Counterfactual Generation on AI-READI (Class 1)
98ValidityLlama*
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Llama*mode=Fine-tuned2026.01 | 98 | 20 | 1.9 | 99 | |
| GPT-4mode=Few-shot2026.01 | 92 | 182 | 4 | 96 | |
| BioMistral*mode=Fine-tuned2026.01 | 91 | 100 | 2.1 | 95 | |
| GPT-4mode=Zero-shot2026.01 | 89 | 150 | 3.8 | 82 | |
| CFNOW2026.01 | 84 | 25 | 3 | 99 | |
| Llamamode=Pre-trained2026.01 | 68 | 130 | 3.8 | 78 | |
| DiCE2026.01 | 58 | 41 | 2.4 | 99 | |
| NICE2026.01 | 53 | 4 | 1.31 | 35 | |
| BioMistralmode=Pre-trained2026.01 | 47 | 150 | 4.1 | 70 |