/opt/venv/lib/python3.10/site-packages/apex/transformer/functional/fused_rope.py:49: UserWarning: Aiter backend is selected for fused RoPE. This has lower precision. To disable aiter, export USE_ROCM_AITER_ROPE_BACKEND=0 warnings.warn("Aiter backend is selected for fused RoPE. This has lower precision. To disable aiter, export USE_ROCM_AITER_ROPE_BACKEND=0", UserWarning) /opt/venv/lib/python3.10/site-packages/timm/models/layers/__init__.py:49: FutureWarning: Importing from timm.models.layers is deprecated, please import via timm.layers warnings.warn(f"Importing from {__name__} is deprecated, please import via timm.layers", FutureWarning) /opt/venv/lib/python3.10/site-packages/torch/functional.py:554: UserWarning: torch.meshgrid: in an upcoming release, it will be required to pass the indexing argument. (Triggered internally at /pytorch/aten/src/ATen/native/TensorShape.cpp:4317.) return _VF.meshgrid(tensors, **kwargs) # type: ignore[attr-defined] ══════════════════════════════════════════════════════════════════════  AgriFM × PASTIS — Fold 1 | 2026-04-17 03:23:59 ══════════════════════════════════════════════════════════════════════ Device : VRAM : 191.7 GB AMP : True Batch size : 16 Epochs : 100 LR : 3e-05 Weight decay : 0.05 Warmup iters : 500 Augmentation : True (flip+rotate) Class weights : True Work dir : ./work_dirs/fold1_v2 ──────────────────────────────────────────────────────────────────────  Building Datasets ────────────────────────────────────────────────────────────────────── PASTIS fold=1 split=train: 1279 patches (augment=True) PASTIS fold=1 split=val: 623 patches (augment=False) PASTIS fold=1 split=test: 531 patches (augment=False) Train: 1279 Val: 623 Test: 531 Train batches: 79 Val batches: 39 ──────────────────────────────────────────────────────────────────────  Building Model ────────────────────────────────────────────────────────────────────── Parameters: 196.2M ──────────────────────────────────────────────────────────────────────  Computing Class Weights ────────────────────────────────────────────────────────────────────── Class weights (fold=1, sampled 300 patches): Class Count Weight ────────────────────────────── ────────── ──────── Background 1978879 0.010 Meadow 1072189 0.018 Soft winter wheat 356296 0.053 Corn 524472 0.036 Winter barley 98052 0.193 Winter rapeseed 68225 0.277 Spring barley 28348 0.666 Sunflower 39655 0.476 Grapevine 69535 0.272 Beet 33292 0.567 Winter triticale 51347 0.368 Winter durum wheat 2307 8.188 Fruits vegetables flowers 18938 0.998 Potatoes 8376 2.255 Leguminous fodder 47354 0.399 Soybeans 88388 0.214 Orchard 8118 2.327 Mixed cereal 32384 0.583 Sorghum 17172 1.100 ══════════════════════════════════════════════════════════════════════  Training | 100 epochs | lr=3e-05 wd=0.05 fold=1 ══════════════════════════════════════════════════════════════════════ Training [█████████████████████████░░░░░░░░░░░░░░░] 63.3% iter 50/79 loss=0.4094 lr=3.90e-06 eta=0:00:12 [████████████████████████████████████████] 100.0% iter 79/79 loss=0.3903 lr=5.58e-06 eta=0:00:00 Done: avg_loss=0.3903 time=0:00:32 Validating... ──────────────────────────────────────────────────────────────────────  Epoch 001 / 100 | Elapsed: 0:00:42 ────────────────────────────────────────────────────────────────────── Metric Value ────────────────── ────────── Train Loss 0.3903 Val Loss 3.5759 ────────────────── ────────── OA   4.75% mIoU   1.06% mFscore   1.91% ★ mPrecision   2.96% mRecall   7.93% Kappa   2.28% ────────────────── ────────── Best mFscore 0.00% Val Time 8.4s ★ New best mFscore: 1.91% → saved best_model.pth Training [█████████████████████████░░░░░░░░░░░░░░░] 63.3% iter 50/79 loss=0.2860 lr=8.48e-06 eta=0:00:10 [████████████████████████████████████████] 100.0% iter 79/79 loss=0.2736 lr=1.02e-05 eta=0:00:00 Done: avg_loss=0.2736 time=0:00:27 Validating... ──────────────────────────────────────────────────────────────────────  Epoch 002 / 100 | Elapsed: 0:01:19 ────────────────────────────────────────────────────────────────────── Metric Value ────────────────── ────────── Train Loss 0.2736 Val Loss 3.3183 ────────────────── ────────── OA   1.78% mIoU   0.83% mFscore   1.60% mPrecision   7.91% mRecall   6.81% Kappa   -0.14% ────────────────── ────────── Best mFscore 1.91% Val Time 6.2s Training [█████████████████████████░░░░░░░░░░░░░░░] 63.3% iter 50/79 loss=0.2220 lr=1.31e-05 eta=0:00:10 [████████████████████████████████████████] 100.0% iter 79/79 loss=0.2256 lr=1.47e-05 eta=0:00:00 Done: avg_loss=0.2256 time=0:00:27 Validating... ──────────────────────────────────────────────────────────────────────  Epoch 003 / 100 | Elapsed: 0:01:55 ────────────────────────────────────────────────────────────────────── Metric Value ────────────────── ────────── Train Loss 0.2256 Val Loss 4.0984 ────────────────── ────────── OA   4.30% mIoU   0.62% mFscore   1.20% mPrecision   2.39% mRecall   6.68% Kappa   0.64% ────────────────── ────────── Best mFscore 1.91% Val Time 6.2s Training [█████████████████████████░░░░░░░░░░░░░░░] 63.3% iter 50/79 loss=0.1965 lr=1.76e-05 eta=0:00:10 [████████████████████████████████████████] 100.0% iter 79/79 loss=0.2048 lr=1.93e-05 eta=0:00:00 Done: avg_loss=0.2048 time=0:00:27 Validating... ──────────────────────────────────────────────────────────────────────  Epoch 004 / 100 | Elapsed: 0:02:31 ────────────────────────────────────────────────────────────────────── Metric Value ────────────────── ────────── Train Loss 0.2048 Val Loss 4.9570 ────────────────── ────────── OA   5.15% mIoU   0.66% mFscore   1.27% mPrecision   9.16% mRecall   5.90% Kappa   0.98% ────────────────── ────────── Best mFscore 1.91% Val Time 6.2s Training [█████████████████████████░░░░░░░░░░░░░░░] 63.3% iter 50/79 loss=0.1945 lr=2.22e-05 eta=0:00:10 [████████████████████████████████████████] 100.0% iter 79/79 loss=0.1903 lr=2.39e-05 eta=0:00:00 Done: avg_loss=0.1903 time=0:00:27 Validating... ──────────────────────────────────────────────────────────────────────  Epoch 005 / 100 | Elapsed: 0:03:07 ────────────────────────────────────────────────────────────────────── Metric Value ────────────────── ────────── Train Loss 0.1903 Val Loss 10.4665 ────────────────── ────────── OA   0.73% mIoU   0.39% mFscore   0.77% mPrecision   11.22% mRecall   5.50% Kappa   0.11%