Opinion Alignment on WoM
75.1Mean AccuracySFT+GRPO
Evaluation results
| Method | Links | |
|---|---|---|
| SFT+GRPOBase model=Magistral 24B2026.03 | 75.1 | |
| SFT+GRPOBase model=Llama 3.1 8B2026.03 | 75.06 | |
| SFTBase model=Magistral 24B2026.03 | 72.56 | |
| GRPOBase model=Magistral 24B2026.03 | 72.05 | |
| SFT+GRPOBase model=Qwen3 8B2026.03 | 71.16 | |
| SFTBase model=Llama 3.1 8B2026.03 | 68.25 | |
| SFTBase model=Qwen3 8B2026.03 | 61.74 | |
| GRPOBase model=Llama 3.1 8B2026.03 | 60.63 | |
| ORPOBase model=Llama 3.1 8B2026.03 | 57.48 | |
| GRPOBase model=Qwen3 8B2026.03 | 53.19 | |
| iclBase model=Qwen3 8B2026.03 | 48.67 | |
| iclBase model=Magistral 24B2026.03 | 44.64 | |
| ORPOBase model=Qwen3 8B2026.03 | 38.56 | |
| ORPOBase model=Magistral 24B2026.03 | 36.1 | |
| randomBase model=Untrained baselines2026.03 | 33.33 | |
| iclBase model=Llama 3.1 8B2026.03 | 31.44 | |
| majorityBase model=Untrained baselines2026.03 | 27.44 |