Language Understanding on MMLU-ProX (T1)
26.4AccuracyNaive Fine-tuning
Evaluation Results
| Method | Links | |
|---|---|---|
| Naive Fine-tuningBackbone=Phi-4-Mini-Instruct-3.8B2026.04 | 26.4 | |
| Full RetrainingBackbone=Phi-4-Mini-Instruct-3.8B2026.04 | 26.1 | |
| EWCBackbone=Phi-4-Mini-Instruct-3.8B2026.04 | 25.9 | |
| COMPASS-ECDABackbone=Phi-4-Mini-Instruct-3.8B2026.04 | 25.8 | |
| Random RehearsalBackbone=Phi-4-Mini-Instruct-3.8B2026.04 | 25.7 | |
| Full RetrainingModel=Qwen2.5-7B-Instruct2026.04 | 0.408 | |
| Naive Fine-tuningModel=Qwen2.5-7B-Instruct2026.04 | 0.406 | |
| COMPASS-ECDAModel=Qwen2.5-7B-Instruct2026.04 | 0.405 | |
| EWCModel=Qwen2.5-7B-Instruct2026.04 | 0.402 | |
| Random RehearsalModel=Qwen2.5-7B-Instruct2026.04 | 0.402 |