Table 11 Performance comparison of different LoRA ranks on Qwen2.5-1.5B-Instruct (similarity threshold=0.8). Metrics reported are micro-averages (mean ± SD) from 5 runs.
From: Qwen TextCNN and BERT models for enhanced multilabel news classification in mobile apps
Metrics | LoRA rank | ||||
|---|---|---|---|---|---|
2 | 4 | 8 | 16 | 32 | |
Precision | \(0.1501 \pm 0.0001\) | \(0.1502 \pm 0.0001\) | \(0.4005 \pm 0.0001\) | \(0.3501 \pm 0.0001\) | \(0.3001 \pm 0.0001\) |
Recall | \(0.1502 \pm 0.0001\) | \(0.1500 \pm 0.0001\) | \(0.3995 \pm 0.0001\) | \(0.3499 \pm 0.0001\) | \(0.2998 \pm 0.0001\) |
F1-score | \(0.1501 \pm 0.0001\) | \(0.1501 \pm 0.0001\) | \(0.4000 \pm 0.0001\) | \(0.3500 \pm 0.0001\) | \(0.2999 \pm 0.0001\) |
Accuracy | \(0.1502 \pm 0.0001\) | \(0.1498 \pm 0.0001\) | \(0.4010 \pm 0.0001\) | \(0.3505 \pm 0.0001\) | \(0.3002 \pm 0.0001\) |
