Natural Language Inference on XNLI French
59.1AccuracyQwen2.5
Evaluation results
| Method | Links | |
|---|---|---|
| Qwen2.5Zero-shot=true, Chat Template=true2025.12 | 59.1 | |
| GamayunZero-shot=true, Chat Template=true2025.12 | 57.9 | |
| Qwen3Zero-shot=true, Chat Template=true2025.12 | 52.7 | |
| HQ seed (LLM Training)Filtering Strategy=HQ seed (LLM Training)2026.04 | 49.04 | |
| Q3 negativesFiltering Strategy=Q3 negatives2026.04 | 48.76 | |
| HQFiltering Strategy=HQ2026.04 | 48.55 | |
| HQ Romance (No Fra)Filtering Strategy=HQ Romance (No Fra)2026.04 | 48.27 | |
| HQ seedFiltering Strategy=HQ seed2026.04 | 48.23 | |
| MLFiltering Strategy=ML2026.04 | 48.07 | |
| ML (15%)Filtering Strategy=ML (15%)2026.04 | 48.07 | |
| NordicFiltering Strategy=Nordic2026.04 | 47.83 | |
| ML (Q3)Filtering Strategy=ML (Q3)2026.04 | 47.67 | |
| MKC-eFiltering Strategy=MKC-e2026.04 | 47.35 | |
| No FilteringFiltering Strategy=No Filtering2026.04 | 47.07 | |
| HQ allFiltering Strategy=HQ all2026.04 | 46.95 | |
| EuroLMZero-shot=true, Chat Template=true2025.12 | 33.4 | |
| Gemma3Zero-shot=true, Chat Template=true2025.12 | 33.3 | |
| Llama3.2Zero-shot=true, Chat Template=true2025.12 | 33.3 |