training_artifacts/README.md

# microsoft/codebert-base Script Detector

## Summary

| Field | Value |
| --- | --- |
| Base model | `microsoft/codebert-base` |
| Phase | `phase1` |
| Target repo | `HID-APP/script-detector-microsoft-codebert-base` |
| Datasets | `HID-APP/All-combined` |
| Text column | `text` |
| Epochs | 3.0000 |
| Max length | 512 |
| Train batch size | 4 |
| Eval batch size | 8 |
| Gradient accumulation steps | 4 |
| Learning rate | 0.0000 |
| Precision | fp16 |
| Optimizer | adamw_torch |
| Dataloader workers | 0 |
| Gradient checkpointing | False |
| Group by length | True |
| LoRA enabled | False |

## Dataset Sources

| Source | Type | Revision | Fingerprint |
| --- | --- | --- | --- |
| HID-APP/All-combined | huggingface | default | 81f16d3fdbb4f97d3574f5c78bce8ada7b15c9ccd625913daa095cae6d545d67 |

## Training And Validation History

| Epoch | Step | Train Loss | Eval Loss | Accuracy | Precision | Recall | F1 |
| --- | --- | --- | --- | --- | --- | --- | --- |
| 1.0000 | 373 | 3.2531 |  |  |  |  |  |
| 1.0000 | 373 |  | 0.5277 | 0.8353 | 0.8144 | 0.9806 | 0.8898 |
| 2.0000 | 746 | 2.7352 |  |  |  |  |  |
| 2.0000 | 746 |  | 0.5214 | 0.8291 | 0.8015 | 0.9943 | 0.8875 |
| 3.0000 | 1119 | 2.6190 |  |  |  |  |  |
| 3.0000 | 1119 |  | 0.5194 | 0.8360 | 0.8383 | 0.9396 | 0.8860 |

## Final Test Summary

| Split | Accuracy | Malicious Precision | Malicious Recall | Malicious F1 | Loss |
| --- | --- | --- | --- | --- | --- |
| test | 0.7846 | 0.7636 | 0.9598 | 0.8505 | 0.7020 |

## Final Test Classification Report

| Class | Precision | Recall | F1 | Support |
| --- | --- | --- | --- | --- |
| benign | 0.8701 | 0.4753 | 0.6147 | 465.0000 |
| malicious | 0.7636 | 0.9598 | 0.8505 | 821.0000 |
| macro avg | 0.8168 | 0.7175 | 0.7326 | 1286.0000 |
| weighted avg | 0.8021 | 0.7846 | 0.7653 | 1286.0000 |
| accuracy |  |  | 0.7846 |  |

## Final Test Confusion Matrix

|  | Pred benign | Pred malicious |
| --- | --- | --- |
| Actual benign | 221 | 244 |
| Actual malicious | 33 | 788 |

## Final Test Grouped By Original Sample

Chunk rows: `1286`; original samples: `1286`.

| Pooling | Original Samples | Accuracy | Malicious Precision | Malicious Recall | Malicious F1 |
| --- | --- | --- | --- | --- | --- |
| max_pool | 1286 | 0.7846 | 0.7636 | 0.9598 | 0.8505 |
| mean_pool | 1286 | 0.7846 | 0.7636 | 0.9598 | 0.8505 |