Table 5 Combined comparison of fine-tuning and inference-stage methods.
From: Accurate discharge summary generation using fine tuned large language models with self evaluation
Model& Method | Fine-Tuning | Inference Optimization | ROUGE-L | BERTScore | Accuracy | Completeness |
|---|---|---|---|---|---|---|
Qwen2-7B Baseline | ✘ | few-shot | 0.391 | 0.866 | 4.0 | 3.9 |
Qwen2-7B + DoRA | ✔ | few-shot | 0.451 | 0.923 | 4.5 | 4.6 |
Qwen2-7B + Self-Evaluation | ✘ | ✔ | 0.451 | 0.923 | 4.7 | 4.9 |
Qwen2-7B + DoRA + Self-Evaluation | ✔ | ✔ | 0.486 | 0.941 | 4.8 | 4.9 |
