metadata
{}
Test-Model
This model is a fine-tuned version of meta-llama/Llama-3.1-8B, and is specialized for document classification. It achieves the following results on the evaluation set:
- Accuracy: 0.924647
- Hit rate: 0.977464
Training Parameters
- Epochs: 5
- Batch Size: 20
- Threshold: 0.975
- Token Length: 512
- Data Set Size: 10000
- Evenly Distributed: True
- Municipalities: Heroey, Kongsberg, Boemlo, Luster, Maalselv
- Learning Rate: 0.0001
- Weight Decay: 0.01
- Eval Accumulation Steps: 4
Evaluation Results
| Label | Accuracy | Accuracy Accumulation | Hit Rate | Hit Rate Accumulation |
|---|---|---|---|---|
| 0 | 0.93 | 0.93 | 0.98 | 0.98 |
| 1 | 0.95 | 0.95 | 0.97 | 0.97 |
| 2 | 0.91 | 0.91 | 0.98 | 0.98 |
| 3 | 0.40 | 0.40 | 0.57 | 0.57 |
| 4 | 0.25 | 0.25 | 0.63 | 0.63 |
| 5 | 0.86 | 0.86 | 0.94 | 0.94 |
| 6 | 0.99 | 0.99 | 1.04 | 1.04 |
| 7 | 0.94 | 0.94 | 0.98 | 0.98 |
| 8 | 0.67 | 0.67 | 0.86 | 0.86 |
| 9 | 0.99 | 0.99 | 0.99 | 0.99 |