0.836 loss=9.7137
[DANN-Gate-TF-100] Ep9/30 λ_grl=0.881 loss=9.7419
[DANN-Gate-TF-100] Ep10/30 λ_grl=0.914 loss=9.6411
[DANN-Gate-TF-100] Ep11/30 λ_grl=0.938 loss=9.6066
[DANN-Gate-TF-100] Ep12/30 λ_grl=0.956 loss=9.5680
[DANN-Gate-TF-100] Ep13/30 λ_grl=0.969 loss=9.5738
[DANN-Gate-TF-100] Ep14/30 λ_grl=0.978 loss=9.4640
[DANN-Gate-TF-100] Ep15/30 λ_grl=0.984 loss=9.5220
[DANN-Gate-TF-100] Ep16/30 λ_grl=0.989 loss=9.4999
[DANN-Gate-TF-100] Ep17/30 λ_grl=0.992 loss=9.4579
[DANN-Gate-TF-100] Ep18/30 λ_grl=0.994 loss=9.3988
[DANN-Gate-TF-100] Ep19/30 λ_grl=0.996 loss=9.4614
[DANN-Gate-TF-100] Ep20/30 λ_grl=0.997 loss=9.4182
[DANN-Gate-TF-100] Ep21/30 λ_grl=0.998 loss=9.4143
[DANN-Gate-TF-100] Ep22/30 λ_grl=0.999 loss=9.3565
[DANN-Gate-TF-100] Ep23/30 λ_grl=0.999 loss=9.3530
[DANN-Gate-TF-100] Ep24/30 λ_grl=0.999 loss=9.3596
[DANN-Gate-TF-100] Ep25/30 λ_grl=0.999 loss=9.4022
[DANN-Gate-TF-100] Ep26/30 λ_grl=1.000 loss=9.3557
[DANN-Gate-TF-100] Ep27/30 λ_grl=1.000 loss=9.3299
[DANN-Gate-TF-100] Ep28/30 λ_grl=1.000 loss=9.3243
[DANN-Gate-TF-100] Ep29/30 λ_grl=1.000 loss=9.3012
[DANN-Gate-TF-100] Ep30/30 λ_grl=1.000 loss=9.2883
[Run 1] LoRA Acc=11.06% | DANN-Gate Acc=11.26%

--- Run 2/5, seed=43 ---
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (Transformer head)...
[LoRA-TF-100] Ep1/30 loss=4.2546
[LoRA-TF-100] Ep2/30 loss=4.0577
[LoRA-TF-100] Ep3/30 loss=3.9480
[LoRA-TF-100] Ep4/30 loss=3.9070
[LoRA-TF-100] Ep5/30 loss=3.8411
[LoRA-TF-100] Ep6/30 loss=3.8086
[LoRA-TF-100] Ep7/30 loss=3.7781
[LoRA-TF-100] Ep8/30 loss=3.7612
[LoRA-TF-100] Ep9/30 loss=3.7217
[LoRA-TF-100] Ep10/30 loss=3.7214
[LoRA-TF-100] Ep11/30 loss=3.7034
[LoRA-TF-100] Ep12/30 loss=3.6809
[LoRA-TF-100] Ep13/30 loss=3.6797
[LoRA-TF-100] Ep14/30 loss=3.6591
[LoRA-TF-100] Ep15/30 loss=3.6457
[LoRA-TF-100] Ep16/30 loss=3.6451
[LoRA-TF-100] Ep17/30 loss=3.6317
[LoRA-TF-100] Ep18/30 loss=3.6285
[LoRA-TF-100] Ep19/30 loss=3.6172
[LoRA-TF-100] Ep20/30 loss=3.6082
[LoRA-TF-100] Ep21/30 loss=3.5980
[LoRA-TF-100] Ep22/30 loss=3.5957
[LoRA-TF-100] Ep23/30 loss=3.5906
[LoRA-TF-100] Ep24/30 loss=3.5915
[LoRA-TF-100] Ep25/30 loss=3.5840
[LoRA-TF-100] Ep26/30 loss=3.5727
[LoRA-TF-100] Ep27/30 loss=3.5701
[LoRA-TF-100] Ep28/30 loss=3.5666
[LoRA-TF-100] Ep29/30 loss=3.5554
[LoRA-TF-100] Ep30/30 loss=3.5557
Training DANN-Gate (Transformer head)...
[DANN-Gate-TF-100] Ep1/30 λ_grl=0.000 loss=10.7239
[DANN-Gate-TF-100] Ep2/30 λ_grl=0.171 loss=10.2553
[DANN-Gate-TF-100] Ep3/30 λ_grl=0.332 loss=10.0814
[DANN-Gate-TF-100] Ep4/30 λ_grl=0.476 loss=9.9699
[DANN-Gate-TF-100] Ep5/30 λ_grl=0.598 loss=9.8167
[DANN-Gate-TF-100] Ep6/30 λ_grl=0.697 loss=9.7395
[DANN-Gate-TF-100] Ep7/30 λ_grl=0.776 loss=9.7144
[DANN-Gate-TF-100] Ep8/30 λ_grl=0.836 loss=9.6461
[DANN-Gate-TF-100] Ep9/30 λ_grl=0.881 loss=9.5617
[DANN-Gate-TF-100] Ep10/30 λ_grl=0.914 loss=9.5953
[DANN-Gate-TF-100] Ep11/30 λ_grl=0.938 loss=9.5013
[DANN-Gate-TF-100] Ep12/30 λ_grl=0.956 loss=9.5136
[DANN-Gate-TF-100] Ep13/30 λ_grl=0.969 loss=9.4684
[DANN-Gate-TF-100] Ep14/30 λ_grl=0.978 loss=9.4996
[DANN-Gate-TF-100] Ep15/30 λ_grl=0.984 loss=9.4720
[DANN-Gate-TF-100] Ep16/30 λ_grl=0.989 loss=9.3979
[DANN-Gate-TF-100] Ep17/30 λ_grl=0.992 loss=9.3899
[DANN-Gate-TF-100] Ep18/30 λ_grl=0.994 loss=9.3949
[DANN-Gate-TF-100] Ep19/30 λ_grl=0.996 loss=9.3441
[DANN-Gate-TF-100] Ep20/30 λ_grl=0.997 loss=9.3267
[DANN-Gate-TF-100] Ep21/30 λ_grl=0.998 loss=9.3099
[DANN-Gate-TF-100] Ep22/30 λ_grl=0.999 loss=9.3082
[DANN-Gate-TF-100] Ep23/30 λ_grl=0.999 loss=9.2830
[DANN-Gate-TF-100] Ep24/30 λ_grl=0.999 loss=9.2777
[DANN-Gate-TF-100] Ep25/30 λ_grl=0.999 loss=9.2663
[DANN-Gate-TF-100] Ep26/30 λ_grl=1.000 loss=9.2497
[DANN-Gate-TF-100] Ep27/30 λ_grl=1.000 loss=9.2565
[DANN-Gate-TF-100] Ep28/30 λ_grl=1.000 loss=9.2273
[DANN-Gate-TF-100] Ep29/30 λ_grl=1.000 loss=9.2320
[DANN-Gate-TF-100] Ep30/30 λ_grl=1.000 loss=9.2276
[Run 2] LoRA Acc=11.53% | DANN-Gate Acc=11.83%

--- Run 3/5, seed=44 ---
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (Transformer head)...
[LoRA-TF-100] Ep1/30 loss=4.2953
[LoRA-TF-100] Ep2/30 loss=4.0901
[LoRA-TF-100] Ep3/30 loss=3.9940
[LoRA-TF-100] Ep4/30 loss=3.9401
[LoRA-TF-100] Ep5/30 loss=3.9077
[LoRA-TF-100] Ep6/30 loss=3.8635
[LoRA-TF-100] Ep7/30 loss=3.8324
[LoRA-TF-100] Ep8/30 loss=3.8043
[LoRA-TF-100] Ep9/30 loss=3.7940
[LoRA-TF-100] Ep10/30 loss=3.7635
[LoRA-TF-100] Ep11/30 loss=3.7504
[LoRA-TF-100] Ep12/30 loss=3.7275
[LoRA-TF-100] Ep13/30 loss=3.7236
[LoRA-TF-100] Ep14/30 loss=3.7132
[LoRA-TF-100] Ep15/30 loss=3.7075
[LoRA-TF-100] Ep16/30 loss=3.6958
[LoRA-TF-100] Ep17/30 loss=3.6826
[LoRA-TF-100] Ep18/30 loss=3.6753
[LoRA-TF-100] Ep19/30 loss=3.6657
[LoRA-TF-100] Ep20/30 loss=3.6620
[LoRA-TF-100] Ep21/30 loss=3.6566
[LoRA-TF-100] Ep22/30 loss=3.6488
[LoRA-TF-100] Ep23/30 loss=3.6433
[LoRA-TF-100] Ep24/30 loss=3.6301
[LoRA-TF-100] Ep25/30 loss=3.6175
[LoRA-TF-100] Ep26/30 loss=3.6257
[LoRA-TF-100] Ep27/30 loss=3.6124
[LoRA-TF-100] Ep28/30 loss=3.6198
[LoRA-TF-100] Ep29/30 loss=3.5968
[LoRA-TF-100] Ep30/30 loss=3.6041
Training DANN-Gate (Transformer head)...
[DANN-Gate-TF-100] Ep1/30 λ_grl=0.000 loss=10.6313
[DANN-Gate-TF-100] Ep2/30 λ_grl=0.171 loss=10.1859
[DANN-Gate-TF-100] Ep3/30 λ_grl=0.332 loss=10.1397
[DANN-Gate-TF-100] Ep4/30 λ_grl=0.476 loss=9.9727
[DANN-Gate-TF-100] Ep5/30 λ_grl=0.598 loss=9.9446
[DANN-Gate-TF-100] Ep6/30 λ_grl=0.697 loss=9.8016
[DANN-Gate-TF-100] Ep7/30 λ_grl=0.776 loss=9.7525
[DANN-Gate-TF-100] Ep8/30 λ_grl=0.836 loss=9.6906
[DANN-Gate-TF-100] Ep9/30 λ_grl=0.881 loss=9.6838
[DANN-Gate-TF-100] Ep10/30 λ_grl=0.914 loss=9.6640
[DANN-Gate-TF-100] Ep11/30 λ_grl=0.938 loss=9.6761
[DANN-Gate-TF-100] Ep12/30 λ_grl=0.956 loss=9.5912
[DANN-Gate-TF-100] Ep13/30 λ_grl=0.969 loss=9.5688
[DANN-Gate-TF-100] Ep14/30 λ_grl=0.978 loss=9.4922
[DANN-Gate-TF-100] Ep15/30 λ_grl=0.984 loss=9.5317
[DANN-Gate-TF-100] Ep16/30 λ_grl=0.989 loss=9.5012
[DANN-Gate-TF-100] Ep17/30 λ_grl=0.992 loss=9.4411
[DANN-Gate-TF-100] Ep18/30 λ_grl=0.994 loss=9.4446
[DANN-Gate-TF-100] Ep19/30 λ_grl=0.996 loss=9.4315
[DANN-Gate-TF-100] Ep20/30 λ_grl=0.997 loss=9.4334
[DANN-Gate-TF-100] Ep21/30 λ_grl=0.998 loss=9.3950
[DANN-Gate-TF-100] Ep22/30 λ_grl=0.999 loss=9.3899
[DANN-Gate-TF-100] Ep23/30 λ_grl=0.999 loss=9.3764
[DANN-Gate-TF-100] Ep24/30 λ_grl=0.999 loss=9.3786
[DANN-Gate-TF-100] Ep25/30 λ_grl=0.999 loss=9.3343
[DANN-Gate-TF-100] Ep26/30 λ_grl=1.000 loss=9.3727
[DANN-Gate-TF-100] Ep27/30 λ_grl=1.000 loss=9.3414
[DANN-Gate-TF-100] Ep28/30 λ_grl=1.000 loss=9.3344
[DANN-Gate-TF-100] Ep29/30 λ_grl=1.000 loss=9.3549
[DANN-Gate-TF-100] Ep30/30 λ_grl=1.000 loss=9.3029
[Run 3] LoRA Acc=11.48% | DANN-Gate Acc=11.54%

--- Run 4/5, seed=45 ---
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (Transformer head)...
[LoRA-TF-100] Ep1/30 loss=4.2682
[LoRA-TF-100] Ep2/30 loss=4.0742
[LoRA-TF-100] Ep3/30 loss=3.9681
[LoRA-TF-100] Ep4/30 loss=3.9298
[LoRA-TF-100] Ep5/30 loss=3.8904
[LoRA-TF-100] Ep6/30 loss=3.8531
[LoRA-TF-100] Ep7/30 loss=3.8259
[LoRA-TF-100] Ep8/30 loss=3.7923
[LoRA-TF-100] Ep9/30 loss=3.7899
[LoRA-TF-100] Ep10/30 loss=3.7703
[LoRA-TF-100] Ep11/30 loss=3.7472
[LoRA-TF-100] Ep12/30 loss=3.7285
[LoRA-TF-100] Ep13/30 loss=3.7163
[LoRA-TF-100] Ep14/30 loss=3.7289
[LoRA-TF-100] Ep15/30 loss=3.6938
[LoRA-TF-100] Ep16/30 loss=3.6912
[LoRA-TF-100] Ep17/30 loss=3.6831
[LoRA-TF-100] Ep18/30 loss=3.6752
[LoRA-TF-100] Ep19/30 loss=3.6638
[LoRA-TF-100] Ep20/30 loss=3.6498
[LoRA-TF-100] Ep21/30 loss=3.6468
[LoRA-TF-100] Ep22/30 loss=3.6531
[LoRA-TF-100] Ep23/30 loss=3.6349
[LoRA-TF-100] Ep24/30 loss=3.6362
[LoRA-TF-100] Ep25/30 loss=3.6313
[LoRA-TF-100] Ep26/30 loss=3.6260
[LoRA-TF-100] Ep27/30 loss=3.6133
[LoRA-TF-100] Ep28/30 loss=3.6173
[LoRA-TF-100] Ep29/30 loss=3.6068
[LoRA-TF-100] Ep30/30 loss=3.5994
Training DANN-Gate (Transformer head)...
[DANN-Gate-TF-100] Ep1/30 λ_grl=0.000 loss=10.4650
[DANN-Gate-TF-100] Ep2/30 λ_grl=0.171 loss=10.1527
[DANN-Gate-TF-100] Ep3/30 λ_grl=0.332 loss=10.0568
[DANN-Gate-TF-100] Ep4/30 λ_grl=0.476 loss=10.0108
[DANN-Gate-TF-100] Ep5/30 λ_grl=0.598 loss=9.8858
[DANN-Gate-TF-100] Ep6/30 λ_grl=0.697 loss=9.8361
[DANN-Gate-TF-100] Ep7/30 λ_grl=0.776 loss=9.8204
[DANN-Gate-TF-100] Ep8/30 λ_grl=0.836 loss=9.7153
[DANN-Gate-TF-100] Ep9/30 λ_grl=0.881 loss=9.6397
[DANN-Gate-TF-100] Ep10/30 λ_grl=0.914 loss=9.6393
[DANN-Gate-TF-100] Ep11/30 λ_grl=0.938 loss=9.5438
[DANN-Gate-TF-100] Ep12/30 λ_grl=0.956 loss=9.5484
[DANN-Gate-TF-100] Ep13/30 λ_grl=0.969 loss=9.5862
[DANN-Gate-TF-100] Ep14/30 λ_grl=0.978 loss=9.5012
[DANN-Gate-TF-100] Ep15/30 λ_grl=0.984 loss=9.5300
[DANN-Gate-TF-100] Ep16/30 λ_grl=0.989 loss=9.4908
[DANN-Gate-TF-100] Ep17/30 λ_grl=0.992 loss=9.4468
[DANN-Gate-TF-100] Ep18/30 λ_grl=0.994 loss=9.4081
[DANN-Gate-TF-100] Ep19/30 λ_grl=0.996 loss=9.4467
[DANN-Gate-TF-100] Ep20/30 λ_grl=0.997 loss=9.4309
[DANN-Gate-TF-100] Ep21/30 λ_grl=0.998 loss=9.3857
[DANN-Gate-TF-100] Ep22/30 λ_grl=0.999 loss=9.3623
[DANN-Gate-TF-100] Ep23/30 λ_grl=0.999 loss=9.3685
[DANN-Gate-TF-100] Ep24/30 λ_grl=0.999 loss=9.3231
[DANN-Gate-TF-100] Ep25/30 λ_grl=0.999 loss=9.3377
[DANN-Gate-TF-100] Ep26/30 λ_grl=1.000 loss=9.3232
[DANN-Gate-TF-100] Ep27/30 λ_grl=1.000 loss=9.3156
[DANN-Gate-TF-100] Ep28/30 λ_grl=1.000 loss=9.3014
[DANN-Gate-TF-100] Ep29/30 λ_grl=1.000 loss=9.2504
[DANN-Gate-TF-100] Ep30/30 λ_grl=1.000 loss=9.2638
[Run 4] LoRA Acc=11.25% | DANN-Gate Acc=11.31%

--- Run 5/5, seed=46 ---
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (Transformer head)...
[LoRA-TF-100] Ep1/30 loss=4.2581
[LoRA-TF-100] Ep2/30 loss=4.0451
[LoRA-TF-100] Ep3/30 loss=3.9649
[LoRA-TF-100] Ep4/30 loss=3.9208
[LoRA-TF-100] Ep5/30 loss=3.8879
[LoRA-TF-100] Ep6/30 loss=3.8582
[LoRA-TF-100] Ep7/30 loss=3.8213
[LoRA-TF-100] Ep8/30 loss=3.8004
[LoRA-TF-100] Ep9/30 loss=3.7698
[LoRA-TF-100] Ep10/30 loss=3.7567
[LoRA-TF-100] Ep11/30 loss=3.7371
[LoRA-TF-100] Ep12/30 loss=3.7328
[LoRA-TF-100] Ep13/30 loss=3.7187
[LoRA-TF-100] Ep14/30 loss=3.7036
[LoRA-TF-100] Ep15/30 loss=3.6913
[LoRA-TF-100] Ep16/30 loss=3.6809
[LoRA-TF-100] Ep17/30 loss=3.6851
[LoRA-TF-100] Ep18/30 loss=3.6749
[LoRA-TF-100] Ep19/30 loss=3.6527
[LoRA-TF-100] Ep20/30 loss=3.6596
[LoRA-TF-100] Ep21/30 loss=3.6533
[LoRA-TF-100] Ep22/30 loss=3.6463
[LoRA-TF-100] Ep23/30 loss=3.6389
[LoRA-TF-100] Ep24/30 loss=3.6219
[LoRA-TF-100] Ep25/30 loss=3.6273
[LoRA-TF-100] Ep26/30 loss=3.6121
[LoRA-TF-100] Ep27/30 loss=3.6119
[LoRA-TF-100] Ep28/30 loss=3.6072
[LoRA-TF-100] Ep29/30 loss=3.6149
[LoRA-TF-100] Ep30/30 loss=3.6115
Training DANN-Gate (Transformer head)...
[DANN-Gate-TF-100] Ep1/30 λ_grl=0.000 loss=10.4159
[DANN-Gate-TF-100] Ep2/30 λ_grl=0.171 loss=10.2287
[DANN-Gate-TF-100] Ep3/30 λ_grl=0.332 loss=10.1398
[DANN-Gate-TF-100] Ep4/30 λ_grl=0.476 loss=9.9648
[DANN-Gate-TF-100] Ep5/30 λ_grl=0.598 loss=9.9506
[DANN-Gate-TF-100] Ep6/30 λ_grl=0.697 loss=9.8899
[DANN-Gate-TF-100] Ep7/30 λ_grl=0.776 loss=9.7323
[DANN-Gate-TF-100] Ep8/30 λ_grl=0.836 loss=9.7701
[DANN-Gate-TF-100] Ep9/30 λ_grl=0.881 loss=9.6544
[DANN-Gate-TF-100] Ep10/30 λ_grl=0.914 loss=9.6570
[DANN-Gate-TF-100] Ep11/30 λ_grl=0.938 loss=9.6243
[DANN-Gate-TF-100] Ep12/30 λ_grl=0.956 loss=9.5686
[DANN-Gate-TF-100] Ep13/30 λ_grl=0.969 loss=9.4951
[DANN-Gate-TF-100] Ep14/30 λ_grl=0.978 loss=9.5048
[DANN-Gate-TF-100] Ep15/30 λ_grl=0.984 loss=9.5219
[DANN-Gate-TF-100] Ep16/30 λ_grl=0.989 loss=9.4869
[DANN-Gate-TF-100] Ep17/30 λ_grl=0.992 loss=9.4523
[DANN-Gate-TF-100] Ep18/30 λ_grl=0.994 loss=9.4426
[DANN-Gate-TF-100] Ep19/30 λ_grl=0.996 loss=9.4085
[DANN-Gate-TF-100] Ep20/30 λ_grl=0.997 loss=9.4127
[DANN-Gate-TF-100] Ep21/30 λ_grl=0.998 loss=9.3495
[DANN-Gate-TF-100] Ep22/30 λ_grl=0.999 loss=9.3968
[DANN-Gate-TF-100] Ep23/30 λ_grl=0.999 loss=9.3608
[DANN-Gate-TF-100] Ep24/30 λ_grl=0.999 loss=9.3542
[DANN-Gate-TF-100] Ep25/30 λ_grl=0.999 loss=9.3295
[DANN-Gate-TF-100] Ep26/30 λ_grl=1.000 loss=9.3343
[DANN-Gate-TF-100] Ep27/30 λ_grl=1.000 loss=9.3345
[DANN-Gate-TF-100] Ep28/30 λ_grl=1.000 loss=9.2942
[DANN-Gate-TF-100] Ep29/30 λ_grl=1.000 loss=9.2621
[DANN-Gate-TF-100] Ep30/30 λ_grl=1.000 loss=9.2560
[Run 5] LoRA Acc=11.50% | DANN-Gate Acc=11.34%

All done. Final mean/std results saved to: ./results_test100_base/adversarial_tf_cifar100_confusion.json
more_baselines/base_adversarial_tf_test100.py completed successfully.
Starting more_baselines/base_imb_tf_test100.py...
Files already downloaded and verified
Files already downloaded and verified
[Epoch 1] Loss: 3.9135
[Epoch 2] Loss: 3.6684
[Epoch 3] Loss: 3.5221
[Epoch 4] Loss: 3.3937
[Epoch 5] Loss: 3.2799
[Epoch 6] Loss: 3.2106
[Epoch 7] Loss: 3.1029
[Epoch 8] Loss: 3.0206
[Epoch 9] Loss: 2.8898
[Epoch 10] Loss: 2.8197
[Epoch 11] Loss: 2.7259
[Epoch 12] Loss: 2.6363
[Epoch 13] Loss: 2.5521
[Epoch 14] Loss: 2.4837
[Epoch 15] Loss: 2.3868
[Epoch 16] Loss: 2.2899
[Epoch 17] Loss: 2.1957
[Epoch 18] Loss: 2.1552
[Epoch 19] Loss: 2.0792
[Epoch 20] Loss: 1.9816
[Epoch 21] Loss: 1.9062
[Epoch 22] Loss: 1.8150
[Epoch 23] Loss: 1.7638
[Epoch 24] Loss: 1.6490
[Epoch 25] Loss: 1.6217
[Epoch 26] Loss: 1.5040
[Epoch 27] Loss: 1.4422
[Epoch 28] Loss: 1.3606
[Epoch 29] Loss: 1.2751
[Epoch 30] Loss: 1.1650
[Epoch 31] Loss: 1.1185
[Epoch 32] Loss: 1.0663
[Epoch 33] Loss: 1.0031
[Epoch 34] Loss: 0.9689
[Epoch 35] Loss: 0.8707
[Epoch 36] Loss: 0.8322
[Epoch 37] Loss: 0.7559
[Epoch 38] Loss: 0.7211
[Epoch 39] Loss: 0.6506
[Epoch 40] Loss: 0.5990
[Epoch 41] Loss: 0.5715
[Epoch 42] Loss: 0.5256
[Epoch 43] Loss: 0.4919
[Epoch 44] Loss: 0.4656
[Epoch 45] Loss: 0.4592
[Epoch 46] Loss: 0.4274
[Epoch 47] Loss: 0.4477
[Epoch 48] Loss: 0.4528
[Epoch 49] Loss: 0.4008
[Epoch 50] Loss: 0.3397
[Epoch 51] Loss: 0.3530
[Epoch 52] Loss: 0.3768
[Epoch 53] Loss: 0.2896
[Epoch 54] Loss: 0.2846
[Epoch 55] Loss: 0.3053
[Epoch 56] Loss: 0.2668
[Epoch 57] Loss: 0.3067
[Epoch 58] Loss: 0.2685
[Epoch 59] Loss: 0.3088
[Epoch 60] Loss: 0.2352
Trained and saved external transformer to: ./model_test100/base_imbalance_tf_cifar100.pt
External Transformer (Pretrained on Imbalanced Data): Acc=20.99% | AUC=0.7766 | F1=0.1661 | MinCAcc=0.00%

=== Run 1/5, seed=42 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (Transformer head)...
[LoRA-TF-100] Ep1/30 loss=6.2399
[LoRA-TF-100] Ep2/30 loss=4.2883
[LoRA-TF-100] Ep3/30 loss=4.1849
[LoRA-TF-100] Ep4/30 loss=4.1579
[LoRA-TF-100] Ep5/30 loss=4.0439
[LoRA-TF-100] Ep6/30 loss=3.9960
[LoRA-TF-100] Ep7/30 loss=3.9280
[LoRA-TF-100] Ep8/30 loss=3.8231
[LoRA-TF-100] Ep9/30 loss=3.7094
[LoRA-TF-100] Ep10/30 loss=3.6269
[LoRA-TF-100] Ep11/30 loss=3.6009
[LoRA-TF-100] Ep12/30 loss=3.5415
[LoRA-TF-100] Ep13/30 loss=3.4213
[LoRA-TF-100] Ep14/30 loss=3.3726
[LoRA-TF-100] Ep15/30 loss=3.3966
[LoRA-TF-100] Ep16/30 loss=3.2929
[LoRA-TF-100] Ep17/30 loss=3.2746
[LoRA-TF-100] Ep18/30 loss=3.2235
[LoRA-TF-100] Ep19/30 loss=3.2105
[LoRA-TF-100] Ep20/30 loss=3.2161
[LoRA-TF-100] Ep21/30 loss=3.1163
[LoRA-TF-100] Ep22/30 loss=3.1144
[LoRA-TF-100] Ep23/30 loss=3.1115
[LoRA-TF-100] Ep24/30 loss=3.1027
[LoRA-TF-100] Ep25/30 loss=3.0304
[LoRA-TF-100] Ep26/30 loss=3.0353
[LoRA-TF-100] Ep27/30 loss=3.0040
[LoRA-TF-100] Ep28/30 loss=2.9754
[LoRA-TF-100] Ep29/30 loss=2.9623
[LoRA-TF-100] Ep30/30 loss=2.9658
Training DANN-Gate (Transformer head)...
[DANN-Gate-TF-100] Ep1/30 λ_grl=0.000 loss=16.0414
[DANN-Gate-TF-100] Ep2/30 λ_grl=0.171 loss=11.6563
[DANN-Gate-TF-100] Ep3/30 λ_grl=0.332 loss=10.8048
[DANN-Gate-TF-100] Ep4/30 λ_grl=0.476 loss=10.5030
[DANN-Gate-TF-100] Ep5/30 λ_grl=0.598 loss=10.2931
[DANN-Gate-TF-100] Ep6/30 λ_grl=0.697 loss=10.1297
[DANN-Gate-TF-100] Ep7/30 λ_grl=0.776 loss=9.8743
[DANN-Gate-TF-100] Ep8/30 λ_grl=0.836 loss=9.6982
[DANN-Gate-TF-100] Ep9/30 λ_grl=0.881 loss=9.6641
[DANN-Gate-TF-100] Ep10/30 λ_grl=0.914 loss=9.4552
[DANN-Gate-TF-100] Ep11/30 λ_grl=0.938 loss=9.3478
[DANN-Gate-TF-100] Ep12/30 λ_grl=0.956 loss=9.3037
[DANN-Gate-TF-100] Ep13/30 λ_grl=0.969 loss=9.1230
[DANN-Gate-TF-100] Ep14/30 λ_grl=0.978 loss=8.9225
[DANN-Gate-TF-100] Ep15/30 λ_grl=0.984 loss=8.9087
[DANN-Gate-TF-100] Ep16/30 λ_grl=0.989 loss=8.8854
[DANN-Gate-TF-100] Ep17/30 λ_grl=0.992 loss=8.8053
[DANN-Gate-TF-100] Ep18/30 λ_grl=0.994 loss=8.6614
[DANN-Gate-TF-100] Ep19/30 λ_grl=0.996 loss=8.5941
[DANN-Gate-TF-100] Ep20/30 λ_grl=0.997 loss=8.4779
[DANN-Gate-TF-100] Ep21/30 λ_grl=0.998 loss=8.4832
[DANN-Gate-TF-100] Ep22/30 λ_grl=0.999 loss=8.3698
[DANN-Gate-TF-100] Ep23/30 λ_grl=0.999 loss=8.3768
[DANN-Gate-TF-100] Ep24/30 λ_grl=0.999 loss=8.3066
[DANN-Gate-TF-100] Ep25/30 λ_grl=0.999 loss=8.2425
[DANN-Gate-TF-100] Ep26/30 λ_grl=1.000 loss=8.2207
[DANN-Gate-TF-100] Ep27/30 λ_grl=1.000 loss=8.1972
[DANN-Gate-TF-100] Ep28/30 λ_grl=1.000 loss=8.1323
[DANN-Gate-TF-100] Ep29/30 λ_grl=1.000 loss=8.0742
[DANN-Gate-TF-100] Ep30/30 λ_grl=1.000 loss=7.9954
[Run 1] LoRA Acc=23.44% | DANN-Gate Acc=23.62%

=== Run 2/5, seed=43 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (Transformer head)...
[LoRA-TF-100] Ep1/30 loss=6.3316
[LoRA-TF-100] Ep2/30 loss=4.3145
[LoRA-TF-100] Ep3/30 loss=4.2931
[LoRA-TF-100] Ep4/30 loss=4.1730
[LoRA-TF-100] Ep5/30 loss=4.1243
[LoRA-TF-100] Ep6/30 loss=4.0632
[LoRA-TF-100] Ep7/30 loss=3.9520
[LoRA-TF-100] Ep8/30 loss=3.8933
[LoRA-TF-100] Ep9/30 loss=3.7747
[LoRA-TF-100] Ep10/30 loss=3.6856
[LoRA-TF-100] Ep11/30 loss=3.5889
[LoRA-TF-100] Ep12/30 loss=3.5210
[LoRA-TF-100] Ep13/30 loss=3.4782
[LoRA-TF-100] Ep14/30 loss=3.4466
[LoRA-TF-100] Ep15/30 loss=3.4378
[LoRA-TF-100] Ep16/30 loss=3.3709
[LoRA-TF-100] Ep17/30 loss=3.3450
[LoRA-TF-100] Ep18/30 loss=3.2582
[LoRA-TF-100] Ep19/30 loss=3.2220
[LoRA-TF-100] Ep20/30 loss=3.2289
[LoRA-TF-100] Ep21/30 loss=3.1513
[LoRA-TF-100] Ep22/30 loss=3.1207
[LoRA-TF-100] Ep23/30 loss=3.1141
[LoRA-TF-100] Ep24/30 loss=3.1057
[LoRA-TF-100] Ep25/30 loss=3.1010
[LoRA-TF-100] Ep26/30 loss=3.0405
[LoRA-TF-100] Ep27/30 loss=3.0681
[LoRA-TF-100] Ep28/30 loss=3.0175
[LoRA-TF-100] Ep29/30 loss=3.0333
[LoRA-TF-100] Ep30/30 loss=2.9640
Training DANN-Gate (Transformer head)...
[DANN-Gate-TF-100] Ep1/30 λ_grl=0.000 loss=16.6221
[DANN-Gate-TF-100] Ep2/30 λ_grl=0.171 loss=11.7941
[DANN-Gate-TF-100] Ep3/30 λ_grl=0.332 loss=10.8523
[DANN-Gate-TF-100] Ep4/30 λ_grl=0.476 loss=10.6163
[DANN-Gate-TF-100] Ep5/30 λ_grl=0.598 loss=10.4005
[DANN-Gate-TF-100] Ep6/30 λ_grl=0.697 loss=10.1067
[DANN-Gate-TF-100] Ep7/30 λ_grl=0.776 loss=10.0756
[DANN-Gate-TF-100] Ep8/30 λ_grl=0.836 loss=9.8496
[DANN-Gate-TF-100] Ep9/30 λ_grl=0.881 loss=9.8532
[DANN-Gate-TF-100] Ep10/30 λ_grl=0.914 loss=9.6589
[DANN-Gate-TF-100] Ep11/30 λ_grl=0.938 loss=9.5322
[DANN-Gate-TF-100] Ep12/30 λ_grl=0.956 loss=9.2867
[DANN-Gate-TF-100] Ep13/30 λ_grl=0.969 loss=9.2334
[DANN-Gate-TF-100] Ep14/30 λ_grl=0.978 loss=9.1113
[DANN-Gate-TF-100] Ep15/30 λ_grl=0.984 loss=8.9781
[DANN-Gate-TF-100] Ep16/30 λ_grl=0.989 loss=8.8707
[DANN-Gate-TF-100] Ep17/30 λ_grl=0.992 loss=8.8711
[DANN-Gate-TF-100] Ep18/30 λ_grl=0.994 loss=8.7815
[DANN-Gate-TF-100] Ep19/30 λ_grl=0.996 loss=8.6498
[DANN-Gate-TF-100] Ep20/30 λ_grl=0.997 loss=8.5110
[DANN-Gate-TF-100] Ep21/30 λ_grl=0.998 loss=8.5782
[DANN-Gate-TF-100] Ep22/30 λ_grl=0.999 loss=8.4338
[DANN-Gate-TF-100] Ep23/30 λ_grl=0.999 loss=8.3889
[DANN-Gate-TF-100] Ep24/30 λ_grl=0.999 loss=8.3652
[DANN-Gate-TF-100] Ep25/30 λ_grl=0.999 loss=8.2308
[DANN-Gate-TF-100] Ep26/30 λ_grl=1.000 loss=8.2922
[DANN-Gate-TF-100] Ep27/30 λ_grl=1.000 loss=8.3006
[DANN-Gate-TF-100] Ep28/30 λ_grl=1.000 loss=8.1934
[DANN-Gate-TF-100] Ep29/30 λ_grl=1.000 loss=8.1783
[DANN-Gate-TF-100] Ep30/30 λ_grl=1.000 loss=8.1889
[Run 2] LoRA Acc=23.58% | DANN-Gate Acc=23.32%

=== Run 3/5, seed=44 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (Transformer head)...
[LoRA-TF-100] Ep1/30 loss=6.2255
[LoRA-TF-100] Ep2/30 loss=4.2784
[LoRA-TF-100] Ep3/30 loss=4.1506
[LoRA-TF-100] Ep4/30 loss=4.0752
[LoRA-TF-100] Ep5/30 loss=4.0040
[LoRA-TF-100] Ep6/30 loss=3.9303
[LoRA-TF-100] Ep7/30 loss=3.8838
[LoRA-TF-100] Ep8/30 loss=3.8174
[LoRA-TF-100] Ep9/30 loss=3.7293
[LoRA-TF-100] Ep10/30 loss=3.6684
[LoRA-TF-100] Ep11/30 loss=3.6194
[LoRA-TF-100] Ep12/30 loss=3.5232
[LoRA-TF-100] Ep13/30 loss=3.4782
[LoRA-TF-100] Ep14/30 loss=3.3849
[LoRA-TF-100] Ep15/30 loss=3.3926
[LoRA-TF-100] Ep16/30 loss=3.3258
[LoRA-TF-100] Ep17/30 loss=3.2709
[LoRA-TF-100] Ep18/30 loss=3.2578
[LoRA-TF-100] Ep19/30 loss=3.2115
[LoRA-TF-100] Ep20/30 loss=3.1774
[LoRA-TF-100] Ep21/30 loss=3.1123
[LoRA-TF-100] Ep22/30 loss=3.1248
[LoRA-TF-100] Ep23/30 loss=3.1059
[LoRA-TF-100] Ep24/30 loss=3.0690
[LoRA-TF-100] Ep25/30 loss=3.0743
[LoRA-TF-100] Ep26/30 loss=2.9919
[LoRA-TF-100] Ep27/30 loss=2.9853
[LoRA-TF-100] Ep28/30 loss=3.0018
[LoRA-TF-100] Ep29/30 loss=2.9564
[LoRA-TF-100] Ep30/30 loss=2.9687
Training DANN-Gate (Transformer head)...
[DANN-Gate-TF-100] Ep1/30 λ_grl=0.000 loss=17.3414
[DANN-Gate-TF-100] Ep2/30 λ_grl=0.171 loss=11.2009
[DANN-Gate-TF-100] Ep3/30 λ_grl=0.332 loss=10.5356
[DANN-Gate-TF-100] Ep4/30 λ_grl=0.476 loss=10.2590
[DANN-Gate-TF-100] Ep5/30 λ_grl=0.598 loss=10.0214
[DANN-Gate-TF-100] Ep6/30 λ_grl=0.697 loss=9.8693
[DANN-Gate-TF-100] Ep7/30 λ_grl=0.776 loss=9.8875
[DANN-Gate-TF-100] Ep8/30 λ_grl=0.836 loss=9.7186
[DANN-Gate-TF-100] Ep9/30 λ_grl=0.881 loss=9.6566
[DANN-Gate-TF-100] Ep10/30 λ_grl=0.914 loss=9.3382
[DANN-Gate-TF-100] Ep11/30 λ_grl=0.938 loss=9.3256
[DANN-Gate-TF-100] Ep12/30 λ_grl=0.956 loss=9.1997
[DANN-Gate-TF-100] Ep13/30 λ_grl=0.969 loss=9.0457
[DANN-Gate-TF-100] Ep14/30 λ_grl=0.978 loss=9.0223
[DANN-Gate-TF-100] Ep15/30 λ_grl=0.984 loss=8.9299
[DANN-Gate-TF-100] Ep16/30 λ_grl=0.989 loss=8.7565
[DANN-Gate-TF-100] Ep17/30 λ_grl=0.992 loss=8.7481
[DANN-Gate-TF-100] Ep18/30 λ_grl=0.994 loss=8.6435
[DANN-Gate-TF-100] Ep19/30 λ_grl=0.996 loss=8.5802
[DANN-Gate-TF-100] Ep20/30 λ_grl=0.997 loss=8.5233
[DANN-Gate-TF-100] Ep21/30 λ_grl=0.998 loss=8.4743
[DANN-Gate-TF-100] Ep22/30 λ_grl=0.999 loss=8.3547
[DANN-Gate-TF-100] Ep23/30 λ_grl=0.999 loss=8.3823
[DANN-Gate-TF-100] Ep24/30 λ_grl=0.999 loss=8.3395
[DANN-Gate-TF-100] Ep25/30 λ_grl=0.999 loss=8.3130
[DANN-Gate-TF-100] Ep26/30 λ_grl=1.000 loss=8.1332
[DANN-Gate-TF-100] Ep27/30 λ_grl=1.000 loss=8.1280
[DANN-Gate-TF-100] Ep28/30 λ_grl=1.000 loss=8.1344
[DANN-Gate-TF-100] Ep29/30 λ_grl=1.000 loss=8.1307
[DANN-Gate-TF-100] Ep30/30 λ_grl=1.000 loss=8.0691
[Run 3] LoRA Acc=23.46% | DANN-Gate Acc=23.58%

=== Run 4/5, seed=45 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (Transformer head)...
[LoRA-TF-100] Ep1/30 loss=6.3410
[LoRA-TF-100] Ep2/30 loss=4.3683
[LoRA-TF-100] Ep3/30 loss=4.2110
[LoRA-TF-100] Ep4/30 loss=4.1124
[LoRA-TF-100] Ep5/30 loss=4.0520
[LoRA-TF-100] Ep6/30 loss=3.9772
[LoRA-TF-100] Ep7/30 loss=3.9121
[LoRA-TF-100] Ep8/30 loss=3.8599
[LoRA-TF-100] Ep9/30 loss=3.7742
[LoRA-TF-100] Ep10/30 loss=3.7240
[LoRA-TF-100] Ep11/30 loss=3.6456
[LoRA-TF-100] Ep12/30 loss=3.6341
[LoRA-TF-100] Ep13/30 loss=3.5513
[LoRA-TF-100] Ep14/30 loss=3.5192
[LoRA-TF-100] Ep15/30 loss=3.4695
[LoRA-TF-100] Ep16/30 loss=3.4297
[LoRA-TF-100] Ep17/30 loss=3.3966
[LoRA-TF-100] Ep18/30 loss=3.3752
[LoRA-TF-100] Ep19/30 loss=3.2955
[LoRA-TF-100] Ep20/30 loss=3.2985
[LoRA-TF-100] Ep21/30 loss=3.2656
[LoRA-TF-100] Ep22/30 loss=3.2412
[LoRA-TF-100] Ep23/30 loss=3.1815
[LoRA-TF-100] Ep24/30 loss=3.1739
[LoRA-TF-100] Ep25/30 loss=3.1487
[LoRA-TF-100] Ep26/30 loss=3.0917
[LoRA-TF-100] Ep27/30 loss=3.1114
[LoRA-TF-100] Ep28/30 loss=3.1117
[LoRA-TF-100] Ep29/30 loss=3.0526
[LoRA-TF-100] Ep30/30 loss=3.0403
Training DANN-Gate (Transformer head)...
[DANN-Gate-TF-100] Ep1/30 λ_grl=0.000 loss=16.1771
[DANN-Gate-TF-100] Ep2/30 λ_grl=0.171 loss=11.6182
[DANN-Gate-TF-100] Ep3/30 λ_grl=0.332 loss=10.7783
[DANN-Gate-TF-100] Ep4/30 λ_grl=0.476 loss=10.7752
[DANN-Gate-TF-100] Ep5/30 λ_grl=0.598 loss=10.5672
[DANN-Gate-TF-100] Ep6/30 λ_grl=0.697 loss=10.3304
[DANN-Gate-TF-100] Ep7/30 λ_grl=0.776 loss=10.0027
[DANN-Gate-TF-100] Ep8/30 λ_grl=0.836 loss=9.8788
[DANN-Gate-TF-100] Ep9/30 λ_grl=0.881 loss=9.8433
[DANN-Gate-TF-100] Ep10/30 λ_grl=0.914 loss=9.5243
[DANN-Gate-TF-100] Ep11/30 λ_grl=0.938 loss=9.5017
[DANN-Gate-TF-100] Ep12/30 λ_grl=0.956 loss=9.3863
[DANN-Gate-TF-100] Ep13/30 λ_grl=0.969 loss=9.2551
[DANN-Gate-TF-100] Ep14/30 λ_grl=0.978 loss=9.1503
[DANN-Gate-TF-100] Ep15/30 λ_grl=0.984 loss=9.0623
[DANN-Gate-TF-100] Ep16/30 λ_grl=0.989 loss=9.0769
[DANN-Gate-TF-100] Ep17/30 λ_grl=0.992 loss=8.8738
[DANN-Gate-TF-100] Ep18/30 λ_grl=0.994 loss=8.8479
[DANN-Gate-TF-100] Ep19/30 λ_grl=0.996 loss=8.7662
[DANN-Gate-TF-100] Ep20/30 λ_grl=0.997 loss=8.6086
[DANN-Gate-TF-100] Ep21/30 λ_grl=0.998 loss=8.7112
[DANN-Gate-TF-100] Ep22/30 λ_grl=0.999 loss=8.5421
[DANN-Gate-TF-100] Ep23/30 λ_grl=0.999 loss=8.4970
[DANN-Gate-TF-100] Ep24/30 λ_grl=0.999 loss=8.4701
[DANN-Gate-TF-100] Ep25/30 λ_grl=0.999 loss=8.4285
[DANN-Gate-TF-100] Ep26/30 λ_grl=1.000 loss=8.3681
[DANN-Gate-TF-100] Ep27/30 λ_grl=1.000 loss=8.2392
[DANN-Gate-TF-100] Ep28/30 λ_grl=1.000 loss=8.1724
[DANN-Gate-TF-100] Ep29/30 λ_grl=1.000 loss=8.1792
[DANN-Gate-TF-100] Ep30/30 λ_grl=1.000 loss=8.1080
[Run 4] LoRA Acc=23.46% | DANN-Gate Acc=23.57%

=== Run 5/5, seed=46 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (Transformer head)...
[LoRA-TF-100] Ep1/30 loss=6.2658
[LoRA-TF-100] Ep2/30 loss=4.1602
[LoRA-TF-100] Ep3/30 loss=4.0492
[LoRA-TF-100] Ep4/30 loss=3.9929
[LoRA-TF-100] Ep5/30 loss=3.9232
[LoRA-TF-100] Ep6/30 loss=3.8941
[LoRA-TF-100] Ep7/30 loss=3.7772
[LoRA-TF-100] Ep8/30 loss=3.7641
[LoRA-TF-100] Ep9/30 loss=3.6395
[LoRA-TF-100] Ep10/30 loss=3.6186
[LoRA-TF-100] Ep11/30 loss=3.5136
[LoRA-TF-100] Ep12/30 loss=3.4672
[LoRA-TF-100] Ep13/30 loss=3.4515
[LoRA-TF-100] Ep14/30 loss=3.3558
[LoRA-TF-100] Ep15/30 loss=3.3527
[LoRA-TF-100] Ep16/30 loss=3.3283
[LoRA-TF-100] Ep17/30 loss=3.2467
[LoRA-TF-100] Ep18/30 loss=3.1759
[LoRA-TF-100] Ep19/30 loss=3.1924
[LoRA-TF-100] Ep20/30 loss=3.2150
[LoRA-TF-100] Ep21/30 loss=3.1308
[LoRA-TF-100] Ep22/30 loss=3.1081
[LoRA-TF-100] Ep23/30 loss=3.1001
[LoRA-TF-100] Ep24/30 loss=3.0631
[LoRA-TF-100] Ep25/30 loss=3.0247
[LoRA-TF-100] Ep26/30 loss=2.9794
[LoRA-TF-100] Ep27/30 loss=3.0169
[LoRA-TF-100] Ep28/30 loss=2.9708
[LoRA-TF-100] Ep29/30 loss=2.9684
[LoRA-TF-100] Ep30/30 loss=2.9499
Training DANN-Gate (Transformer head)...
[DANN-Gate-TF-100] Ep1/30 λ_grl=0.000 loss=16.1433
[DANN-Gate-TF-100] Ep2/30 λ_grl=0.171 loss=11.2798
[DANN-Gate-TF-100] Ep3/30 λ_grl=0.332 loss=10.5391
[DANN-Gate-TF-100] Ep4/30 λ_grl=0.476 loss=10.1412
[DANN-Gate-TF-100] Ep5/30 λ_grl=0.598 loss=10.0409
[DANN-Gate-TF-100] Ep6/30 λ_grl=0.697 loss=9.8049
[DANN-Gate-TF-100] Ep7/30 λ_grl=0.776 loss=9.5204
[DANN-Gate-TF-100] Ep8/30 λ_grl=0.836 loss=9.4401
[DANN-Gate-TF-100] Ep9/30 λ_grl=0.881 loss=9.3490
[DANN-Gate-TF-100] Ep10/30 λ_grl=0.914 loss=9.3034
[DANN-Gate-TF-100] Ep11/30 λ_grl=0.938 loss=9.1820
[DANN-Gate-TF-100] Ep12/30 λ_grl=0.956 loss=9.0392
[DANN-Gate-TF-100] Ep13/30 λ_grl=0.969 loss=8.9946
[DANN-Gate-TF-100] Ep14/30 λ_grl=0.978 loss=8.8999
[DANN-Gate-TF-100] Ep15/30 λ_grl=0.984 loss=8.7983
[DANN-Gate-TF-100] Ep16/30 λ_grl=0.989 loss=8.7950
[DANN-Gate-TF-100] Ep17/30 λ_grl=0.992 loss=8.6871
[DANN-Gate-TF-100] Ep18/30 λ_grl=0.994 loss=8.5912
[DANN-Gate-TF-100] Ep19/30 λ_grl=0.996 loss=8.5002
[DANN-Gate-TF-100] Ep20/30 λ_grl=0.997 loss=8.4388
[DANN-Gate-TF-100] Ep21/30 λ_grl=0.998 loss=8.4739
[DANN-Gate-TF-100] Ep22/30 λ_grl=0.999 loss=8.2824
[DANN-Gate-TF-100] Ep23/30 λ_grl=0.999 loss=8.2354
[DANN-Gate-TF-100] Ep24/30 λ_grl=0.999 loss=8.2683
[DANN-Gate-TF-100] Ep25/30 λ_grl=0.999 loss=8.2653
[DANN-Gate-TF-100] Ep26/30 λ_grl=1.000 loss=8.2063
[DANN-Gate-TF-100] Ep27/30 λ_grl=1.000 loss=8.1437
[DANN-Gate-TF-100] Ep28/30 λ_grl=1.000 loss=7.9911
[DANN-Gate-TF-100] Ep29/30 λ_grl=1.000 loss=7.8994
[DANN-Gate-TF-100] Ep30/30 λ_grl=1.000 loss=8.0096
[Run 5] LoRA Acc=23.67% | DANN-Gate Acc=23.33%

All done. Results saved to: ./results_test100_base/imbalance_tf_cifar100.json
more_baselines/base_imb_tf_test100.py completed successfully.
Starting more_baselines/base_mismatch_tf_test100.py...

=== Pretraining Teacher (BigTransformer-10) on CIFAR-10 subset ===
Files already downloaded and verified
[Epoch 1] Loss: 2.1171
[Epoch 2] Loss: 2.0019
[Epoch 3] Loss: 1.9035
[Epoch 4] Loss: 1.8162
[Epoch 5] Loss: 1.7313
[Epoch 6] Loss: 1.6690
[Epoch 7] Loss: 1.6221
[Epoch 8] Loss: 1.5791
[Epoch 9] Loss: 1.5354
[Epoch 10] Loss: 1.4789
[Epoch 11] Loss: 1.4486
[Epoch 12] Loss: 1.3982
[Epoch 13] Loss: 1.3459
[Epoch 14] Loss: 1.2940
[Epoch 15] Loss: 1.2623
[Epoch 16] Loss: 1.2272
[Epoch 17] Loss: 1.1892
[Epoch 18] Loss: 1.1485
[Epoch 19] Loss: 1.1178
[Epoch 20] Loss: 1.0767
[Epoch 21] Loss: 1.0564
[Epoch 22] Loss: 1.0307
[Epoch 23] Loss: 0.9926
[Epoch 24] Loss: 0.9596
[Epoch 25] Loss: 0.9247
[Epoch 26] Loss: 0.8552
[Epoch 27] Loss: 0.8258
[Epoch 28] Loss: 0.8224
[Epoch 29] Loss: 0.7998
[Epoch 30] Loss: 0.7290
[Epoch 31] Loss: 0.6979
[Epoch 32] Loss: 0.7229
[Epoch 33] Loss: 0.6841
[Epoch 34] Loss: 0.6359
[Epoch 35] Loss: 0.6070
[Epoch 36] Loss: 0.5775
[Epoch 37] Loss: 0.5588
[Epoch 38] Loss: 0.5525
[Epoch 39] Loss: 0.5068
[Epoch 40] Loss: 0.4829
[Epoch 41] Loss: 0.4785
[Epoch 42] Loss: 0.4409
[Epoch 43] Loss: 0.4325
[Epoch 44] Loss: 0.4728
[Epoch 45] Loss: 0.4451
[Epoch 46] Loss: 0.3758
[Epoch 47] Loss: 0.3776
[Epoch 48] Loss: 0.3260
[Epoch 49] Loss: 0.3001
[Epoch 50] Loss: 0.3238
[Epoch 51] Loss: 0.3248
[Epoch 52] Loss: 0.3001
[Epoch 53] Loss: 0.2831
[Epoch 54] Loss: 0.2853
[Epoch 55] Loss: 0.2774
[Epoch 56] Loss: 0.2972
[Epoch 57] Loss: 0.2619
[Epoch 58] Loss: 0.2598
[Epoch 59] Loss: 0.2458
[Epoch 60] Loss: 0.2691
Trained and saved teacher transformer to: ./model_test100/base_mismatch_tf_teacher_cifar10.pt
Files already downloaded and verified
Files already downloaded and verified

=== Run 1/5, seed=42 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (convert teacher head to 100, then LoRA-head)...
[LoRA-TF-100] Ep1/30 loss=4.5418
[LoRA-TF-100] Ep2/30 loss=4.1202
[LoRA-TF-100] Ep3/30 loss=3.8584
[LoRA-TF-100] Ep4/30 loss=3.7235
[LoRA-TF-100] Ep5/30 loss=3.6384
[LoRA-TF-100] Ep6/30 loss=3.5732
[LoRA-TF-100] Ep7/30 loss=3.5060
[LoRA-TF-100] Ep8/30 loss=3.4649
[LoRA-TF-100] Ep9/30 loss=3.4298
[LoRA-TF-100] Ep10/30 loss=3.3954
[LoRA-TF-100] Ep11/30 loss=3.3621
[LoRA-TF-100] Ep12/30 loss=3.3332
[LoRA-TF-100] Ep13/30 loss=3.3098
[LoRA-TF-100] Ep14/30 loss=3.2985
[LoRA-TF-100] Ep15/30 loss=3.2776
[LoRA-TF-100] Ep16/30 loss=3.2379
[LoRA-TF-100] Ep17/30 loss=3.2324
[LoRA-TF-100] Ep18/30 loss=3.2210
[LoRA-TF-100] Ep19/30 loss=3.2163
[LoRA-TF-100] Ep20/30 loss=3.1933
[LoRA-TF-100] Ep21/30 loss=3.1742
[LoRA-TF-100] Ep22/30 loss=3.1633
[LoRA-TF-100] Ep23/30 loss=3.1588
[LoRA-TF-100] Ep24/30 loss=3.1378
[LoRA-TF-100] Ep25/30 loss=3.1411
[LoRA-TF-100] Ep26/30 loss=3.1244
[LoRA-TF-100] Ep27/30 loss=3.1103
[LoRA-TF-100] Ep28/30 loss=3.1103
[LoRA-TF-100] Ep29/30 loss=3.0969
[LoRA-TF-100] Ep30/30 loss=3.0943
Training DANN-Gate (convert teacher head to 100, head-only training)...
[DANN-Gate-TF-100] Ep1/30 λ_grl=0.000 loss=11.2934
[DANN-Gate-TF-100] Ep2/30 λ_grl=0.171 loss=10.3840
[DANN-Gate-TF-100] Ep3/30 λ_grl=0.332 loss=9.7935
[DANN-Gate-TF-100] Ep4/30 λ_grl=0.476 loss=9.5713
[DANN-Gate-TF-100] Ep5/30 λ_grl=0.598 loss=9.4238
[DANN-Gate-TF-100] Ep6/30 λ_grl=0.697 loss=9.2764
[DANN-Gate-TF-100] Ep7/30 λ_grl=0.776 loss=9.1467
[DANN-Gate-TF-100] Ep8/30 λ_grl=0.836 loss=9.0307
[DANN-Gate-TF-100] Ep9/30 λ_grl=0.881 loss=8.9273
[DANN-Gate-TF-100] Ep10/30 λ_grl=0.914 loss=8.9295
[DANN-Gate-TF-100] Ep11/30 λ_grl=0.938 loss=8.8480
[DANN-Gate-TF-100] Ep12/30 λ_grl=0.956 loss=8.7291
[DANN-Gate-TF-100] Ep13/30 λ_grl=0.969 loss=8.6844
[DANN-Gate-TF-100] Ep14/30 λ_grl=0.978 loss=8.7057
[DANN-Gate-TF-100] Ep15/30 λ_grl=0.984 loss=8.6895
[DANN-Gate-TF-100] Ep16/30 λ_grl=0.989 loss=8.5614
[DANN-Gate-TF-100] Ep17/30 λ_grl=0.992 loss=8.5344
[DANN-Gate-TF-100] Ep18/30 λ_grl=0.994 loss=8.5583
[DANN-Gate-TF-100] Ep19/30 λ_grl=0.996 loss=8.5097
[DANN-Gate-TF-100] Ep20/30 λ_grl=0.997 loss=8.5117
[DANN-Gate-TF-100] Ep21/30 λ_grl=0.998 loss=8.4860
[DANN-Gate-TF-100] Ep22/30 λ_grl=0.999 loss=8.3786
[DANN-Gate-TF-100] Ep23/30 λ_grl=0.999 loss=8.4086
[DANN-Gate-TF-100] Ep24/30 λ_grl=0.999 loss=8.4075
[DANN-Gate-TF-100] Ep25/30 λ_grl=0.999 loss=8.3277
[DANN-Gate-TF-100] Ep26/30 λ_grl=1.000 loss=8.3075
[DANN-Gate-TF-100] Ep27/30 λ_grl=1.000 loss=8.3507
[DANN-Gate-TF-100] Ep28/30 λ_grl=1.000 loss=8.2816
[DANN-Gate-TF-100] Ep29/30 λ_grl=1.000 loss=8.3074
[DANN-Gate-TF-100] Ep30/30 λ_grl=1.000 loss=8.2784
[Run 1] LoRA Acc=14.60% | DANN-Gate Acc=14.61%

=== Run 2/5, seed=43 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (convert teacher head to 100, then LoRA-head)...
[LoRA-TF-100] Ep1/30 loss=4.5540
[LoRA-TF-100] Ep2/30 loss=4.1357
[LoRA-TF-100] Ep3/30 loss=3.8490
[LoRA-TF-100] Ep4/30 loss=3.7045
[LoRA-TF-100] Ep5/30 loss=3.6066
[LoRA-TF-100] Ep6/30 loss=3.5362
[LoRA-TF-100] Ep7/30 loss=3.4893
[LoRA-TF-100] Ep8/30 loss=3.4347
[LoRA-TF-100] Ep9/30 loss=3.3981
[LoRA-TF-100] Ep10/30 loss=3.3761
[LoRA-TF-100] Ep11/30 loss=3.3547
[LoRA-TF-100] Ep12/30 loss=3.3218
[LoRA-TF-100] Ep13/30 loss=3.2969
[LoRA-TF-100] Ep14/30 loss=3.2825
[LoRA-TF-100] Ep15/30 loss=3.2542
[LoRA-TF-100] Ep16/30 loss=3.2487
[LoRA-TF-100] Ep17/30 loss=3.2303
[LoRA-TF-100] Ep18/30 loss=3.2130
[LoRA-TF-100] Ep19/30 loss=3.1990
[LoRA-TF-100] Ep20/30 loss=3.1676
[LoRA-TF-100] Ep21/30 loss=3.1534
[LoRA-TF-100] Ep22/30 loss=3.1477
[LoRA-TF-100] Ep23/30 loss=3.1455
[LoRA-TF-100] Ep24/30 loss=3.1379
[LoRA-TF-100] Ep25/30 loss=3.1141
[LoRA-TF-100] Ep26/30 loss=3.1171
[LoRA-TF-100] Ep27/30 loss=3.0929
[LoRA-TF-100] Ep28/30 loss=3.0828
[LoRA-TF-100] Ep29/30 loss=3.0910
[LoRA-TF-100] Ep30/30 loss=3.0908
Training DANN-Gate (convert teacher head to 100, head-only training)...
[DANN-Gate-TF-100] Ep1/30 λ_grl=0.000 loss=11.2538
[DANN-Gate-TF-100] Ep2/30 λ_grl=0.171 loss=10.3463
[DANN-Gate-TF-100] Ep3/30 λ_grl=0.332 loss=9.7805
[DANN-Gate-TF-100] Ep4/30 λ_grl=0.476 loss=9.4725
[DANN-Gate-TF-100] Ep5/30 λ_grl=0.598 loss=9.3682
[DANN-Gate-TF-100] Ep6/30 λ_grl=0.697 loss=9.2204
[DANN-Gate-TF-100] Ep7/30 λ_grl=0.776 loss=9.0999
[DANN-Gate-TF-100] Ep8/30 λ_grl=0.836 loss=9.0171
[DANN-Gate-TF-100] Ep9/30 λ_grl=0.881 loss=8.9488
[DANN-Gate-TF-100] Ep10/30 λ_grl=0.914 loss=8.8619
[DANN-Gate-TF-100] Ep11/30 λ_grl=0.938 loss=8.7707
[DANN-Gate-TF-100] Ep12/30 λ_grl=0.956 loss=8.7655
[DANN-Gate-TF-100] Ep13/30 λ_grl=0.969 loss=8.6407
[DANN-Gate-TF-100] Ep14/30 λ_grl=0.978 loss=8.6255
[DANN-Gate-TF-100] Ep15/30 λ_grl=0.984 loss=8.5873
[DANN-Gate-TF-100] Ep16/30 λ_grl=0.989 loss=8.5275
[DANN-Gate-TF-100] Ep17/30 λ_grl=0.992 loss=8.4825
[DANN-Gate-TF-100] Ep18/30 λ_grl=0.994 loss=8.4568
[DANN-Gate-TF-100] Ep19/30 λ_grl=0.996 loss=8.4256
[DANN-Gate-TF-100] Ep20/30 λ_grl=0.997 loss=8.4329
[DANN-Gate-TF-100] Ep21/30 λ_grl=0.998 loss=8.3616
[DANN-Gate-TF-100] Ep22/30 λ_grl=0.999 loss=8.3170
[DANN-Gate-TF-100] Ep23/30 λ_grl=0.999 loss=8.3906
[DANN-Gate-TF-100] Ep24/30 λ_grl=0.999 loss=8.3586
[DANN-Gate-TF-100] Ep25/30 λ_grl=0.999 loss=8.3073
[DANN-Gate-TF-100] Ep26/30 λ_grl=1.000 loss=8.2949
[DANN-Gate-TF-100] Ep27/30 λ_grl=1.000 loss=8.2776
[DANN-Gate-TF-100] Ep28/30 λ_grl=1.000 loss=8.3255
[DANN-Gate-TF-100] Ep29/30 λ_grl=1.000 loss=8.1973
[DANN-Gate-TF-100] Ep30/30 λ_grl=1.000 loss=8.2309
[Run 2] LoRA Acc=14.29% | DANN-Gate Acc=13.78%

=== Run 3/5, seed=44 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (convert teacher head to 100, then LoRA-head)...
[LoRA-TF-100] Ep1/30 loss=4.5293
[LoRA-TF-100] Ep2/30 loss=4.1116
[LoRA-TF-100] Ep3/30 loss=3.8423
[LoRA-TF-100] Ep4/30 loss=3.7171
[LoRA-TF-100] Ep5/30 loss=3.6115
[LoRA-TF-100] Ep6/30 loss=3.5313
[LoRA-TF-100] Ep7/30 loss=3.4763
[LoRA-TF-100] Ep8/30 loss=3.4242
[LoRA-TF-100] Ep9/30 loss=3.3748
[LoRA-TF-100] Ep10/30 loss=3.3566
[LoRA-TF-100] Ep11/30 loss=3.3219
[LoRA-TF-100] Ep12/30 loss=3.2853
[LoRA-TF-100] Ep13/30 loss=3.2769
[LoRA-TF-100] Ep14/30 loss=3.2534
[LoRA-TF-100] Ep15/30 loss=3.2370
[LoRA-TF-100] Ep16/30 loss=3.2158
[LoRA-TF-100] Ep17/30 loss=3.1856
[LoRA-TF-100] Ep18/30 loss=3.1790
[LoRA-TF-100] Ep19/30 loss=3.1731
[LoRA-TF-100] Ep20/30 loss=3.1516
[LoRA-TF-100] Ep21/30 loss=3.1466
[LoRA-TF-100] Ep22/30 loss=3.1236
[LoRA-TF-100] Ep23/30 loss=3.1210
[LoRA-TF-100] Ep24/30 loss=3.0875
[LoRA-TF-100] Ep25/30 loss=3.0914
[LoRA-TF-100] Ep26/30 loss=3.0693
[LoRA-TF-100] Ep27/30 loss=3.0529
[LoRA-TF-100] Ep28/30 loss=3.0541
[LoRA-TF-100] Ep29/30 loss=3.0729
[LoRA-TF-100] Ep30/30 loss=3.0393
Training DANN-Gate (convert teacher head to 100, head-only training)...
[DANN-Gate-TF-100] Ep1/30 λ_grl=0.000 loss=11.3846
[DANN-Gate-TF-100] Ep2/30 λ_grl=0.171 loss=10.2593
[DANN-Gate-TF-100] Ep3/30 λ_grl=0.332 loss=9.8467
[DANN-Gate-TF-100] Ep4/30 λ_grl=0.476 loss=9.5121
[DANN-Gate-TF-100] Ep5/30 λ_grl=0.598 loss=9.3711
[DANN-Gate-TF-100] Ep6/30 λ_grl=0.697 loss=9.1624
[DANN-Gate-TF-100] Ep7/30 λ_grl=0.776 loss=9.1006
[DANN-Gate-TF-100] Ep8/30 λ_grl=0.836 loss=9.0052
[DANN-Gate-TF-100] Ep9/30 λ_grl=0.881 loss=8.8913
[DANN-Gate-TF-100] Ep10/30 λ_grl=0.914 loss=8.9025
[DANN-Gate-TF-100] Ep11/30 λ_grl=0.938 loss=8.8212
[DANN-Gate-TF-100] Ep12/30 λ_grl=0.956 loss=8.7568
[DANN-Gate-TF-100] Ep13/30 λ_grl=0.969 loss=8.6721
[DANN-Gate-TF-100] Ep14/30 λ_grl=0.978 loss=8.6433
[DANN-Gate-TF-100] Ep15/30 λ_grl=0.984 loss=8.6610
[DANN-Gate-TF-100] Ep16/30 λ_grl=0.989 loss=8.5538
[DANN-Gate-TF-100] Ep17/30 λ_grl=0.992 loss=8.4867
[DANN-Gate-TF-100] Ep18/30 λ_grl=0.994 loss=8.5145
[DANN-Gate-TF-100] Ep19/30 λ_grl=0.996 loss=8.4859
[DANN-Gate-TF-100] Ep20/30 λ_grl=0.997 loss=8.4032
[DANN-Gate-TF-100] Ep21/30 λ_grl=0.998 loss=8.4067
[DANN-Gate-TF-100] Ep22/30 λ_grl=0.999 loss=8.3789
[DANN-Gate-TF-100] Ep23/30 λ_grl=0.999 loss=8.3438
[DANN-Gate-TF-100] Ep24/30 λ_grl=0.999 loss=8.3063
[DANN-Gate-TF-100] Ep25/30 λ_grl=0.999 loss=8.3522
[DANN-Gate-TF-100] Ep26/30 λ_grl=1.000 loss=8.2923
[DANN-Gate-TF-100] Ep27/30 λ_grl=1.000 loss=8.2795
[DANN-Gate-TF-100] Ep28/30 λ_grl=1.000 loss=8.2713
[DANN-Gate-TF-100] Ep29/30 λ_grl=1.000 loss=8.1902
[DANN-Gate-TF-100] Ep30/30 λ_grl=1.000 loss=8.1878
[Run 3] LoRA Acc=14.36% | DANN-Gate Acc=14.00%

=== Run 4/5, seed=45 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (convert teacher head to 100, then LoRA-head)...
[LoRA-TF-100] Ep1/30 loss=4.5208
[LoRA-TF-100] Ep2/30 loss=4.0954
[LoRA-TF-100] Ep3/30 loss=3.8988
[LoRA-TF-100] Ep4/30 loss=3.7974
[LoRA-TF-100] Ep5/30 loss=3.7142
[LoRA-TF-100] Ep6/30 loss=3.6413
[LoRA-TF-100] Ep7/30 loss=3.5973
[LoRA-TF-100] Ep8/30 loss=3.5433
[LoRA-TF-100] Ep9/30 loss=3.4957
[LoRA-TF-100] Ep10/30 loss=3.4581
[LoRA-TF-100] Ep11/30 loss=3.4278
[LoRA-TF-100] Ep12/30 loss=3.3879
[LoRA-TF-100] Ep13/30 loss=3.3705
[LoRA-TF-100] Ep14/30 loss=3.3369
[LoRA-TF-100] Ep15/30 loss=3.3187
[LoRA-TF-100] Ep16/30 loss=3.2847
[LoRA-TF-100] Ep17/30 loss=3.2943
[LoRA-TF-100] Ep18/30 loss=3.2637
[LoRA-TF-100] Ep19/30 loss=3.2518
[LoRA-TF-100] Ep20/30 loss=3.2377
[LoRA-TF-100] Ep21/30 loss=3.2351
[LoRA-TF-100] Ep22/30 loss=3.2045
[LoRA-TF-100] Ep23/30 loss=3.1945
[LoRA-TF-100] Ep24/30 loss=3.1751
[LoRA-TF-100] Ep25/30 loss=3.1816
[LoRA-TF-100] Ep26/30 loss=3.1614
[LoRA-TF-100] Ep27/30 loss=3.1600
[LoRA-TF-100] Ep28/30 loss=3.1535
[LoRA-TF-100] Ep29/30 loss=3.1126
[LoRA-TF-100] Ep30/30 loss=3.1217
Training DANN-Gate (convert teacher head to 100, head-only training)...
[DANN-Gate-TF-100] Ep1/30 λ_grl=0.000 loss=11.0175
[DANN-Gate-TF-100] Ep2/30 λ_grl=0.171 loss=10.3114
[DANN-Gate-TF-100] Ep3/30 λ_grl=0.332 loss=9.9223
[DANN-Gate-TF-100] Ep4/30 λ_grl=0.476 loss=9.6761
[DANN-Gate-TF-100] Ep5/30 λ_grl=0.598 loss=9.4607
[DANN-Gate-TF-100] Ep6/30 λ_grl=0.697 loss=9.3655
[DANN-Gate-TF-100] Ep7/30 λ_grl=0.776 loss=9.2931
[DANN-Gate-TF-100] Ep8/30 λ_grl=0.836 loss=9.0979
[DANN-Gate-TF-100] Ep9/30 λ_grl=0.881 loss=9.1076
[DANN-Gate-TF-100] Ep10/30 λ_grl=0.914 loss=8.9865
[DANN-Gate-TF-100] Ep11/30 λ_grl=0.938 loss=8.9124
[DANN-Gate-TF-100] Ep12/30 λ_grl=0.956 loss=8.8728
[DANN-Gate-TF-100] Ep13/30 λ_grl=0.969 loss=8.9357
[DANN-Gate-TF-100] Ep14/30 λ_grl=0.978 loss=8.7818
[DANN-Gate-TF-100] Ep15/30 λ_grl=0.984 loss=8.7729
[DANN-Gate-TF-100] Ep16/30 λ_grl=0.989 loss=8.6772
[DANN-Gate-TF-100] Ep17/30 λ_grl=0.992 loss=8.6278
[DANN-Gate-TF-100] Ep18/30 λ_grl=0.994 loss=8.6191
[DANN-Gate-TF-100] Ep19/30 λ_grl=0.996 loss=8.5855
[DANN-Gate-TF-100] Ep20/30 λ_grl=0.997 loss=8.5534
[DANN-Gate-TF-100] Ep21/30 λ_grl=0.998 loss=8.5334
[DANN-Gate-TF-100] Ep22/30 λ_grl=0.999 loss=8.5156
[DANN-Gate-TF-100] Ep23/30 λ_grl=0.999 loss=8.4630
[DANN-Gate-TF-100] Ep24/30 λ_grl=0.999 loss=8.5242
[DANN-Gate-TF-100] Ep25/30 λ_grl=0.999 loss=8.4283
[DANN-Gate-TF-100] Ep26/30 λ_grl=1.000 loss=8.4283
[DANN-Gate-TF-100] Ep27/30 λ_grl=1.000 loss=8.4083
[DANN-Gate-TF-100] Ep28/30 λ_grl=1.000 loss=8.4065
[DANN-Gate-TF-100] Ep29/30 λ_grl=1.000 loss=8.3739
[DANN-Gate-TF-100] Ep30/30 λ_grl=1.000 loss=8.3289
[Run 4] LoRA Acc=13.99% | DANN-Gate Acc=13.62%

=== Run 5/5, seed=46 ===
Files already downloaded and verified
Files already downloaded and verified
Training LoRA (convert teacher head to 100, then LoRA-head)...
[LoRA-TF-100] Ep1/30 loss=4.5627
[LoRA-TF-100] Ep2/30 loss=4.1185
[LoRA-TF-100] Ep3/30 loss=3.8769
[LoRA-TF-100] Ep4/30 loss=3.7426
[LoRA-TF-100] Ep5/30 loss=3.6525
[LoRA-TF-100] Ep6/30 loss=3.5788
[LoRA-TF-100] Ep7/30 loss=3.5167
[LoRA-TF-100] Ep8/30 loss=3.4658
[LoRA-TF-100] Ep9/30 loss=3.4343
[LoRA-TF-100] Ep10/30 loss=3.4094
[LoRA-TF-100] Ep11/30 loss=3.3889
[LoRA-TF-100] Ep12/30 loss=3.3660
[LoRA-TF-100] Ep13/30 loss=3.3340
[LoRA-TF-100] Ep14/30 loss=3.3098
[LoRA-TF-100] Ep15/30 loss=3.2992
[LoRA-TF-100] Ep16/30 loss=3.2699
[LoRA-TF-100] Ep17/30 loss=3.2669
[LoRA-TF-100] Ep18/30 loss=3.2294
[LoRA-TF-100] Ep19/30 loss=3.2265
[LoRA-TF-100] Ep20/30 loss=3.2052
[LoRA-TF-100] Ep21/30 loss=3.1878
[LoRA-TF-100] Ep22/30 loss=3.1863
[LoRA-TF-100] Ep23/30 loss=3.1658
[LoRA-TF-100] Ep24/30 loss=3.1667
[LoRA-TF-100] Ep25/30 loss=3.1499
[LoRA-TF-100] Ep26/30 loss=3.1504
[LoRA-TF-100] Ep27/30 loss=3.1268
[LoRA-TF-100] Ep28/30 loss=3.1298
[LoRA-TF-100] Ep29/30 loss=3.1159
[LoRA-TF-100] Ep30/30 loss=3.1188
Training DANN-Gate (convert teacher head to 100, head-only training)...
[DANN-Gate-TF-100] Ep1/30 λ_grl=0.000 loss=11.0519
[DANN-Gate-TF-100] Ep2/30 λ_grl=0.171 loss=10.4255
[DANN-Gate-TF-100] Ep3/30 λ_grl=0.332 loss=9.9944
[DANN-Gate-TF-100] Ep4/30 λ_grl=0.476 loss=9.7096
[DANN-Gate-TF-100] Ep5/30 λ_grl=0.598 loss=9.4984
[DANN-Gate-TF-100] Ep6/30 λ_grl=0.697 loss=9.4383
[DANN-Gate-TF-100] Ep7/30 λ_grl=0.776 loss=9.2285
[DANN-Gate-TF-100] Ep8/30 λ_grl=0.836 loss=9.1922
[DANN-Gate-TF-100] Ep9/30 λ_grl=0.881 loss=9.0714
[DANN-Gate-TF-100] Ep10/30 λ_grl=0.914 loss=8.9686
[DANN-Gate-TF-100] Ep11/30 λ_grl=0.938 loss=8.8817
[DANN-Gate-TF-100] Ep12/30 λ_grl=0.956 loss=8.8814
[DANN-Gate-TF-100] Ep13/30 λ_grl=0.969 loss=8.7379
[DANN-Gate-TF-100] Ep14/30 λ_grl=0.978 loss=8.7167
[DANN-Gate-TF-100] Ep15/30 λ_grl=0.984 loss=8.6855
[DANN-Gate-TF-100] Ep16/30 λ_grl=0.989 loss=8.6413
[DANN-Gate-TF-100] Ep17/30 λ_grl=0.992 loss=8.6381
[DANN-Gate-TF-100] Ep18/30 λ_grl=0.994 loss=8.5837
[DANN-Gate-TF-100] Ep19/30 λ_grl=0.996 loss=8.5131
[DANN-Gate-TF-100] Ep20/30 λ_grl=0.997 loss=8.5445
[DANN-Gate-TF-100] Ep21/30 λ_grl=0.998 loss=8.4497
[DANN-Gate-TF-100] Ep22/30 λ_grl=0.999 loss=8.4254
[DANN-Gate-TF-100] Ep23/30 λ_grl=0.999 loss=8.4618
[DANN-Gate-TF-100] Ep24/30 λ_grl=0.999 loss=8.3908
[DANN-Gate-TF-100] Ep25/30 λ_grl=0.999 loss=8.4160
[DANN-Gate-TF-100] Ep26/30 λ_grl=1.000 loss=8.3822
[DANN-Gate-TF-100] Ep27/30 λ_grl=1.000 loss=8.3770
[DANN-Gate-TF-100] Ep28/30 λ_grl=1.000 loss=8.3099
[DANN-Gate-TF-100] Ep29/30 λ_grl=1.000 loss=8.3142
[DANN-Gate-TF-100] Ep30/30 λ_grl=1.000 loss=8.2804
[Run 5] LoRA Acc=13.87% | DANN-Gate Acc=14.39%

All runs complete. Results saved to ./results_test100_base/mismatch_tf_test100.json
more_baselines/base_mismatch_tf_test100.py completed successfully.
