Neutrino Event Classification on Simulated LArTPC detector data
87AccuracyLLaMA 3.2 Vision
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| LLaMA 3.2 Vision# of Trainable Parameters=29.5M (QLoRA), Training Regime=PEFT, 1 epoch, Inference Memory Usage (GB)=12.91, Time per Sample (ms)=34122025.09 | 87 | 87 | 87 | 96 | |
| ViT-h/14# of Trainable Parameters=632M, Training Regime=Full, 10 epochs, Inference Memory Usage (GB)=2.56, Time per Sample (ms)=299.12025.09 | 86 | 86 | 85 | 96 | |
| CNN# of Trainable Parameters=21.7M, Training Regime=Full, 300 epochs, Inference Memory Usage (GB)=2.44, Time per Sample (ms)=23.902025.09 | 80 | 80 | 79 | 94 |