=== EXP 2 (arch AB x ds=1) start 2026-07-15T11:31:28+00:00 ===
[main] device=cuda
[main] paired train=11610  val=3582
Setting up [LPIPS] perceptual loss: trunk [alex], v[0.1], spatial [off]
/usr/local/lib/python3.12/dist-packages/torchvision/models/_utils.py:208: UserWarning: The parameter 'pretrained' is deprecated since 0.13 and may be removed in the future, please use 'weights' instead.
  warnings.warn(
/usr/local/lib/python3.12/dist-packages/torchvision/models/_utils.py:223: UserWarning: Arguments other than a weight enum or `None` for 'weights' are deprecated since 0.13 and may be removed in the future. The current behavior is equivalent to passing `weights=AlexNet_Weights.IMAGENET1K_V1`. You can also use `weights=AlexNet_Weights.DEFAULT` to get the most up-to-date weights.
  warnings.warn(msg)
Downloading: "https://download.pytorch.org/models/alexnet-owt-7be5be79.pth" to /home/ubuntu/.cache/torch/hub/checkpoints/alexnet-owt-7be5be79.pth

  0%|          | 0.00/233M [00:00<?, ?B/s]
  3%|▎         | 6.75M/233M [00:00<00:03, 70.5MB/s]
 13%|█▎        | 30.9M/233M [00:00<00:01, 177MB/s] 
 25%|██▌       | 58.5M/233M [00:00<00:00, 228MB/s]
 37%|███▋      | 86.8M/233M [00:00<00:00, 255MB/s]
 48%|████▊     | 113M/233M [00:00<00:00, 261MB/s] 
 59%|█████▉    | 138M/233M [00:00<00:00, 237MB/s]
 70%|██████▉   | 163M/233M [00:00<00:00, 244MB/s]
 82%|████████▏ | 190M/233M [00:00<00:00, 257MB/s]
 94%|█████████▎| 218M/233M [00:00<00:00, 268MB/s]
100%|██████████| 233M/233M [00:00<00:00, 246MB/s]
Loading model from: /home/ubuntu/.local/lib/python3.12/site-packages/lpips/weights/v0.1/alex.pth
[arch=A ds=1] params = 11.12M
[arch=A ds=1] step   500  val_l1=0.1432  val_lpips=0.3509  best=0.3509  (249s)
[arch=A ds=1] step  1000  val_l1=0.1349  val_lpips=0.3188  best=0.3188  (497s)
[arch=A ds=1] step  1500  val_l1=0.1308  val_lpips=0.2996  best=0.2996  (748s)
[arch=A ds=1] step  2000  val_l1=0.1264  val_lpips=0.2920  best=0.2920  (995s)
[arch=A ds=1] step  2500  val_l1=0.1239  val_lpips=0.2677  best=0.2677  (1242s)
[arch=A ds=1] step  3000  val_l1=0.1235  val_lpips=0.2607  best=0.2607  (1489s)
[arch=A ds=1] step  3500  val_l1=0.1221  val_lpips=0.2509  best=0.2509  (1732s)
[arch=A ds=1] step  4000  val_l1=0.1224  val_lpips=0.2529  best=0.2509  (1976s)
[arch=A ds=1] step  4500  val_l1=0.1216  val_lpips=0.2450  best=0.2450  (2223s)
[arch=A ds=1] step  5000  val_l1=0.1229  val_lpips=0.2488  best=0.2450  (2473s)
[arch=A ds=1] step  5500  val_l1=0.1212  val_lpips=0.2344  best=0.2344  (2719s)
[arch=A ds=1] step  6000  val_l1=0.1210  val_lpips=0.2414  best=0.2344  (2962s)
[arch=A ds=1] step  6500  val_l1=0.1216  val_lpips=0.2359  best=0.2344  (3216s)
[arch=A ds=1] step  7000  val_l1=0.1211  val_lpips=0.2331  best=0.2331  (3468s)
[arch=A ds=1] step  7500  val_l1=0.1215  val_lpips=0.2300  best=0.2300  (3716s)
[arch=A ds=1] step  8000  val_l1=0.1226  val_lpips=0.2273  best=0.2273  (3961s)
[arch=A ds=1] DONE. best_val_lpips=0.2273 final=0.2273 elapsed=4007s
[arch=B ds=1] params = 11.12M
[arch=B ds=1] step   500  val_l1=0.1523  val_lpips=0.3325  best=0.3325  (245s)
[arch=B ds=1] step  1000  val_l1=0.1535  val_lpips=0.3081  best=0.3081  (482s)
[arch=B ds=1] step  1500  val_l1=0.1531  val_lpips=0.3034  best=0.3034  (698s)
[arch=B ds=1] step  2000  val_l1=0.1522  val_lpips=0.3062  best=0.3034  (951s)
[arch=B ds=1] step  2500  val_l1=0.1524  val_lpips=0.3021  best=0.3021  (1197s)
[arch=B ds=1] step  3000  val_l1=0.1518  val_lpips=0.2994  best=0.2994  (1444s)
[arch=B ds=1] step  3500  val_l1=0.1476  val_lpips=0.3042  best=0.2994  (1686s)
[arch=B ds=1] step  4000  val_l1=0.1441  val_lpips=0.3010  best=0.2994  (1928s)
[arch=B ds=1] step  4500  val_l1=0.1406  val_lpips=0.3019  best=0.2994  (2172s)
[arch=B ds=1] step  5000  val_l1=0.1387  val_lpips=0.2995  best=0.2994  (2408s)
[arch=B ds=1] step  5500  val_l1=0.1399  val_lpips=0.3038  best=0.2994  (2647s)
[arch=B ds=1] step  6000  val_l1=0.1364  val_lpips=0.2983  best=0.2983  (2881s)
[arch=B ds=1] step  6500  val_l1=0.1371  val_lpips=0.2954  best=0.2954  (3123s)
[arch=B ds=1] step  7000  val_l1=0.1393  val_lpips=0.3018  best=0.2954  (3363s)
[arch=B ds=1] step  7500  val_l1=0.1326  val_lpips=0.2949  best=0.2949  (3576s)
[arch=B ds=1] step  8000  val_l1=0.1323  val_lpips=0.2955  best=0.2949  (3778s)
[arch=B ds=1] DONE. best_val_lpips=0.2949 final=0.2955 elapsed=3814s

=== LEADERBOARD (best val LPIPS, lower better) ===
arch   ds   best_lpips   final_lpips   final_l1  params(M)
   A    1       0.2273        0.2273     0.1226      11.12
   B    1       0.2949        0.2955     0.1323      11.12
=== EXP 2 done 2026-07-15T13:42:10+00:00 exit=0 ===