pcam-interpretability/dino-vits16-val08483-vit-tuned-sota

dino-vits16

Best Validation Accuracy: 0.8483

Metadata

• Model Name: dino-vits16
• Optimizer: adamw
• Scheduler: cosine
• Weight Decay: 0.001
• Warmup Epochs: 10
• Patience: 10
• Amp: True
• Seed: 42
• Batch Size: 352
• Initial Lr: 0.001
• Total Epochs Ran: 17
• Early Stopped: True
• Training Time Seconds: 9935.389509916306
• Num Parameters: 21666049
• Device: NVIDIA A100-SXM4-40GB
• Run Id: vit-tuned-sota

Training Configuration

• Epochs: 17
• Batch size: 352
• Learning rate (initial): 0.001

Training Logs (Per Epoch)

Epoch	Train Loss	Train Acc	Val Loss	Val Acc	LR
1	0.5760	0.7394	0.4960	0.7463	0.000100
2	0.4235	0.8045	0.4166	0.7999	0.000200
3	0.3990	0.8191	0.4240	0.7902	0.000300
4	0.3827	0.8273	0.3552	0.8308	0.000400
5	0.3745	0.8315	0.3966	0.8094	0.000500
6	0.3716	0.8328	0.4255	0.7925	0.000600
7	0.3632	0.8375	0.3360	0.8483	0.000700
8	0.3603	0.8397	0.4131	0.7990	0.000800
9	0.3562	0.8418	0.3642	0.8295	0.000900
10	0.3505	0.8442	0.3885	0.8094	0.001000
11	0.3432	0.8490	0.4165	0.7999	0.001000
12	0.3381	0.8514	0.3754	0.8293	0.001000
13	0.3334	0.8536	0.3611	0.8345	0.000999
14	0.3292	0.8557	0.4204	0.8109	0.000998
15	0.3205	0.8598	0.3687	0.8343	0.000996
16	nan	0.8588	0.3862	0.8230	0.000994
17	0.3147	0.8636	0.3794	0.8328	0.000991