pcam-interpretability/dino-vits16-val07415-vit-tuned

dino-vits16

Best Validation Accuracy: 0.7415

Metadata

• Model Name: dino-vits16
• Optimizer: adamw
• Scheduler: cosine
• Weight Decay: 0.001
• Warmup Epochs: 3
• Patience: 10
• Amp: True
• Seed: 42
• Batch Size: 352
• Initial Lr: 0.0012
• Total Epochs Ran: 12
• Early Stopped: True
• Training Time Seconds: 7287.24094581604
• Num Parameters: 21666049
• Device: NVIDIA A100-SXM4-40GB
• Run Id: vit-tuned

Training Configuration

• Epochs: 12
• Batch size: 352
• Learning rate (initial): 0.0012

Training Logs (Per Epoch)

Epoch	Train Loss	Train Acc	Val Loss	Val Acc	LR
1	0.6663	0.6782	0.5234	0.7309	0.000400
2	0.5246	0.7383	0.5054	0.7415	0.000800
3	nan	0.7173	0.6398	0.6699	0.001200
4	nan	0.5349	nan	0.5005	0.001200
5	nan	0.5000	nan	0.5005	0.001199
6	nan	0.5000	nan	0.5005	0.001195
7	nan	0.5000	nan	0.5005	0.001189
8	nan	0.5000	nan	0.5005	0.001181
9	nan	0.5000	nan	0.5005	0.001171
10	nan	0.5000	nan	0.5005	0.001158
11	nan	0.5000	nan	0.5005	0.001143
12	nan	0.5000	nan	0.5005	0.001126