Upload folder using huggingface_hub

2026-04-12-154000-flow-warp-ar-curriculum/config.json

@@ -0,0 +1 @@
+{"in_channels": 24, "channels": [32, 64, 128, 256], "context_len": 8, "model_class": "FlowWarpAttnUNet"}

2026-04-12-154000-flow-warp-ar-curriculum/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ac2e2d6dd460bcf391eb6a0d7e17bde8bbbf0b1d82eb6f7aa7402c70855c7f9f
+size 15231294

2026-04-12-154000-flow-warp-ar-curriculum/predict.py

@@ -0,0 +1,59 @@
+"""Inference for AR curriculum model + TTA."""
+import json
+import numpy as np
+import torch
+import sys
+sys.path.insert(0, "/home/coder/code")
+from flow_warp_attn_model import FlowWarpAttnUNet
+
+
+def load_model(model_dir: str):
+    with open(f"{model_dir}/config.json") as f:
+        config = json.load(f)
+    model = FlowWarpAttnUNet(in_channels=config["in_channels"], channels=config["channels"])
+    sd = torch.load(f"{model_dir}/model.pt", map_location="cpu", weights_only=True)
+    sd = {k: v.float() for k, v in sd.items()}
+    model.load_state_dict(sd)
+    model.eval()
+    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+    model = model.to(device)
+    return {"model": model, "device": device, "context_len": config["context_len"]}
+
+
+def _prepare_input(context_frames, context_len):
+    N = len(context_frames)
+    if N >= context_len:
+        frames = context_frames[-context_len:]
+    else:
+        pad = np.repeat(context_frames[:1], context_len - N, axis=0)
+        frames = np.concatenate([pad, context_frames], axis=0)
+    frames_f = frames.astype(np.float32) / 255.0
+    frames_f = np.transpose(frames_f, (0, 3, 1, 2))
+    context = frames_f.reshape(1, -1, 64, 64)
+    last_frame = frames_f[-1:]
+    return context, last_frame
+
+
+def predict_next_frame(model_dict, context_frames: np.ndarray) -> np.ndarray:
+    model = model_dict["model"]
+    device = model_dict["device"]
+    context_len = model_dict["context_len"]
+
+    ctx, last = _prepare_input(context_frames, context_len)
+    with torch.no_grad():
+        ctx_t = torch.from_numpy(ctx).to(device)
+        last_t = torch.from_numpy(last).to(device)
+        pred1, _ = model(ctx_t, last_t)
+
+    flipped_frames = context_frames[:, :, ::-1, :].copy()
+    ctx_f, last_f = _prepare_input(flipped_frames, context_len)
+    with torch.no_grad():
+        ctx_ft = torch.from_numpy(ctx_f).to(device)
+        last_ft = torch.from_numpy(last_f).to(device)
+        pred2, _ = model(ctx_ft, last_ft)
+        pred2 = pred2.flip(-1)
+
+    pred = (pred1 + pred2) / 2.0
+    pred_np = pred[0].cpu().numpy()
+    pred_np = np.transpose(pred_np, (1, 2, 0))
+    return (pred_np * 255.0).clip(0, 255).astype(np.uint8)

2026-04-12-154000-flow-warp-ar-curriculum/train.log

@@ -0,0 +1,254 @@
+[15:40:44] Model: FlowWarpAttnUNet, 7,596,742 params
+[15:40:44] === Phase 1: Single-step (80 epochs) ===
+[15:40:49] Train: 44392, Val: 5456
+[15:41:38] P1 Ep 1/80 | Train: 0.112908 | Val: 0.092624 | LR: 3.00e-04
+[15:41:38]   -> Saved P1 (val=0.092624)
+[15:42:16] P1 Ep 2/80 | Train: 0.086134 | Val: 0.081807 | LR: 3.00e-04
+[15:42:16]   -> Saved P1 (val=0.081807)
+[15:42:41] P1 Ep 3/80 | Train: 0.080104 | Val: 0.077833 | LR: 2.99e-04
+[15:42:41]   -> Saved P1 (val=0.077833)
+[15:43:27] P1 Ep 4/80 | Train: 0.076518 | Val: 0.075106 | LR: 2.98e-04
+[15:43:27]   -> Saved P1 (val=0.075106)
+[15:44:25] P1 Ep 5/80 | Train: 0.073637 | Val: 0.073171 | LR: 2.97e-04
+[15:44:25]   -> Saved P1 (val=0.073171)
+[15:45:17] P1 Ep 6/80 | Train: 0.071194 | Val: 0.071489 | LR: 2.96e-04
+[15:45:17]   -> Saved P1 (val=0.071489)
+[15:46:12] P1 Ep 7/80 | Train: 0.069269 | Val: 0.069629 | LR: 2.94e-04
+[15:46:12]   -> Saved P1 (val=0.069629)
+[15:47:07] P1 Ep 8/80 | Train: 0.067358 | Val: 0.068144 | LR: 2.93e-04
+[15:47:07]   -> Saved P1 (val=0.068144)
+[15:48:01] P1 Ep 9/80 | Train: 0.065801 | Val: 0.066217 | LR: 2.91e-04
+[15:48:01]   -> Saved P1 (val=0.066217)
+[15:48:48] P1 Ep 10/80 | Train: 0.064228 | Val: 0.064792 | LR: 2.89e-04
+[15:48:48]   -> Saved P1 (val=0.064792)
+[15:49:11] P1 Ep 11/80 | Train: 0.062906 | Val: 0.064103 | LR: 2.86e-04
+[15:49:11]   -> Saved P1 (val=0.064103)
+[15:49:47] P1 Ep 12/80 | Train: 0.061461 | Val: 0.063365 | LR: 2.84e-04
+[15:49:47]   -> Saved P1 (val=0.063365)
+[15:50:31] P1 Ep 13/80 | Train: 0.060517 | Val: 0.062029 | LR: 2.81e-04
+[15:50:31]   -> Saved P1 (val=0.062029)
+[15:51:25] P1 Ep 14/80 | Train: 0.059288 | Val: 0.062736 | LR: 2.78e-04
+[15:52:11] P1 Ep 15/80 | Train: 0.058430 | Val: 0.060402 | LR: 2.75e-04
+[15:52:11]   -> Saved P1 (val=0.060402)
+[15:52:35] P1 Ep 16/80 | Train: 0.057286 | Val: 0.058811 | LR: 2.71e-04
+[15:52:35]   -> Saved P1 (val=0.058811)
+[15:53:00] P1 Ep 17/80 | Train: 0.056255 | Val: 0.058606 | LR: 2.68e-04
+[15:53:00]   -> Saved P1 (val=0.058606)
+[15:53:22] P1 Ep 18/80 | Train: 0.055517 | Val: 0.058299 | LR: 2.64e-04
+[15:53:22]   -> Saved P1 (val=0.058299)
+[15:53:46] P1 Ep 19/80 | Train: 0.054703 | Val: 0.056898 | LR: 2.60e-04
+[15:53:46]   -> Saved P1 (val=0.056898)
+[15:54:15] P1 Ep 20/80 | Train: 0.053777 | Val: 0.057646 | LR: 2.56e-04
+[15:55:04] P1 Ep 21/80 | Train: 0.053153 | Val: 0.056199 | LR: 2.52e-04
+[15:55:04]   -> Saved P1 (val=0.056199)
+[15:55:41] P1 Ep 22/80 | Train: 0.052464 | Val: 0.055376 | LR: 2.48e-04
+[15:55:41]   -> Saved P1 (val=0.055376)
+[15:56:18] P1 Ep 23/80 | Train: 0.051747 | Val: 0.055485 | LR: 2.43e-04
+[15:56:50] P1 Ep 24/80 | Train: 0.050900 | Val: 0.055214 | LR: 2.38e-04
+[15:56:50]   -> Saved P1 (val=0.055214)
+[15:57:14] P1 Ep 25/80 | Train: 0.050435 | Val: 0.054996 | LR: 2.34e-04
+[15:57:14]   -> Saved P1 (val=0.054996)
+[15:57:43] P1 Ep 26/80 | Train: 0.049864 | Val: 0.055223 | LR: 2.29e-04
+[15:58:33] P1 Ep 27/80 | Train: 0.049200 | Val: 0.054005 | LR: 2.24e-04
+[15:58:33]   -> Saved P1 (val=0.054005)
+[15:59:19] P1 Ep 28/80 | Train: 0.048483 | Val: 0.053483 | LR: 2.18e-04
+[15:59:19]   -> Saved P1 (val=0.053483)
+[16:00:05] P1 Ep 29/80 | Train: 0.047984 | Val: 0.052937 | LR: 2.13e-04
+[16:00:05]   -> Saved P1 (val=0.052937)
+[16:00:39] P1 Ep 30/80 | Train: 0.047200 | Val: 0.052694 | LR: 2.08e-04
+[16:00:39]   -> Saved P1 (val=0.052694)
+[16:01:11] P1 Ep 31/80 | Train: 0.046624 | Val: 0.053359 | LR: 2.02e-04
+[16:02:07] P1 Ep 32/80 | Train: 0.045999 | Val: 0.052335 | LR: 1.97e-04
+[16:02:07]   -> Saved P1 (val=0.052335)
+[16:03:05] P1 Ep 33/80 | Train: 0.045480 | Val: 0.052796 | LR: 1.91e-04
+[16:03:58] P1 Ep 34/80 | Train: 0.044809 | Val: 0.052662 | LR: 1.85e-04
+[16:04:27] P1 Ep 35/80 | Train: 0.044315 | Val: 0.051852 | LR: 1.80e-04
+[16:04:27]   -> Saved P1 (val=0.051852)
+[16:05:16] P1 Ep 36/80 | Train: 0.043599 | Val: 0.051811 | LR: 1.74e-04
+[16:05:16]   -> Saved P1 (val=0.051811)
+[16:06:03] P1 Ep 37/80 | Train: 0.043138 | Val: 0.051202 | LR: 1.68e-04
+[16:06:03]   -> Saved P1 (val=0.051202)
+[16:06:41] P1 Ep 38/80 | Train: 0.042539 | Val: 0.050464 | LR: 1.62e-04
+[16:06:41]   -> Saved P1 (val=0.050464)
+[16:07:38] P1 Ep 39/80 | Train: 0.042066 | Val: 0.051059 | LR: 1.56e-04
+[16:08:34] P1 Ep 40/80 | Train: 0.041315 | Val: 0.050918 | LR: 1.51e-04
+[16:09:27] P1 Ep 41/80 | Train: 0.040930 | Val: 0.050821 | LR: 1.45e-04
+[16:10:15] P1 Ep 42/80 | Train: 0.040304 | Val: 0.050680 | LR: 1.39e-04
+[16:11:10] P1 Ep 43/80 | Train: 0.039895 | Val: 0.050563 | LR: 1.33e-04
+[16:12:05] P1 Ep 44/80 | Train: 0.039291 | Val: 0.050782 | LR: 1.27e-04
+[16:13:00] P1 Ep 45/80 | Train: 0.038875 | Val: 0.050635 | LR: 1.21e-04
+[16:13:54] P1 Ep 46/80 | Train: 0.038366 | Val: 0.050127 | LR: 1.16e-04
+[16:13:54]   -> Saved P1 (val=0.050127)
+[16:14:29] P1 Ep 47/80 | Train: 0.037889 | Val: 0.050041 | LR: 1.10e-04
+[16:14:29]   -> Saved P1 (val=0.050041)
+[16:15:23] P1 Ep 48/80 | Train: 0.037326 | Val: 0.050060 | LR: 1.04e-04
+[16:16:12] P1 Ep 49/80 | Train: 0.036947 | Val: 0.050365 | LR: 9.88e-05
+[16:17:00] P1 Ep 50/80 | Train: 0.036355 | Val: 0.049527 | LR: 9.33e-05
+[16:17:00]   -> Saved P1 (val=0.049527)
+[16:17:57] P1 Ep 51/80 | Train: 0.036024 | Val: 0.049895 | LR: 8.79e-05
+[16:18:50] P1 Ep 52/80 | Train: 0.035437 | Val: 0.049741 | LR: 8.26e-05
+[16:19:35] P1 Ep 53/80 | Train: 0.034996 | Val: 0.049694 | LR: 7.75e-05
+[16:20:19] P1 Ep 54/80 | Train: 0.034717 | Val: 0.049697 | LR: 7.24e-05
+[16:21:00] P1 Ep 55/80 | Train: 0.034225 | Val: 0.049615 | LR: 6.74e-05
+[16:21:55] P1 Ep 56/80 | Train: 0.033776 | Val: 0.049514 | LR: 6.26e-05
+[16:21:55]   -> Saved P1 (val=0.049514)
+[16:22:42] P1 Ep 57/80 | Train: 0.033460 | Val: 0.049664 | LR: 5.79e-05
+[16:23:34] P1 Ep 58/80 | Train: 0.033085 | Val: 0.049812 | LR: 5.34e-05
+[16:24:24] P1 Ep 59/80 | Train: 0.032784 | Val: 0.049797 | LR: 4.90e-05
+[16:24:49] P1 Ep 60/80 | Train: 0.032414 | Val: 0.049671 | LR: 4.48e-05
+[16:25:33] P1 Ep 61/80 | Train: 0.032062 | Val: 0.049591 | LR: 4.07e-05
+[16:26:25] P1 Ep 62/80 | Train: 0.031821 | Val: 0.050111 | LR: 3.68e-05
+[16:27:10] P1 Ep 63/80 | Train: 0.031511 | Val: 0.049890 | LR: 3.31e-05
+[16:27:55] P1 Ep 64/80 | Train: 0.031259 | Val: 0.050195 | LR: 2.96e-05
+[16:28:34] P1 Ep 65/80 | Train: 0.031016 | Val: 0.049906 | LR: 2.62e-05
+[16:29:32] P1 Ep 66/80 | Train: 0.030782 | Val: 0.049883 | LR: 2.30e-05
+[16:30:27] P1 Ep 67/80 | Train: 0.030587 | Val: 0.049873 | LR: 2.01e-05
+[16:31:13] P1 Ep 68/80 | Train: 0.030401 | Val: 0.049917 | LR: 1.73e-05
+[16:31:40] P1 Ep 69/80 | Train: 0.030268 | Val: 0.049848 | LR: 1.47e-05
+[16:32:31] P1 Ep 70/80 | Train: 0.030057 | Val: 0.050109 | LR: 1.24e-05
+[16:33:27] P1 Ep 71/80 | Train: 0.029958 | Val: 0.050045 | LR: 1.02e-05
+[16:34:17] P1 Ep 72/80 | Train: 0.029819 | Val: 0.050060 | LR: 8.32e-06
+[16:34:49] P1 Ep 73/80 | Train: 0.029710 | Val: 0.050139 | LR: 6.61e-06
+[16:35:11] P1 Ep 74/80 | Train: 0.029600 | Val: 0.050150 | LR: 5.13e-06
+[16:35:48] P1 Ep 75/80 | Train: 0.029548 | Val: 0.050126 | LR: 3.87e-06
+[16:36:41] P1 Ep 76/80 | Train: 0.029521 | Val: 0.050131 | LR: 2.84e-06
+[16:37:25] P1 Ep 77/80 | Train: 0.029491 | Val: 0.050194 | LR: 2.04e-06
+[16:38:05] P1 Ep 78/80 | Train: 0.029450 | Val: 0.050196 | LR: 1.46e-06
+[16:38:35] P1 Ep 79/80 | Train: 0.029430 | Val: 0.050184 | LR: 1.12e-06
+[16:39:27] P1 Ep 80/80 | Train: 0.029404 | Val: 0.050189 | LR: 1.00e-06
+[16:39:27] === Phase 2: 2-step AR (40 epochs, noise=0.0078) ===
+[16:39:32] Train: 44034, Val: 5412
+[16:40:26] P2 Ep 1/40 | Train: 0.047285 | Val: 0.070376 | LR: 9.98e-05
+[16:40:26]   -> Saved P2 (val=0.070376)
+[16:41:17] P2 Ep 2/40 | Train: 0.046644 | Val: 0.069826 | LR: 9.94e-05
+[16:41:17]   -> Saved P2 (val=0.069826)
+[16:42:04] P2 Ep 3/40 | Train: 0.046104 | Val: 0.069656 | LR: 9.86e-05
+[16:42:04]   -> Saved P2 (val=0.069656)
+[16:43:01] P2 Ep 4/40 | Train: 0.045620 | Val: 0.070126 | LR: 9.76e-05
+[16:43:58] P2 Ep 5/40 | Train: 0.045143 | Val: 0.069811 | LR: 9.62e-05
+[16:44:42] P2 Ep 6/40 | Train: 0.044883 | Val: 0.069925 | LR: 9.46e-05
+[16:45:26] P2 Ep 7/40 | Train: 0.044342 | Val: 0.070318 | LR: 9.27e-05
+[16:46:14] P2 Ep 8/40 | Train: 0.043952 | Val: 0.070639 | LR: 9.05e-05
+[16:46:59] P2 Ep 9/40 | Train: 0.043475 | Val: 0.070085 | LR: 8.81e-05
+[16:47:45] P2 Ep 10/40 | Train: 0.043026 | Val: 0.069909 | LR: 8.55e-05
+[16:48:27] P2 Ep 11/40 | Train: 0.042577 | Val: 0.070436 | LR: 8.26e-05
+[16:49:15] P2 Ep 12/40 | Train: 0.042101 | Val: 0.071131 | LR: 7.96e-05
+[16:50:02] P2 Ep 13/40 | Train: 0.041770 | Val: 0.070430 | LR: 7.64e-05
+[16:50:47] P2 Ep 14/40 | Train: 0.041258 | Val: 0.070631 | LR: 7.30e-05
+[16:51:29] P2 Ep 15/40 | Train: 0.040938 | Val: 0.071101 | LR: 6.94e-05
+[16:52:15] P2 Ep 16/40 | Train: 0.040437 | Val: 0.070787 | LR: 6.58e-05
+[16:52:59] P2 Ep 17/40 | Train: 0.040043 | Val: 0.071084 | LR: 6.21e-05
+[16:53:41] P2 Ep 18/40 | Train: 0.039619 | Val: 0.070765 | LR: 5.82e-05
+[16:54:24] P2 Ep 19/40 | Train: 0.039209 | Val: 0.070925 | LR: 5.44e-05
+[16:55:05] P2 Ep 20/40 | Train: 0.038831 | Val: 0.070665 | LR: 5.05e-05
+[16:55:48] P2 Ep 21/40 | Train: 0.038462 | Val: 0.071706 | LR: 4.66e-05
+[16:56:29] P2 Ep 22/40 | Train: 0.038149 | Val: 0.070848 | LR: 4.28e-05
+[16:57:10] P2 Ep 23/40 | Train: 0.037821 | Val: 0.071886 | LR: 3.89e-05
+[16:57:48] P2 Ep 24/40 | Train: 0.037490 | Val: 0.071343 | LR: 3.52e-05
+[16:58:28] P2 Ep 25/40 | Train: 0.037162 | Val: 0.071360 | LR: 3.16e-05
+[16:59:07] P2 Ep 26/40 | Train: 0.036869 | Val: 0.071390 | LR: 2.80e-05
+[16:59:46] P2 Ep 27/40 | Train: 0.036626 | Val: 0.071814 | LR: 2.46e-05
+[17:00:25] P2 Ep 28/40 | Train: 0.036370 | Val: 0.071899 | LR: 2.14e-05
+[17:01:08] P2 Ep 29/40 | Train: 0.036102 | Val: 0.072189 | LR: 1.84e-05
+[17:01:50] P2 Ep 30/40 | Train: 0.035936 | Val: 0.071791 | LR: 1.55e-05
+[17:02:28] P2 Ep 31/40 | Train: 0.035759 | Val: 0.072592 | LR: 1.29e-05
+[17:03:07] P2 Ep 32/40 | Train: 0.035663 | Val: 0.072196 | LR: 1.05e-05
+[17:04:04] P2 Ep 33/40 | Train: 0.035482 | Val: 0.072056 | LR: 8.29e-06
+[17:04:50] P2 Ep 34/40 | Train: 0.035372 | Val: 0.072318 | LR: 6.40e-06
+[17:05:37] P2 Ep 35/40 | Train: 0.035295 | Val: 0.072516 | LR: 4.77e-06
+[17:06:23] P2 Ep 36/40 | Train: 0.035200 | Val: 0.072428 | LR: 3.42e-06
+[17:07:03] P2 Ep 37/40 | Train: 0.035132 | Val: 0.072473 | LR: 2.37e-06
+[17:08:03] P2 Ep 38/40 | Train: 0.035041 | Val: 0.072513 | LR: 1.61e-06
+[17:08:52] P2 Ep 39/40 | Train: 0.035027 | Val: 0.072532 | LR: 1.15e-06
+[17:09:36] P2 Ep 40/40 | Train: 0.035019 | Val: 0.072462 | LR: 1.00e-06
+[17:09:36] === Phase 3: 4-step AR (40 epochs, noise=0.0039) ===
+[17:09:41] Train: 10919, Val: 1342
+[17:10:22] P3 Ep 1/40 | Train: 0.066989 | Val: 0.105359 | LR: 4.99e-05
+[17:10:22]   -> Saved P3 (val=0.105359)
+[17:11:01] P3 Ep 2/40 | Train: 0.065233 | Val: 0.105719 | LR: 4.97e-05
+[17:11:41] P3 Ep 3/40 | Train: 0.064332 | Val: 0.105288 | LR: 4.93e-05
+[17:11:41]   -> Saved P3 (val=0.105288)
+[17:12:20] P3 Ep 4/40 | Train: 0.063536 | Val: 0.106669 | LR: 4.88e-05
+[17:12:57] P3 Ep 5/40 | Train: 0.063039 | Val: 0.106552 | LR: 4.81e-05
+[17:13:32] P3 Ep 6/40 | Train: 0.062253 | Val: 0.106537 | LR: 4.73e-05
+[17:14:09] P3 Ep 7/40 | Train: 0.061609 | Val: 0.107229 | LR: 4.64e-05
+[17:14:47] P3 Ep 8/40 | Train: 0.061302 | Val: 0.106256 | LR: 4.53e-05
+[17:15:27] P3 Ep 9/40 | Train: 0.060679 | Val: 0.108110 | LR: 4.41e-05
+[17:16:06] P3 Ep 10/40 | Train: 0.060154 | Val: 0.109534 | LR: 4.28e-05
+[17:16:44] P3 Ep 11/40 | Train: 0.059723 | Val: 0.108979 | LR: 4.14e-05
+[17:17:18] P3 Ep 12/40 | Train: 0.059178 | Val: 0.107284 | LR: 3.99e-05
+[17:17:51] P3 Ep 13/40 | Train: 0.058841 | Val: 0.108774 | LR: 3.83e-05
+[17:18:25] P3 Ep 14/40 | Train: 0.058297 | Val: 0.109313 | LR: 3.66e-05
+[17:18:59] P3 Ep 15/40 | Train: 0.057755 | Val: 0.108477 | LR: 3.49e-05
+[17:19:35] P3 Ep 16/40 | Train: 0.057522 | Val: 0.108420 | LR: 3.31e-05
+[17:20:11] P3 Ep 17/40 | Train: 0.057237 | Val: 0.110192 | LR: 3.12e-05
+[17:20:48] P3 Ep 18/40 | Train: 0.056578 | Val: 0.108483 | LR: 2.93e-05
+[17:21:27] P3 Ep 19/40 | Train: 0.056250 | Val: 0.108837 | LR: 2.74e-05
+[17:22:06] P3 Ep 20/40 | Train: 0.055933 | Val: 0.109871 | LR: 2.55e-05
+[17:22:44] P3 Ep 21/40 | Train: 0.055610 | Val: 0.109395 | LR: 2.36e-05
+[17:23:21] P3 Ep 22/40 | Train: 0.055233 | Val: 0.109832 | LR: 2.17e-05
+[17:23:58] P3 Ep 23/40 | Train: 0.054883 | Val: 0.110014 | LR: 1.98e-05
+[17:24:36] P3 Ep 24/40 | Train: 0.054613 | Val: 0.109150 | LR: 1.79e-05
+[17:25:15] P3 Ep 25/40 | Train: 0.054338 | Val: 0.110191 | LR: 1.61e-05
+[17:25:54] P3 Ep 26/40 | Train: 0.054093 | Val: 0.109644 | LR: 1.44e-05
+[17:26:34] P3 Ep 27/40 | Train: 0.053798 | Val: 0.110244 | LR: 1.27e-05
+[17:27:13] P3 Ep 28/40 | Train: 0.053642 | Val: 0.110176 | LR: 1.11e-05
+[17:27:53] P3 Ep 29/40 | Train: 0.053352 | Val: 0.110111 | LR: 9.59e-06
+[17:28:33] P3 Ep 30/40 | Train: 0.053254 | Val: 0.110057 | LR: 8.18e-06
+[17:29:13] P3 Ep 31/40 | Train: 0.053071 | Val: 0.110303 | LR: 6.87e-06
+[17:29:50] P3 Ep 32/40 | Train: 0.052807 | Val: 0.110612 | LR: 5.68e-06
+[17:30:25] P3 Ep 33/40 | Train: 0.052767 | Val: 0.110462 | LR: 4.61e-06
+[17:31:03] P3 Ep 34/40 | Train: 0.052605 | Val: 0.110954 | LR: 3.67e-06
+[17:31:41] P3 Ep 35/40 | Train: 0.052573 | Val: 0.110939 | LR: 2.86e-06
+[17:32:19] P3 Ep 36/40 | Train: 0.052437 | Val: 0.110889 | LR: 2.20e-06
+[17:32:57] P3 Ep 37/40 | Train: 0.052463 | Val: 0.111071 | LR: 1.68e-06
+[17:33:35] P3 Ep 38/40 | Train: 0.052356 | Val: 0.110856 | LR: 1.30e-06
+[17:34:13] P3 Ep 39/40 | Train: 0.052375 | Val: 0.111090 | LR: 1.08e-06
+[17:34:52] P3 Ep 40/40 | Train: 0.052406 | Val: 0.111096 | LR: 1.00e-06
+[17:34:52] === Phase 4: 8-step AR (40 epochs, noise=0.0000) ===
+[17:34:57] Train: 2685, Val: 330
+[17:35:27] P4 Ep 1/40 | Train: 0.101111 | Val: 0.162571 | LR: 2.00e-05
+[17:35:27]   -> Saved P4 (val=0.162571)
+[17:35:57] P4 Ep 2/40 | Train: 0.098553 | Val: 0.160688 | LR: 1.99e-05
+[17:35:58]   -> Saved P4 (val=0.160688)
+[17:36:27] P4 Ep 3/40 | Train: 0.097715 | Val: 0.159298 | LR: 1.97e-05
+[17:36:27]   -> Saved P4 (val=0.159298)
+[17:36:57] P4 Ep 4/40 | Train: 0.096368 | Val: 0.159610 | LR: 1.95e-05
+[17:37:27] P4 Ep 5/40 | Train: 0.095401 | Val: 0.159243 | LR: 1.93e-05
+[17:37:27]   -> Saved P4 (val=0.159243)
+[17:37:58] P4 Ep 6/40 | Train: 0.094479 | Val: 0.160074 | LR: 1.90e-05
+[17:38:27] P4 Ep 7/40 | Train: 0.093345 | Val: 0.160185 | LR: 1.86e-05
+[17:38:57] P4 Ep 8/40 | Train: 0.092778 | Val: 0.163288 | LR: 1.82e-05
+[17:39:27] P4 Ep 9/40 | Train: 0.091566 | Val: 0.163134 | LR: 1.77e-05
+[17:39:57] P4 Ep 10/40 | Train: 0.091514 | Val: 0.161680 | LR: 1.72e-05
+[17:40:27] P4 Ep 11/40 | Train: 0.090770 | Val: 0.162345 | LR: 1.67e-05
+[17:40:57] P4 Ep 12/40 | Train: 0.090370 | Val: 0.163773 | LR: 1.61e-05
+[17:41:27] P4 Ep 13/40 | Train: 0.089234 | Val: 0.161309 | LR: 1.55e-05
+[17:41:57] P4 Ep 14/40 | Train: 0.088393 | Val: 0.164177 | LR: 1.48e-05
+[17:42:27] P4 Ep 15/40 | Train: 0.087783 | Val: 0.161810 | LR: 1.41e-05
+[17:42:57] P4 Ep 16/40 | Train: 0.087447 | Val: 0.164077 | LR: 1.34e-05
+[17:43:26] P4 Ep 17/40 | Train: 0.087248 | Val: 0.162098 | LR: 1.27e-05
+[17:43:56] P4 Ep 18/40 | Train: 0.086734 | Val: 0.163517 | LR: 1.20e-05
+[17:44:26] P4 Ep 19/40 | Train: 0.086703 | Val: 0.163718 | LR: 1.12e-05
+[17:44:56] P4 Ep 20/40 | Train: 0.085615 | Val: 0.162694 | LR: 1.05e-05
+[17:45:26] P4 Ep 21/40 | Train: 0.085481 | Val: 0.163687 | LR: 9.75e-06
+[17:45:55] P4 Ep 22/40 | Train: 0.085017 | Val: 0.164132 | LR: 9.01e-06
+[17:46:25] P4 Ep 23/40 | Train: 0.084124 | Val: 0.164077 | LR: 8.28e-06
+[17:46:56] P4 Ep 24/40 | Train: 0.084426 | Val: 0.164232 | LR: 7.56e-06
+[17:47:26] P4 Ep 25/40 | Train: 0.084260 | Val: 0.164044 | LR: 6.86e-06
+[17:47:56] P4 Ep 26/40 | Train: 0.083558 | Val: 0.164989 | LR: 6.19e-06
+[17:48:26] P4 Ep 27/40 | Train: 0.083491 | Val: 0.165989 | LR: 5.54e-06
+[17:48:56] P4 Ep 28/40 | Train: 0.082750 | Val: 0.165438 | LR: 4.92e-06
+[17:49:26] P4 Ep 29/40 | Train: 0.082852 | Val: 0.165827 | LR: 4.33e-06
+[17:49:56] P4 Ep 30/40 | Train: 0.082666 | Val: 0.166003 | LR: 3.78e-06
+[17:50:26] P4 Ep 31/40 | Train: 0.082712 | Val: 0.165605 | LR: 3.28e-06
+[17:50:56] P4 Ep 32/40 | Train: 0.082512 | Val: 0.166602 | LR: 2.81e-06
+[17:51:26] P4 Ep 33/40 | Train: 0.081745 | Val: 0.166517 | LR: 2.40e-06
+[17:51:55] P4 Ep 34/40 | Train: 0.082233 | Val: 0.166963 | LR: 2.04e-06
+[17:52:25] P4 Ep 35/40 | Train: 0.081910 | Val: 0.167685 | LR: 1.72e-06
+[17:52:55] P4 Ep 36/40 | Train: 0.081687 | Val: 0.167599 | LR: 1.46e-06
+[17:53:33] P4 Ep 37/40 | Train: 0.081693 | Val: 0.166911 | LR: 1.26e-06
+[17:54:05] P4 Ep 38/40 | Train: 0.081608 | Val: 0.167503 | LR: 1.12e-06
+[17:54:35] P4 Ep 39/40 | Train: 0.081655 | Val: 0.167355 | LR: 1.03e-06
+[17:55:04] P4 Ep 40/40 | Train: 0.081653 | Val: 0.167608 | LR: 1.00e-06
+[17:55:04] Training complete. Now score each phase to find best.