Hugging Face's logo

yuriyvnv
/
3layers_NO_Alignment

Model card Files
xet
Community
yuriyvnv commited on
Commit
9206b9f
·
verified ·
1 Parent(s): 7221070

Upload folder using huggingface_hub

Browse files
Files changed (40) hide show
  1. best_model_gap.pt +3 -0
  2. best_model_loss.pt +3 -0
  3. checkpoint_epoch_15.pt +3 -0
  4. checkpoint_epoch_30.pt +3 -0
  5. clean_corrupt_progress.png +0 -0
  6. final_model.pt +3 -0
  7. similarity_dist_epoch_10.png +0 -0
  8. similarity_dist_epoch_12.png +0 -0
  9. similarity_dist_epoch_14.png +0 -0
  10. similarity_dist_epoch_16.png +0 -0
  11. similarity_dist_epoch_18.png +0 -0
  12. similarity_dist_epoch_2.png +0 -0
  13. similarity_dist_epoch_20.png +0 -0
  14. similarity_dist_epoch_22.png +0 -0
  15. similarity_dist_epoch_24.png +0 -0
  16. similarity_dist_epoch_26.png +0 -0
  17. similarity_dist_epoch_28.png +0 -0
  18. similarity_dist_epoch_30.png +0 -0
  19. similarity_dist_epoch_4.png +0 -0
  20. similarity_dist_epoch_6.png +0 -0
  21. similarity_dist_epoch_8.png +0 -0
  22. test_metrics.json +26 -0
  23. test_similarity_dist_best_gap.png +0 -0
  24. test_similarity_dist_best_loss.png +0 -0
  25. training.log +448 -0
  26. validation_alignment_epoch_10.png +0 -0
  27. validation_alignment_epoch_12.png +0 -0
  28. validation_alignment_epoch_14.png +0 -0
  29. validation_alignment_epoch_16.png +0 -0
  30. validation_alignment_epoch_18.png +0 -0
  31. validation_alignment_epoch_2.png +0 -0
  32. validation_alignment_epoch_20.png +0 -0
  33. validation_alignment_epoch_22.png +0 -0
  34. validation_alignment_epoch_24.png +0 -0
  35. validation_alignment_epoch_26.png +0 -0
  36. validation_alignment_epoch_28.png +0 -0
  37. validation_alignment_epoch_30.png +0 -0
  38. validation_alignment_epoch_4.png +0 -0
  39. validation_alignment_epoch_6.png +0 -0
  40. validation_alignment_epoch_8.png +0 -0
best_model_gap.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7ca4ca0982cffd8936eb2c8670ca391948f6bded8abafeeceaf1e26a51317436
3
+ size 5787099300
best_model_loss.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2b50c7a3098be0e949bfcbecdd9f147fd2df66d196f81f01c8c8cf1a6c9ad816
3
+ size 5787100783
checkpoint_epoch_15.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8253cdd4676d48a619e67842e3b3e7f7711c6371b8b8a6e85a3bcefd1b1f64b2
3
+ size 5787106715
checkpoint_epoch_30.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ca418d80f4f35fd2a2b86f950552a86af6769f06857f7bb3433f3270e038a2f2
3
+ size 5787106715
clean_corrupt_progress.png ADDED
final_model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:639fca6990a876a464735b7c71a55a060548c5ff72220c8a393a9bcce034ce93
3
+ size 5787094851
similarity_dist_epoch_10.png ADDED
similarity_dist_epoch_12.png ADDED
similarity_dist_epoch_14.png ADDED
similarity_dist_epoch_16.png ADDED
similarity_dist_epoch_18.png ADDED
similarity_dist_epoch_2.png ADDED
similarity_dist_epoch_20.png ADDED
similarity_dist_epoch_22.png ADDED
similarity_dist_epoch_24.png ADDED
similarity_dist_epoch_26.png ADDED
similarity_dist_epoch_28.png ADDED
similarity_dist_epoch_30.png ADDED
similarity_dist_epoch_4.png ADDED
similarity_dist_epoch_6.png ADDED
similarity_dist_epoch_8.png ADDED
test_metrics.json ADDED
@@ -0,0 +1,26 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_loss_model": {
3
+ "loss": 0.1638984185715997,
4
+ "avg_similarity": 0.6754336357116699,
5
+ "median_similarity": 0.9327878952026367,
6
+ "std_similarity": 0.40030691027641296,
7
+ "clean_similarity": 0.6754336357116699,
8
+ "corrupt_similarity": 0.25681251287460327,
9
+ "similarity_gap": 0.41862112283706665,
10
+ "avg_pos_alignment": 0.0,
11
+ "avg_neg_alignment": 0.0,
12
+ "alignment_gap": 0.0
13
+ },
14
+ "best_gap_model": {
15
+ "loss": 0.1585616613090435,
16
+ "avg_similarity": 0.6903528571128845,
17
+ "median_similarity": 0.9347754716873169,
18
+ "std_similarity": 0.38960909843444824,
19
+ "clean_similarity": 0.6903528571128845,
20
+ "corrupt_similarity": 0.23637232184410095,
21
+ "similarity_gap": 0.45398053526878357,
22
+ "avg_pos_alignment": 0.0,
23
+ "avg_neg_alignment": 0.0,
24
+ "alignment_gap": 0.0
25
+ }
26
+ }
test_similarity_dist_best_gap.png ADDED
test_similarity_dist_best_loss.png ADDED
training.log ADDED
@@ -0,0 +1,448 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2025-07-09 10:02:50,441 - INFO - Training with parameters:
2
+ 2025-07-09 10:02:50,442 - INFO - Text model: sentence-transformers/paraphrase-multilingual-mpnet-base-v2
3
+ 2025-07-09 10:02:50,442 - INFO - Audio model: facebook/w2v-bert-2.0
4
+ 2025-07-09 10:02:50,442 - INFO - Freeze encoders: partial
5
+ 2025-07-09 10:02:50,442 - INFO - Text layers to unfreeze: 3
6
+ 2025-07-09 10:02:50,442 - INFO - Audio layers to unfreeze: 3
7
+ 2025-07-09 10:02:50,442 - INFO - Use cross-modal attention: False
8
+ 2025-07-09 10:02:50,442 - INFO - Use attentive pooling: False
9
+ 2025-07-09 10:02:50,442 - INFO - Use word-level alignment: False
10
+ 2025-07-09 10:02:50,442 - INFO - Batch size: 48
11
+ 2025-07-09 10:02:50,442 - INFO - Gradient accumulation steps: 15
12
+ 2025-07-09 10:02:50,442 - INFO - Effective batch size: 720
13
+ 2025-07-09 10:02:50,442 - INFO - Mixed precision training: False
14
+ 2025-07-09 10:02:50,442 - INFO - Learning rate: 0.0008
15
+ 2025-07-09 10:02:50,442 - INFO - Temperature: 0.1
16
+ 2025-07-09 10:02:50,442 - INFO - Projection dimension: 768
17
+ 2025-07-09 10:02:50,442 - INFO - Training samples: 21968
18
+ 2025-07-09 10:02:50,442 - INFO - Validation samples: 9464
19
+ 2025-07-09 10:02:50,442 - INFO - Test samples: 9467
20
+ 2025-07-09 10:02:50,442 - INFO - Max audio length: 480000 samples (30.00 seconds at 16kHz)
21
+ 2025-07-09 10:02:50,442 - INFO - Loading tokenizer and feature extractor...
22
+ 2025-07-09 10:02:51,406 - INFO - Feature extractor output keys: ['input_features', 'attention_mask']
23
+ 2025-07-09 10:02:51,406 - INFO - Creating datasets...
24
+ 2025-07-09 10:02:51,406 - INFO - Feature extractor output keys: ['input_features', 'attention_mask']
25
+ 2025-07-09 10:02:51,407 - INFO - Feature extractor output keys: ['input_features', 'attention_mask']
26
+ 2025-07-09 10:02:51,407 - INFO - Feature extractor output keys: ['input_features', 'attention_mask']
27
+ 2025-07-09 10:02:51,407 - INFO - Creating data loaders...
28
+ 2025-07-09 10:02:51,407 - INFO - Checking a sample batch...
29
+ 2025-07-09 10:05:56,755 - INFO - Training with parameters:
30
+ 2025-07-09 10:05:56,755 - INFO - Text model: sentence-transformers/paraphrase-multilingual-mpnet-base-v2
31
+ 2025-07-09 10:05:56,755 - INFO - Audio model: facebook/w2v-bert-2.0
32
+ 2025-07-09 10:05:56,755 - INFO - Freeze encoders: partial
33
+ 2025-07-09 10:05:56,755 - INFO - Text layers to unfreeze: 3
34
+ 2025-07-09 10:05:56,755 - INFO - Audio layers to unfreeze: 3
35
+ 2025-07-09 10:05:56,755 - INFO - Use cross-modal attention: False
36
+ 2025-07-09 10:05:56,755 - INFO - Use attentive pooling: False
37
+ 2025-07-09 10:05:56,755 - INFO - Use word-level alignment: False
38
+ 2025-07-09 10:05:56,755 - INFO - Batch size: 48
39
+ 2025-07-09 10:05:56,755 - INFO - Gradient accumulation steps: 15
40
+ 2025-07-09 10:05:56,755 - INFO - Effective batch size: 720
41
+ 2025-07-09 10:05:56,755 - INFO - Mixed precision training: False
42
+ 2025-07-09 10:05:56,755 - INFO - Learning rate: 0.0008
43
+ 2025-07-09 10:05:56,755 - INFO - Temperature: 0.1
44
+ 2025-07-09 10:05:56,755 - INFO - Projection dimension: 768
45
+ 2025-07-09 10:05:56,755 - INFO - Training samples: 21968
46
+ 2025-07-09 10:05:56,755 - INFO - Validation samples: 9464
47
+ 2025-07-09 10:05:56,755 - INFO - Test samples: 9467
48
+ 2025-07-09 10:05:56,755 - INFO - Max audio length: 480000 samples (30.00 seconds at 16kHz)
49
+ 2025-07-09 10:05:56,755 - INFO - Loading tokenizer and feature extractor...
50
+ 2025-07-09 10:05:57,689 - INFO - Feature extractor output keys: ['input_features', 'attention_mask']
51
+ 2025-07-09 10:05:57,689 - INFO - Creating datasets...
52
+ 2025-07-09 10:05:57,689 - INFO - Feature extractor output keys: ['input_features', 'attention_mask']
53
+ 2025-07-09 10:05:57,690 - INFO - Feature extractor output keys: ['input_features', 'attention_mask']
54
+ 2025-07-09 10:05:57,690 - INFO - Feature extractor output keys: ['input_features', 'attention_mask']
55
+ 2025-07-09 10:05:57,690 - INFO - Creating data loaders...
56
+ 2025-07-09 10:05:57,690 - INFO - Checking a sample batch...
57
+ 2025-07-09 10:06:15,500 - INFO - input_ids_pos: torch.Size([48, 128])
58
+ 2025-07-09 10:06:15,500 - INFO - attention_mask_pos: torch.Size([48, 128])
59
+ 2025-07-09 10:06:15,500 - INFO - input_ids_neg: torch.Size([48, 128])
60
+ 2025-07-09 10:06:15,500 - INFO - attention_mask_neg: torch.Size([48, 128])
61
+ 2025-07-09 10:06:15,500 - INFO - input_values: torch.Size([48, 473, 160])
62
+ 2025-07-09 10:06:15,500 - INFO - attention_mask_audio: torch.Size([48, 473])
63
+ 2025-07-09 10:06:15,500 - INFO - is_corrupted: torch.Size([48])
64
+ 2025-07-09 10:06:15,500 - INFO - Initializing model...
65
+ 2025-07-09 10:06:16,250 - INFO - Text encoder hidden dim: 768
66
+ 2025-07-09 10:06:16,250 - INFO - Audio encoder hidden dim: 1024
67
+ 2025-07-09 10:06:16,250 - INFO - Partial freezing: unfreezing last 3 text layers and 3 audio layers
68
+ 2025-07-09 10:06:16,250 - INFO - Unfreezing text encoder layer 9
69
+ 2025-07-09 10:06:16,250 - INFO - Unfreezing text encoder layer 10
70
+ 2025-07-09 10:06:16,250 - INFO - Unfreezing text encoder layer 11
71
+ 2025-07-09 10:06:16,251 - INFO - Unfreezing audio encoder layer 21
72
+ 2025-07-09 10:06:16,251 - INFO - Unfreezing audio encoder layer 22
73
+ 2025-07-09 10:06:16,251 - INFO - Unfreezing audio encoder layer 23
74
+ 2025-07-09 10:06:16,281 - INFO - Model initialized with 292,079,360 trainable parameters out of 863,656,256 total
75
+ 2025-07-09 10:06:17,120 - INFO - Using discriminative learning rates: encoder_lr=4e-05, main_lr=0.0008
76
+ 2025-07-09 10:06:17,120 - INFO - Encoder parameters: 156, Non-encoder parameters: 12
77
+ 2025-07-09 10:06:17,120 - INFO - Scheduler setup:
78
+ 2025-07-09 10:06:17,121 - INFO - Batches per epoch: 457
79
+ 2025-07-09 10:06:17,121 - INFO - Accumulation steps: 15
80
+ 2025-07-09 10:06:17,121 - INFO - Optimizer steps per epoch: 31
81
+ 2025-07-09 10:06:17,121 - INFO - Total optimizer steps: 930
82
+ 2025-07-09 10:06:17,121 - INFO - Warmup steps: 1000
83
+ 2025-07-09 10:06:17,121 - INFO - Validating gradient accumulation setup...
84
+ 2025-07-09 10:06:17,121 - INFO - Validating gradient accumulation with 15 steps...
85
+ 2025-07-09 10:06:35,061 - WARNING - Not enough test batches (10) for accumulation_steps (15)
86
+ 2025-07-09 10:06:35,061 - INFO - Starting training for 30 epochs
87
+ 2025-07-09 10:18:51,576 - INFO - Epoch 1: Total optimizer steps: 31
88
+ 2025-07-09 10:22:08,662 - INFO - Validation metrics:
89
+ 2025-07-09 10:22:08,663 - INFO - Loss: 0.5501
90
+ 2025-07-09 10:22:08,663 - INFO - Average similarity: 0.0474
91
+ 2025-07-09 10:22:08,663 - INFO - Median similarity: 0.0319
92
+ 2025-07-09 10:22:08,663 - INFO - Clean sample similarity: 0.0474
93
+ 2025-07-09 10:22:08,663 - INFO - Corrupted sample similarity: 0.0339
94
+ 2025-07-09 10:22:08,663 - INFO - Similarity gap (clean - corrupt): 0.0134
95
+ 2025-07-09 10:22:08,790 - INFO - Epoch 1/30 - Train Loss: 0.6551, Val Loss: 0.5501, Clean Sim: 0.0474, Corrupt Sim: 0.0339, Gap: 0.0134, Time: 933.73s
96
+ 2025-07-09 10:22:08,790 - INFO - New best validation loss: 0.5501
97
+ 2025-07-09 10:22:15,090 - INFO - New best similarity gap: 0.0134
98
+ 2025-07-09 10:34:33,415 - INFO - Epoch 2: Total optimizer steps: 31
99
+ 2025-07-09 10:37:49,380 - INFO - Validation metrics:
100
+ 2025-07-09 10:37:49,380 - INFO - Loss: 0.3848
101
+ 2025-07-09 10:37:49,380 - INFO - Average similarity: 0.3295
102
+ 2025-07-09 10:37:49,380 - INFO - Median similarity: 0.1824
103
+ 2025-07-09 10:37:49,380 - INFO - Clean sample similarity: 0.3295
104
+ 2025-07-09 10:37:49,380 - INFO - Corrupted sample similarity: 0.1992
105
+ 2025-07-09 10:37:49,380 - INFO - Similarity gap (clean - corrupt): 0.1303
106
+ 2025-07-09 10:37:49,528 - INFO - Epoch 2/30 - Train Loss: 0.5121, Val Loss: 0.3848, Clean Sim: 0.3295, Corrupt Sim: 0.1992, Gap: 0.1303, Time: 928.07s
107
+ 2025-07-09 10:37:49,528 - INFO - New best validation loss: 0.3848
108
+ 2025-07-09 10:37:56,439 - INFO - New best similarity gap: 0.1303
109
+ 2025-07-09 10:40:47,309 - INFO - Epoch 2 Validation Alignment: Pos=0.000, Neg=0.000, Gap=0.000
110
+ 2025-07-09 10:52:54,869 - INFO - Epoch 3: Total optimizer steps: 31
111
+ 2025-07-09 10:56:10,304 - INFO - Validation metrics:
112
+ 2025-07-09 10:56:10,304 - INFO - Loss: 0.3347
113
+ 2025-07-09 10:56:10,304 - INFO - Average similarity: 0.4569
114
+ 2025-07-09 10:56:10,304 - INFO - Median similarity: 0.3819
115
+ 2025-07-09 10:56:10,304 - INFO - Clean sample similarity: 0.4569
116
+ 2025-07-09 10:56:10,304 - INFO - Corrupted sample similarity: 0.2749
117
+ 2025-07-09 10:56:10,304 - INFO - Similarity gap (clean - corrupt): 0.1820
118
+ 2025-07-09 10:56:10,443 - INFO - Epoch 3/30 - Train Loss: 0.4332, Val Loss: 0.3347, Clean Sim: 0.4569, Corrupt Sim: 0.2749, Gap: 0.1820, Time: 923.13s
119
+ 2025-07-09 10:56:10,443 - INFO - New best validation loss: 0.3347
120
+ 2025-07-09 10:56:17,152 - INFO - New best similarity gap: 0.1820
121
+ 2025-07-09 11:08:36,466 - INFO - Epoch 4: Total optimizer steps: 31
122
+ 2025-07-09 11:11:52,118 - INFO - Validation metrics:
123
+ 2025-07-09 11:11:52,118 - INFO - Loss: 0.3008
124
+ 2025-07-09 11:11:52,118 - INFO - Average similarity: 0.4935
125
+ 2025-07-09 11:11:52,118 - INFO - Median similarity: 0.4840
126
+ 2025-07-09 11:11:52,118 - INFO - Clean sample similarity: 0.4935
127
+ 2025-07-09 11:11:52,118 - INFO - Corrupted sample similarity: 0.2809
128
+ 2025-07-09 11:11:52,118 - INFO - Similarity gap (clean - corrupt): 0.2126
129
+ 2025-07-09 11:11:52,234 - INFO - Epoch 4/30 - Train Loss: 0.3885, Val Loss: 0.3008, Clean Sim: 0.4935, Corrupt Sim: 0.2809, Gap: 0.2126, Time: 927.71s
130
+ 2025-07-09 11:11:52,234 - INFO - New best validation loss: 0.3008
131
+ 2025-07-09 11:11:58,985 - INFO - New best similarity gap: 0.2126
132
+ 2025-07-09 11:14:50,013 - INFO - Epoch 4 Validation Alignment: Pos=0.000, Neg=0.000, Gap=0.000
133
+ 2025-07-09 11:26:58,690 - INFO - Epoch 5: Total optimizer steps: 31
134
+ 2025-07-09 11:30:15,510 - INFO - Validation metrics:
135
+ 2025-07-09 11:30:15,510 - INFO - Loss: 0.2910
136
+ 2025-07-09 11:30:15,510 - INFO - Average similarity: 0.5942
137
+ 2025-07-09 11:30:15,510 - INFO - Median similarity: 0.8150
138
+ 2025-07-09 11:30:15,510 - INFO - Clean sample similarity: 0.5942
139
+ 2025-07-09 11:30:15,510 - INFO - Corrupted sample similarity: 0.3597
140
+ 2025-07-09 11:30:15,510 - INFO - Similarity gap (clean - corrupt): 0.2344
141
+ 2025-07-09 11:30:15,637 - INFO - Epoch 5/30 - Train Loss: 0.3716, Val Loss: 0.2910, Clean Sim: 0.5942, Corrupt Sim: 0.3597, Gap: 0.2344, Time: 925.62s
142
+ 2025-07-09 11:30:15,637 - INFO - New best validation loss: 0.2910
143
+ 2025-07-09 11:30:22,374 - INFO - New best similarity gap: 0.2344
144
+ 2025-07-09 11:42:45,432 - INFO - Epoch 6: Total optimizer steps: 31
145
+ 2025-07-09 11:46:01,110 - INFO - Validation metrics:
146
+ 2025-07-09 11:46:01,111 - INFO - Loss: 0.2737
147
+ 2025-07-09 11:46:01,111 - INFO - Average similarity: 0.5773
148
+ 2025-07-09 11:46:01,111 - INFO - Median similarity: 0.7764
149
+ 2025-07-09 11:46:01,111 - INFO - Clean sample similarity: 0.5773
150
+ 2025-07-09 11:46:01,111 - INFO - Corrupted sample similarity: 0.3289
151
+ 2025-07-09 11:46:01,111 - INFO - Similarity gap (clean - corrupt): 0.2484
152
+ 2025-07-09 11:46:01,242 - INFO - Epoch 6/30 - Train Loss: 0.3509, Val Loss: 0.2737, Clean Sim: 0.5773, Corrupt Sim: 0.3289, Gap: 0.2484, Time: 931.59s
153
+ 2025-07-09 11:46:01,242 - INFO - New best validation loss: 0.2737
154
+ 2025-07-09 11:46:08,171 - INFO - New best similarity gap: 0.2484
155
+ 2025-07-09 11:48:58,321 - INFO - Epoch 6 Validation Alignment: Pos=0.000, Neg=0.000, Gap=0.000
156
+ 2025-07-09 12:01:16,208 - INFO - Epoch 7: Total optimizer steps: 31
157
+ 2025-07-09 12:04:31,657 - INFO - Validation metrics:
158
+ 2025-07-09 12:04:31,657 - INFO - Loss: 0.2616
159
+ 2025-07-09 12:04:31,657 - INFO - Average similarity: 0.6094
160
+ 2025-07-09 12:04:31,657 - INFO - Median similarity: 0.8658
161
+ 2025-07-09 12:04:31,657 - INFO - Clean sample similarity: 0.6094
162
+ 2025-07-09 12:04:31,658 - INFO - Corrupted sample similarity: 0.3416
163
+ 2025-07-09 12:04:31,658 - INFO - Similarity gap (clean - corrupt): 0.2678
164
+ 2025-07-09 12:04:31,765 - INFO - Epoch 7/30 - Train Loss: 0.3341, Val Loss: 0.2616, Clean Sim: 0.6094, Corrupt Sim: 0.3416, Gap: 0.2678, Time: 933.44s
165
+ 2025-07-09 12:04:31,765 - INFO - New best validation loss: 0.2616
166
+ 2025-07-09 12:04:38,550 - INFO - New best similarity gap: 0.2678
167
+ 2025-07-09 12:17:00,511 - INFO - Epoch 8: Total optimizer steps: 31
168
+ 2025-07-09 12:20:16,262 - INFO - Validation metrics:
169
+ 2025-07-09 12:20:16,262 - INFO - Loss: 0.2580
170
+ 2025-07-09 12:20:16,262 - INFO - Average similarity: 0.6054
171
+ 2025-07-09 12:20:16,262 - INFO - Median similarity: 0.8577
172
+ 2025-07-09 12:20:16,262 - INFO - Clean sample similarity: 0.6054
173
+ 2025-07-09 12:20:16,262 - INFO - Corrupted sample similarity: 0.3324
174
+ 2025-07-09 12:20:16,262 - INFO - Similarity gap (clean - corrupt): 0.2730
175
+ 2025-07-09 12:20:16,374 - INFO - Epoch 8/30 - Train Loss: 0.3235, Val Loss: 0.2580, Clean Sim: 0.6054, Corrupt Sim: 0.3324, Gap: 0.2730, Time: 930.51s
176
+ 2025-07-09 12:20:16,374 - INFO - New best validation loss: 0.2580
177
+ 2025-07-09 12:20:23,227 - INFO - New best similarity gap: 0.2730
178
+ 2025-07-09 12:23:13,199 - INFO - Epoch 8 Validation Alignment: Pos=0.000, Neg=0.000, Gap=0.000
179
+ 2025-07-09 12:35:29,937 - INFO - Epoch 9: Total optimizer steps: 31
180
+ 2025-07-09 12:38:45,452 - INFO - Validation metrics:
181
+ 2025-07-09 12:38:45,452 - INFO - Loss: 0.2467
182
+ 2025-07-09 12:38:45,452 - INFO - Average similarity: 0.6282
183
+ 2025-07-09 12:38:45,452 - INFO - Median similarity: 0.9019
184
+ 2025-07-09 12:38:45,452 - INFO - Clean sample similarity: 0.6282
185
+ 2025-07-09 12:38:45,452 - INFO - Corrupted sample similarity: 0.3381
186
+ 2025-07-09 12:38:45,452 - INFO - Similarity gap (clean - corrupt): 0.2901
187
+ 2025-07-09 12:38:45,580 - INFO - Epoch 9/30 - Train Loss: 0.3127, Val Loss: 0.2467, Clean Sim: 0.6282, Corrupt Sim: 0.3381, Gap: 0.2901, Time: 932.38s
188
+ 2025-07-09 12:38:45,580 - INFO - New best validation loss: 0.2467
189
+ 2025-07-09 12:38:52,326 - INFO - New best similarity gap: 0.2901
190
+ 2025-07-09 12:51:11,687 - INFO - Epoch 10: Total optimizer steps: 31
191
+ 2025-07-09 12:54:27,759 - INFO - Validation metrics:
192
+ 2025-07-09 12:54:27,760 - INFO - Loss: 0.2337
193
+ 2025-07-09 12:54:27,760 - INFO - Average similarity: 0.6409
194
+ 2025-07-09 12:54:27,760 - INFO - Median similarity: 0.9230
195
+ 2025-07-09 12:54:27,760 - INFO - Clean sample similarity: 0.6409
196
+ 2025-07-09 12:54:27,760 - INFO - Corrupted sample similarity: 0.3338
197
+ 2025-07-09 12:54:27,760 - INFO - Similarity gap (clean - corrupt): 0.3071
198
+ 2025-07-09 12:54:27,863 - INFO - Epoch 10/30 - Train Loss: 0.3025, Val Loss: 0.2337, Clean Sim: 0.6409, Corrupt Sim: 0.3338, Gap: 0.3071, Time: 928.26s
199
+ 2025-07-09 12:54:27,863 - INFO - New best validation loss: 0.2337
200
+ 2025-07-09 12:54:34,605 - INFO - New best similarity gap: 0.3071
201
+ 2025-07-09 12:57:23,489 - INFO - Epoch 10 Validation Alignment: Pos=0.000, Neg=0.000, Gap=0.000
202
+ 2025-07-09 13:09:34,426 - INFO - Epoch 11: Total optimizer steps: 31
203
+ 2025-07-09 13:12:51,270 - INFO - Validation metrics:
204
+ 2025-07-09 13:12:51,270 - INFO - Loss: 0.2312
205
+ 2025-07-09 13:12:51,270 - INFO - Average similarity: 0.6329
206
+ 2025-07-09 13:12:51,270 - INFO - Median similarity: 0.9068
207
+ 2025-07-09 13:12:51,270 - INFO - Clean sample similarity: 0.6329
208
+ 2025-07-09 13:12:51,270 - INFO - Corrupted sample similarity: 0.3190
209
+ 2025-07-09 13:12:51,270 - INFO - Similarity gap (clean - corrupt): 0.3138
210
+ 2025-07-09 13:12:51,384 - INFO - Epoch 11/30 - Train Loss: 0.2904, Val Loss: 0.2312, Clean Sim: 0.6329, Corrupt Sim: 0.3190, Gap: 0.3138, Time: 927.89s
211
+ 2025-07-09 13:12:51,384 - INFO - New best validation loss: 0.2312
212
+ 2025-07-09 13:12:58,336 - INFO - New best similarity gap: 0.3138
213
+ 2025-07-09 13:25:17,357 - INFO - Epoch 12: Total optimizer steps: 31
214
+ 2025-07-09 13:28:35,622 - INFO - Validation metrics:
215
+ 2025-07-09 13:28:35,622 - INFO - Loss: 0.2223
216
+ 2025-07-09 13:28:35,622 - INFO - Average similarity: 0.6732
217
+ 2025-07-09 13:28:35,622 - INFO - Median similarity: 0.9485
218
+ 2025-07-09 13:28:35,622 - INFO - Clean sample similarity: 0.6732
219
+ 2025-07-09 13:28:35,622 - INFO - Corrupted sample similarity: 0.3435
220
+ 2025-07-09 13:28:35,622 - INFO - Similarity gap (clean - corrupt): 0.3297
221
+ 2025-07-09 13:28:35,738 - INFO - Epoch 12/30 - Train Loss: 0.2779, Val Loss: 0.2223, Clean Sim: 0.6732, Corrupt Sim: 0.3435, Gap: 0.3297, Time: 930.02s
222
+ 2025-07-09 13:28:35,738 - INFO - New best validation loss: 0.2223
223
+ 2025-07-09 13:28:42,677 - INFO - New best similarity gap: 0.3297
224
+ 2025-07-09 13:31:33,532 - INFO - Epoch 12 Validation Alignment: Pos=0.000, Neg=0.000, Gap=0.000
225
+ 2025-07-09 13:43:51,607 - INFO - Epoch 13: Total optimizer steps: 31
226
+ 2025-07-09 13:47:08,884 - INFO - Validation metrics:
227
+ 2025-07-09 13:47:08,885 - INFO - Loss: 0.2143
228
+ 2025-07-09 13:47:08,885 - INFO - Average similarity: 0.6180
229
+ 2025-07-09 13:47:08,885 - INFO - Median similarity: 0.8872
230
+ 2025-07-09 13:47:08,885 - INFO - Clean sample similarity: 0.6180
231
+ 2025-07-09 13:47:08,885 - INFO - Corrupted sample similarity: 0.2805
232
+ 2025-07-09 13:47:08,885 - INFO - Similarity gap (clean - corrupt): 0.3375
233
+ 2025-07-09 13:47:09,002 - INFO - Epoch 13/30 - Train Loss: 0.2746, Val Loss: 0.2143, Clean Sim: 0.6180, Corrupt Sim: 0.2805, Gap: 0.3375, Time: 935.47s
234
+ 2025-07-09 13:47:09,003 - INFO - New best validation loss: 0.2143
235
+ 2025-07-09 13:47:16,005 - INFO - New best similarity gap: 0.3375
236
+ 2025-07-09 13:59:44,255 - INFO - Epoch 14: Total optimizer steps: 31
237
+ 2025-07-09 14:03:01,055 - INFO - Validation metrics:
238
+ 2025-07-09 14:03:01,055 - INFO - Loss: 0.2056
239
+ 2025-07-09 14:03:01,055 - INFO - Average similarity: 0.6810
240
+ 2025-07-09 14:03:01,055 - INFO - Median similarity: 0.9569
241
+ 2025-07-09 14:03:01,055 - INFO - Clean sample similarity: 0.6810
242
+ 2025-07-09 14:03:01,055 - INFO - Corrupted sample similarity: 0.3251
243
+ 2025-07-09 14:03:01,055 - INFO - Similarity gap (clean - corrupt): 0.3559
244
+ 2025-07-09 14:03:01,156 - INFO - Epoch 14/30 - Train Loss: 0.2641, Val Loss: 0.2056, Clean Sim: 0.6810, Corrupt Sim: 0.3251, Gap: 0.3559, Time: 937.64s
245
+ 2025-07-09 14:03:01,156 - INFO - New best validation loss: 0.2056
246
+ 2025-07-09 14:03:07,933 - INFO - New best similarity gap: 0.3559
247
+ 2025-07-09 14:05:58,018 - INFO - Epoch 14 Validation Alignment: Pos=0.000, Neg=0.000, Gap=0.000
248
+ 2025-07-09 14:18:08,049 - INFO - Epoch 15: Total optimizer steps: 31
249
+ 2025-07-09 14:21:24,577 - INFO - Validation metrics:
250
+ 2025-07-09 14:21:24,577 - INFO - Loss: 0.2027
251
+ 2025-07-09 14:21:24,577 - INFO - Average similarity: 0.6446
252
+ 2025-07-09 14:21:24,577 - INFO - Median similarity: 0.9029
253
+ 2025-07-09 14:21:24,577 - INFO - Clean sample similarity: 0.6446
254
+ 2025-07-09 14:21:24,577 - INFO - Corrupted sample similarity: 0.2914
255
+ 2025-07-09 14:21:24,578 - INFO - Similarity gap (clean - corrupt): 0.3532
256
+ 2025-07-09 14:21:24,706 - INFO - Epoch 15/30 - Train Loss: 0.2543, Val Loss: 0.2027, Clean Sim: 0.6446, Corrupt Sim: 0.2914, Gap: 0.3532, Time: 926.69s
257
+ 2025-07-09 14:21:24,706 - INFO - New best validation loss: 0.2027
258
+ 2025-07-09 14:33:59,853 - INFO - Epoch 16: Total optimizer steps: 31
259
+ 2025-07-09 14:37:18,100 - INFO - Validation metrics:
260
+ 2025-07-09 14:37:18,100 - INFO - Loss: 0.1950
261
+ 2025-07-09 14:37:18,100 - INFO - Average similarity: 0.7201
262
+ 2025-07-09 14:37:18,100 - INFO - Median similarity: 0.9757
263
+ 2025-07-09 14:37:18,100 - INFO - Clean sample similarity: 0.7201
264
+ 2025-07-09 14:37:18,100 - INFO - Corrupted sample similarity: 0.3419
265
+ 2025-07-09 14:37:18,100 - INFO - Similarity gap (clean - corrupt): 0.3782
266
+ 2025-07-09 14:37:18,222 - INFO - Epoch 16/30 - Train Loss: 0.2506, Val Loss: 0.1950, Clean Sim: 0.7201, Corrupt Sim: 0.3419, Gap: 0.3782, Time: 941.32s
267
+ 2025-07-09 14:37:18,222 - INFO - New best validation loss: 0.1950
268
+ 2025-07-09 14:37:24,624 - INFO - New best similarity gap: 0.3782
269
+ 2025-07-09 14:40:15,594 - INFO - Epoch 16 Validation Alignment: Pos=0.000, Neg=0.000, Gap=0.000
270
+ 2025-07-09 14:52:44,776 - INFO - Epoch 17: Total optimizer steps: 31
271
+ 2025-07-09 14:56:02,178 - INFO - Validation metrics:
272
+ 2025-07-09 14:56:02,178 - INFO - Loss: 0.1946
273
+ 2025-07-09 14:56:02,178 - INFO - Average similarity: 0.6943
274
+ 2025-07-09 14:56:02,178 - INFO - Median similarity: 0.9603
275
+ 2025-07-09 14:56:02,178 - INFO - Clean sample similarity: 0.6943
276
+ 2025-07-09 14:56:02,178 - INFO - Corrupted sample similarity: 0.3080
277
+ 2025-07-09 14:56:02,178 - INFO - Similarity gap (clean - corrupt): 0.3864
278
+ 2025-07-09 14:56:02,281 - INFO - Epoch 17/30 - Train Loss: 0.2467, Val Loss: 0.1946, Clean Sim: 0.6943, Corrupt Sim: 0.3080, Gap: 0.3864, Time: 946.69s
279
+ 2025-07-09 14:56:02,282 - INFO - New best validation loss: 0.1946
280
+ 2025-07-09 14:56:08,783 - INFO - New best similarity gap: 0.3864
281
+ 2025-07-09 15:08:42,755 - INFO - Epoch 18: Total optimizer steps: 31
282
+ 2025-07-09 15:12:01,042 - INFO - Validation metrics:
283
+ 2025-07-09 15:12:01,043 - INFO - Loss: 0.1850
284
+ 2025-07-09 15:12:01,043 - INFO - Average similarity: 0.7162
285
+ 2025-07-09 15:12:01,043 - INFO - Median similarity: 0.9751
286
+ 2025-07-09 15:12:01,043 - INFO - Clean sample similarity: 0.7162
287
+ 2025-07-09 15:12:01,043 - INFO - Corrupted sample similarity: 0.3205
288
+ 2025-07-09 15:12:01,043 - INFO - Similarity gap (clean - corrupt): 0.3957
289
+ 2025-07-09 15:12:01,157 - INFO - Epoch 18/30 - Train Loss: 0.2353, Val Loss: 0.1850, Clean Sim: 0.7162, Corrupt Sim: 0.3205, Gap: 0.3957, Time: 945.76s
290
+ 2025-07-09 15:12:01,157 - INFO - New best validation loss: 0.1850
291
+ 2025-07-09 15:12:07,833 - INFO - New best similarity gap: 0.3957
292
+ 2025-07-09 15:14:57,923 - INFO - Epoch 18 Validation Alignment: Pos=0.000, Neg=0.000, Gap=0.000
293
+ 2025-07-09 15:27:22,903 - INFO - Epoch 19: Total optimizer steps: 31
294
+ 2025-07-09 15:30:40,855 - INFO - Validation metrics:
295
+ 2025-07-09 15:30:40,855 - INFO - Loss: 0.1868
296
+ 2025-07-09 15:30:40,855 - INFO - Average similarity: 0.7511
297
+ 2025-07-09 15:30:40,855 - INFO - Median similarity: 0.9870
298
+ 2025-07-09 15:30:40,855 - INFO - Clean sample similarity: 0.7511
299
+ 2025-07-09 15:30:40,855 - INFO - Corrupted sample similarity: 0.3606
300
+ 2025-07-09 15:30:40,855 - INFO - Similarity gap (clean - corrupt): 0.3905
301
+ 2025-07-09 15:30:40,988 - INFO - Epoch 19/30 - Train Loss: 0.2352, Val Loss: 0.1868, Clean Sim: 0.7511, Corrupt Sim: 0.3606, Gap: 0.3905, Time: 943.06s
302
+ 2025-07-09 15:43:07,937 - INFO - Epoch 20: Total optimizer steps: 31
303
+ 2025-07-09 15:46:25,284 - INFO - Validation metrics:
304
+ 2025-07-09 15:46:25,285 - INFO - Loss: 0.1752
305
+ 2025-07-09 15:46:25,285 - INFO - Average similarity: 0.6895
306
+ 2025-07-09 15:46:25,285 - INFO - Median similarity: 0.9676
307
+ 2025-07-09 15:46:25,285 - INFO - Clean sample similarity: 0.6895
308
+ 2025-07-09 15:46:25,285 - INFO - Corrupted sample similarity: 0.2810
309
+ 2025-07-09 15:46:25,285 - INFO - Similarity gap (clean - corrupt): 0.4085
310
+ 2025-07-09 15:46:25,420 - INFO - Epoch 20/30 - Train Loss: 0.2317, Val Loss: 0.1752, Clean Sim: 0.6895, Corrupt Sim: 0.2810, Gap: 0.4085, Time: 944.43s
311
+ 2025-07-09 15:46:25,420 - INFO - New best validation loss: 0.1752
312
+ 2025-07-09 15:46:32,053 - INFO - New best similarity gap: 0.4085
313
+ 2025-07-09 15:49:22,949 - INFO - Epoch 20 Validation Alignment: Pos=0.000, Neg=0.000, Gap=0.000
314
+ 2025-07-09 16:01:50,850 - INFO - Epoch 21: Total optimizer steps: 31
315
+ 2025-07-09 16:05:08,533 - INFO - Validation metrics:
316
+ 2025-07-09 16:05:08,533 - INFO - Loss: 0.1761
317
+ 2025-07-09 16:05:08,533 - INFO - Average similarity: 0.6436
318
+ 2025-07-09 16:05:08,533 - INFO - Median similarity: 0.9055
319
+ 2025-07-09 16:05:08,533 - INFO - Clean sample similarity: 0.6436
320
+ 2025-07-09 16:05:08,533 - INFO - Corrupted sample similarity: 0.2365
321
+ 2025-07-09 16:05:08,533 - INFO - Similarity gap (clean - corrupt): 0.4070
322
+ 2025-07-09 16:05:08,641 - INFO - Epoch 21/30 - Train Loss: 0.2267, Val Loss: 0.1761, Clean Sim: 0.6436, Corrupt Sim: 0.2365, Gap: 0.4070, Time: 945.69s
323
+ 2025-07-09 16:17:36,939 - INFO - Epoch 22: Total optimizer steps: 31
324
+ 2025-07-09 16:20:54,509 - INFO - Validation metrics:
325
+ 2025-07-09 16:20:54,509 - INFO - Loss: 0.1725
326
+ 2025-07-09 16:20:54,509 - INFO - Average similarity: 0.7292
327
+ 2025-07-09 16:20:54,509 - INFO - Median similarity: 0.9750
328
+ 2025-07-09 16:20:54,509 - INFO - Clean sample similarity: 0.7292
329
+ 2025-07-09 16:20:54,509 - INFO - Corrupted sample similarity: 0.3020
330
+ 2025-07-09 16:20:54,509 - INFO - Similarity gap (clean - corrupt): 0.4272
331
+ 2025-07-09 16:20:54,647 - INFO - Epoch 22/30 - Train Loss: 0.2244, Val Loss: 0.1725, Clean Sim: 0.7292, Corrupt Sim: 0.3020, Gap: 0.4272, Time: 946.01s
332
+ 2025-07-09 16:20:54,647 - INFO - New best validation loss: 0.1725
333
+ 2025-07-09 16:21:01,162 - INFO - New best similarity gap: 0.4272
334
+ 2025-07-09 16:23:50,529 - INFO - Epoch 22 Validation Alignment: Pos=0.000, Neg=0.000, Gap=0.000
335
+ 2025-07-09 16:36:19,562 - INFO - Epoch 23: Total optimizer steps: 31
336
+ 2025-07-09 16:39:36,938 - INFO - Validation metrics:
337
+ 2025-07-09 16:39:36,938 - INFO - Loss: 0.1698
338
+ 2025-07-09 16:39:36,938 - INFO - Average similarity: 0.6585
339
+ 2025-07-09 16:39:36,938 - INFO - Median similarity: 0.9260
340
+ 2025-07-09 16:39:36,938 - INFO - Clean sample similarity: 0.6585
341
+ 2025-07-09 16:39:36,938 - INFO - Corrupted sample similarity: 0.2528
342
+ 2025-07-09 16:39:36,938 - INFO - Similarity gap (clean - corrupt): 0.4057
343
+ 2025-07-09 16:39:37,057 - INFO - Epoch 23/30 - Train Loss: 0.2147, Val Loss: 0.1698, Clean Sim: 0.6585, Corrupt Sim: 0.2528, Gap: 0.4057, Time: 946.53s
344
+ 2025-07-09 16:39:37,057 - INFO - New best validation loss: 0.1698
345
+ 2025-07-09 16:52:02,629 - INFO - Epoch 24: Total optimizer steps: 31
346
+ 2025-07-09 16:55:20,171 - INFO - Validation metrics:
347
+ 2025-07-09 16:55:20,171 - INFO - Loss: 0.1622
348
+ 2025-07-09 16:55:20,171 - INFO - Average similarity: 0.6216
349
+ 2025-07-09 16:55:20,171 - INFO - Median similarity: 0.8589
350
+ 2025-07-09 16:55:20,171 - INFO - Clean sample similarity: 0.6216
351
+ 2025-07-09 16:55:20,171 - INFO - Corrupted sample similarity: 0.2012
352
+ 2025-07-09 16:55:20,171 - INFO - Similarity gap (clean - corrupt): 0.4203
353
+ 2025-07-09 16:55:20,303 - INFO - Epoch 24/30 - Train Loss: 0.2169, Val Loss: 0.1622, Clean Sim: 0.6216, Corrupt Sim: 0.2012, Gap: 0.4203, Time: 936.70s
354
+ 2025-07-09 16:55:20,304 - INFO - New best validation loss: 0.1622
355
+ 2025-07-09 16:58:12,057 - INFO - Epoch 24 Validation Alignment: Pos=0.000, Neg=0.000, Gap=0.000
356
+ 2025-07-09 17:10:36,292 - INFO - Epoch 25: Total optimizer steps: 31
357
+ 2025-07-09 17:13:54,630 - INFO - Validation metrics:
358
+ 2025-07-09 17:13:54,630 - INFO - Loss: 0.1701
359
+ 2025-07-09 17:13:54,630 - INFO - Average similarity: 0.6491
360
+ 2025-07-09 17:13:54,630 - INFO - Median similarity: 0.8909
361
+ 2025-07-09 17:13:54,630 - INFO - Clean sample similarity: 0.6491
362
+ 2025-07-09 17:13:54,630 - INFO - Corrupted sample similarity: 0.2283
363
+ 2025-07-09 17:13:54,630 - INFO - Similarity gap (clean - corrupt): 0.4207
364
+ 2025-07-09 17:13:54,735 - INFO - Epoch 25/30 - Train Loss: 0.2145, Val Loss: 0.1701, Clean Sim: 0.6491, Corrupt Sim: 0.2283, Gap: 0.4207, Time: 942.68s
365
+ 2025-07-09 17:26:12,647 - INFO - Epoch 26: Total optimizer steps: 31
366
+ 2025-07-09 17:29:28,828 - INFO - Validation metrics:
367
+ 2025-07-09 17:29:28,828 - INFO - Loss: 0.1657
368
+ 2025-07-09 17:29:28,828 - INFO - Average similarity: 0.7093
369
+ 2025-07-09 17:29:28,828 - INFO - Median similarity: 0.9734
370
+ 2025-07-09 17:29:28,828 - INFO - Clean sample similarity: 0.7093
371
+ 2025-07-09 17:29:28,828 - INFO - Corrupted sample similarity: 0.2811
372
+ 2025-07-09 17:29:28,828 - INFO - Similarity gap (clean - corrupt): 0.4282
373
+ 2025-07-09 17:29:28,950 - INFO - Epoch 26/30 - Train Loss: 0.2103, Val Loss: 0.1657, Clean Sim: 0.7093, Corrupt Sim: 0.2811, Gap: 0.4282, Time: 934.21s
374
+ 2025-07-09 17:29:28,950 - INFO - New best similarity gap: 0.4282
375
+ 2025-07-09 17:32:19,140 - INFO - Epoch 26 Validation Alignment: Pos=0.000, Neg=0.000, Gap=0.000
376
+ 2025-07-09 17:44:44,050 - INFO - Epoch 27: Total optimizer steps: 31
377
+ 2025-07-09 17:48:03,080 - INFO - Validation metrics:
378
+ 2025-07-09 17:48:03,080 - INFO - Loss: 0.1599
379
+ 2025-07-09 17:48:03,080 - INFO - Average similarity: 0.7226
380
+ 2025-07-09 17:48:03,080 - INFO - Median similarity: 0.9653
381
+ 2025-07-09 17:48:03,080 - INFO - Clean sample similarity: 0.7226
382
+ 2025-07-09 17:48:03,080 - INFO - Corrupted sample similarity: 0.2710
383
+ 2025-07-09 17:48:03,080 - INFO - Similarity gap (clean - corrupt): 0.4516
384
+ 2025-07-09 17:48:03,205 - INFO - Epoch 27/30 - Train Loss: 0.2110, Val Loss: 0.1599, Clean Sim: 0.7226, Corrupt Sim: 0.2710, Gap: 0.4516, Time: 944.06s
385
+ 2025-07-09 17:48:03,206 - INFO - New best validation loss: 0.1599
386
+ 2025-07-09 17:48:09,874 - INFO - New best similarity gap: 0.4516
387
+ 2025-07-09 18:00:42,666 - INFO - Epoch 28: Total optimizer steps: 31
388
+ 2025-07-09 18:04:01,807 - INFO - Validation metrics:
389
+ 2025-07-09 18:04:01,808 - INFO - Loss: 0.1578
390
+ 2025-07-09 18:04:01,808 - INFO - Average similarity: 0.6821
391
+ 2025-07-09 18:04:01,808 - INFO - Median similarity: 0.9215
392
+ 2025-07-09 18:04:01,808 - INFO - Clean sample similarity: 0.6821
393
+ 2025-07-09 18:04:01,808 - INFO - Corrupted sample similarity: 0.2303
394
+ 2025-07-09 18:04:01,808 - INFO - Similarity gap (clean - corrupt): 0.4518
395
+ 2025-07-09 18:04:01,924 - INFO - Epoch 28/30 - Train Loss: 0.2068, Val Loss: 0.1578, Clean Sim: 0.6821, Corrupt Sim: 0.2303, Gap: 0.4518, Time: 944.98s
396
+ 2025-07-09 18:04:01,924 - INFO - New best validation loss: 0.1578
397
+ 2025-07-09 18:04:08,704 - INFO - New best similarity gap: 0.4518
398
+ 2025-07-09 18:06:59,355 - INFO - Epoch 28 Validation Alignment: Pos=0.000, Neg=0.000, Gap=0.000
399
+ 2025-07-09 18:19:22,621 - INFO - Epoch 29: Total optimizer steps: 31
400
+ 2025-07-09 18:22:40,643 - INFO - Validation metrics:
401
+ 2025-07-09 18:22:40,644 - INFO - Loss: 0.1575
402
+ 2025-07-09 18:22:40,644 - INFO - Average similarity: 0.6681
403
+ 2025-07-09 18:22:40,644 - INFO - Median similarity: 0.9217
404
+ 2025-07-09 18:22:40,644 - INFO - Clean sample similarity: 0.6681
405
+ 2025-07-09 18:22:40,644 - INFO - Corrupted sample similarity: 0.2448
406
+ 2025-07-09 18:22:40,644 - INFO - Similarity gap (clean - corrupt): 0.4233
407
+ 2025-07-09 18:22:40,759 - INFO - Epoch 29/30 - Train Loss: 0.2089, Val Loss: 0.1575, Clean Sim: 0.6681, Corrupt Sim: 0.2448, Gap: 0.4233, Time: 941.40s
408
+ 2025-07-09 18:22:40,760 - INFO - New best validation loss: 0.1575
409
+ 2025-07-09 18:35:06,440 - INFO - Epoch 30: Total optimizer steps: 31
410
+ 2025-07-09 18:38:24,646 - INFO - Validation metrics:
411
+ 2025-07-09 18:38:24,647 - INFO - Loss: 0.1578
412
+ 2025-07-09 18:38:24,647 - INFO - Average similarity: 0.7543
413
+ 2025-07-09 18:38:24,647 - INFO - Median similarity: 0.9870
414
+ 2025-07-09 18:38:24,647 - INFO - Clean sample similarity: 0.7543
415
+ 2025-07-09 18:38:24,647 - INFO - Corrupted sample similarity: 0.3034
416
+ 2025-07-09 18:38:24,647 - INFO - Similarity gap (clean - corrupt): 0.4509
417
+ 2025-07-09 18:38:24,749 - INFO - Epoch 30/30 - Train Loss: 0.2037, Val Loss: 0.1578, Clean Sim: 0.7543, Corrupt Sim: 0.3034, Gap: 0.4509, Time: 937.46s
418
+ 2025-07-09 18:41:15,378 - INFO - Epoch 30 Validation Alignment: Pos=0.000, Neg=0.000, Gap=0.000
419
+ 2025-07-09 18:41:15,378 - INFO - Training completed!
420
+ 2025-07-09 18:41:21,296 - INFO - Evaluating best models on test set...
421
+ 2025-07-09 18:41:25,008 - INFO - Loaded best loss model from epoch 29
422
+ 2025-07-09 18:44:59,190 - INFO - Test (Best Loss) metrics:
423
+ 2025-07-09 18:44:59,190 - INFO - Loss: 0.1639
424
+ 2025-07-09 18:44:59,190 - INFO - Average similarity: 0.6754
425
+ 2025-07-09 18:44:59,190 - INFO - Median similarity: 0.9328
426
+ 2025-07-09 18:44:59,190 - INFO - Clean sample similarity: 0.6754
427
+ 2025-07-09 18:44:59,190 - INFO - Corrupted sample similarity: 0.2568
428
+ 2025-07-09 18:44:59,190 - INFO - Similarity gap (clean - corrupt): 0.4186
429
+ 2025-07-09 18:48:07,330 - INFO - Loaded best gap model from epoch 28
430
+ 2025-07-09 18:51:43,726 - INFO - Test (Best Gap) metrics:
431
+ 2025-07-09 18:51:43,726 - INFO - Loss: 0.1586
432
+ 2025-07-09 18:51:43,726 - INFO - Average similarity: 0.6904
433
+ 2025-07-09 18:51:43,726 - INFO - Median similarity: 0.9348
434
+ 2025-07-09 18:51:43,726 - INFO - Clean sample similarity: 0.6904
435
+ 2025-07-09 18:51:43,726 - INFO - Corrupted sample similarity: 0.2364
436
+ 2025-07-09 18:51:43,726 - INFO - Similarity gap (clean - corrupt): 0.4540
437
+ 2025-07-09 18:54:45,139 - INFO - Evaluation completed!
438
+ 2025-07-09 18:54:45,139 - INFO - Test results for best_loss_model:
439
+ 2025-07-09 18:54:45,139 - INFO - Loss: 0.1639
440
+ 2025-07-09 18:54:45,139 - INFO - Clean Sample Similarity: 0.6754
441
+ 2025-07-09 18:54:45,139 - INFO - Corrupted Sample Similarity: 0.2568
442
+ 2025-07-09 18:54:45,139 - INFO - Similarity Gap: 0.4186
443
+ 2025-07-09 18:54:45,139 - INFO - Test results for best_gap_model:
444
+ 2025-07-09 18:54:45,139 - INFO - Loss: 0.1586
445
+ 2025-07-09 18:54:45,139 - INFO - Clean Sample Similarity: 0.6904
446
+ 2025-07-09 18:54:45,139 - INFO - Corrupted Sample Similarity: 0.2364
447
+ 2025-07-09 18:54:45,139 - INFO - Similarity Gap: 0.4540
448
+ 2025-07-09 18:54:45,544 - INFO - All tasks completed!
validation_alignment_epoch_10.png ADDED
validation_alignment_epoch_12.png ADDED
validation_alignment_epoch_14.png ADDED
validation_alignment_epoch_16.png ADDED
validation_alignment_epoch_18.png ADDED
validation_alignment_epoch_2.png ADDED
validation_alignment_epoch_20.png ADDED
validation_alignment_epoch_22.png ADDED
validation_alignment_epoch_24.png ADDED
validation_alignment_epoch_26.png ADDED
validation_alignment_epoch_28.png ADDED
validation_alignment_epoch_30.png ADDED
validation_alignment_epoch_4.png ADDED
validation_alignment_epoch_6.png ADDED
validation_alignment_epoch_8.png ADDED