TrainLens Report
Model detected: FineTunedTransformer
Framework: huggingface
Training summary
- Detected a fine-tuned transformer training run.
- Training accuracy improved from 0.742 to 0.914.
- Validation loss stabilized after epoch 4.
Result explanation
- Training and validation loss are close, so the current run does not show a large loss gap.
- Training and validation accuracy are reasonably aligned for this snapshot.
- TrainLens found 2 signal(s), including 1 warning or critical item(s).
- Execution trace evidence is available, so the metric readings can be tied back to concrete training events.
Metrics
| Metric | Value |
|---|---|
| train_accuracy | 0.914 |
| train_loss | 0.214 |
| validation_accuracy | 0.887 |
| validation_loss | 0.291 |
Execution trace
| Step | Epoch | Event | Metrics |
|---|---|---|---|
| 100 | 1.00 | eval | validation_loss=0.412, validation_accuracy=0.801 |
| 500 | 5.00 | eval | validation_loss=0.291, validation_accuracy=0.887 |
Potential issues
- Possible mild overfitting: Training accuracy is higher than validation accuracy, but the gap is still manageable.
- Validation trend stable: Validation loss changed only slightly in the last epochs.
Top features
1. token_overlap
2. sequence_length
3. domain_keyword_count
Recommended next steps
- Run a validation error analysis. _Inspect false positives and false negatives before changing architecture._
- Try a lower learning rate for two extra epochs. _The curve suggests room for small convergence gains._
Improvement plan
1. Run a validation error analysis. Confidence: 82%. Why: Inspect false positives and false negatives before changing architecture.
2. Try a lower learning rate for two extra epochs. Confidence: 64%. Why: The curve suggests room for small convergence gains.