Multilingual Reasoning and Language Understanding on Downstream target-language tasks Italian
69.9MATH100 ScoreSmolLM3 + GRPO
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| SmolLM3 + GRPOLanguage=IT, Stage=GRPO2026.04 | 69.9 | 40.9 | 88.2 | 88.4 | 70 | 68.8 | 71 | 100 | |
| SmolLM3Language=IT, Stage=Base2026.04 | 64.4 | 39.4 | 82.9 | 75.9 | 63.7 | 63.1 | 64.9 | 0 | |
| SmolLM3 + SFTLanguage=IT, Stage=SFT2026.04 | 63.8 | 36.6 | 85.1 | 86.4 | 59.6 | 63.7 | 65.9 | 100 |