Span and Type Prediction on Drug-ACE (test)
51.2Hard F1 ScoreGemma3-4BRCLoRA
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Gemma3-4BRCLoRATrainable Params=15.46 M, Evaluation Protocol=LoRA Fine-tuning2026.06 | 51.2 | 59.11 | |
| Qwen3-4BRCLoRATrainable Params=17.11 M, Evaluation Protocol=LoRA Fine-tuning2026.06 | 50.91 | 59.78 | |
| Gemma2-9BRCLoRATrainable Params=28.39 M, Evaluation Protocol=LoRA Fine-tuning2026.06 | 49.57 | 58.83 | |
| Qwen3-4BLoRATrainable Params=16.52 M, Evaluation Protocol=LoRA Fine-tuning2026.06 | 49.19 | 59.18 | |
| Gemma3-4BLoRATrainable Params=14.90 M, Evaluation Protocol=LoRA Fine-tuning2026.06 | 49.01 | 56.1 | |
| MedGemma-4BRCLoRATrainable Params=15.46 M, Evaluation Protocol=LoRA Fine-tuning2026.06 | 48.4 | 56.56 | |
| Gemma2-9BLoRATrainable Params=27.01 M, Evaluation Protocol=LoRA Fine-tuning2026.06 | 47.6 | 58.3 | |
| Qwen2.5-7BRCLoRATrainable Params=20.42 M, Evaluation Protocol=LoRA Fine-tuning2026.06 | 46.52 | 56.23 | |
| Qwen2.5-7BLoRATrainable Params=20.19 M, Evaluation Protocol=LoRA Fine-tuning2026.06 | 46.42 | 55.51 | |
| MedGemma-4BLoRATrainable Params=14.90 M, Evaluation Protocol=LoRA Fine-tuning2026.06 | 46.39 | 54.51 | |
| BiomedBERT-baseTrainable Params=108.26 M, Evaluation Protocol=SpanMarker2026.06 | 39.35 | 52.91 | |
| RoBERTa-largeTrainable Params=355.39 M, Evaluation Protocol=SpanMarker2026.06 | 37.81 | 51.9 | |
| RoBERTa-baseTrainable Params=124.67 M, Evaluation Protocol=SpanMarker2026.06 | 34.81 | 47.45 | |
| BioBERT-base-casedTrainable Params=108.33 M, Evaluation Protocol=SpanMarker2026.06 | 33.38 | 45.08 | |
| BERT-base-uncasedTrainable Params=109.51 M, Evaluation Protocol=SpanMarker2026.06 | 32.96 | 44.8 | |
| Bio_ClinicalBERTTrainable Params=108.33 M, Evaluation Protocol=SpanMarker2026.06 | 32.94 | 46.03 | |
| BERT-base-casedTrainable Params=108.33 M, Evaluation Protocol=SpanMarker2026.06 | 30.69 | 44.93 | |
| Llama3.3-70BEvaluation Protocol=2-Shot Prompting2026.06 | 23.71 | 32.12 | |
| Qwen2.5-72BEvaluation Protocol=2-Shot Prompting2026.06 | 22.82 | 30.92 | |
| DeepSeek-R1-70BEvaluation Protocol=2-Shot Prompting2026.06 | 13.73 | 35 |