Continual Learning on Nemotron Science, KodCode, and Nemotron Math (Order B Test Set)
91.2Science Test PerformanceExpeL
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| ExpeLTrained On=Stage 3 Math (Nemotron Math), Base Model=Qwen3-4B-Instruct-25072026.05 | 91.2 | 60.2 | 31.2 | 60.87 | 2.5 | 1.1 | -3.4 | |
| ExpeLTrained On=Stage 2 Code (KodCode), Base Model=Qwen3-4B-Instruct-25072026.05 | 91 | 58.6 | 30.2 | 59.93 | -0.4 | 0.4 | -6.6 | |
| ExpeLTrained On=Stage 1 Science (Nemotron Science), Base Model=Qwen3-4B-Instruct-25072026.05 | 90.6 | 65.4 | 32 | 62.67 | — | — | — | |
| MemGenTrained On=Stage 1 Science (Nemotron Science), Base Model=Qwen3-4B-Instruct-25072026.05 | 90 | 62.2 | 34 | 62.07 | — | — | — | |
| MoLEMTrained On=Stage 1 Science (Nemotron Science), Base Model=Qwen3-4B-Instruct-25072026.05 | 88.4 | 72 | 30.8 | 63.73 | — | — | — | |
| MoLEMTrained On=Stage 2 Code (KodCode), Base Model=Qwen3-4B-Instruct-25072026.05 | 88.4 | 74.2 | 30.8 | 64.47 | 0 | 0 | 0 | |
| MoLEMTrained On=Stage 3 Math (Nemotron Math), Base Model=Qwen3-4B-Instruct-25072026.05 | 88.4 | 74.2 | 59.2 | 73.93 | 0 | 0 | 0.2 | |
| VanillaTrained On=Vanilla, Base Model=Qwen3-4B-Instruct-25072026.05 | 88.2 | 72 | 30.4 | 63.53 | — | — | — | |
| MemGenTrained On=Stage 2 Code (KodCode), Base Model=Qwen3-4B-Instruct-25072026.05 | 87.4 | 73.4 | 23 | 61.27 | 2.6 | -2.6 | -9.8 | |
| SFTTrained On=Stage 1 Science (Nemotron Science), Base Model=Qwen3-4B-Instruct-25072026.05 | 86.6 | 27.2 | 31.2 | 48.33 | — | — | — | |
| MemGenTrained On=Stage 3 Math (Nemotron Math), Base Model=Qwen3-4B-Instruct-25072026.05 | 84.4 | 68.6 | 56.2 | 69.73 | 5.2 | -5.2 | -8.6 | |
| SFTTrained On=Stage 3 Math (Nemotron Math), Base Model=Qwen3-4B-Instruct-25072026.05 | 73.6 | 59.4 | 54.8 | 62.6 | 6.4 | -6.4 | -27.5 | |
| SFTTrained On=Stage 2 Code (KodCode), Base Model=Qwen3-4B-Instruct-25072026.05 | 62 | 59.2 | 20.2 | 47.13 | 24.6 | -24.6 | -44.8 |