Science Question Answering on ARC-C (Accuracy)
51.47Accuracy (ARC-C)Pre-trained
Evaluation Results
| Method | Links | |
|---|---|---|
| Pre-trainedModel=Gemma-3-4B2026.06 | 51.47 | |
| AlphaTokenModel=Gemma-3-4B2026.06 | 50.74 | |
| Token CleaningModel=Gemma-3-4B2026.06 | 50.15 | |
| LESSModel=Gemma-3-4B2026.06 | 49.97 | |
| ssTOKENModel=Gemma-3-4B2026.06 | 49.74 | |
| STMModel=Gemma-3-4B2026.06 | 49.53 | |
| LoRAModel=Gemma-3-4B2026.06 | 48.96 | |
| XTFModel=Gemma-3-4B2026.06 | 48.56 | |
| Standard FTModel=Gemma-3-4B2026.06 | 43.52 | |
| HybKDDistillation Pair=Llama3.1-8B → Llama3.2-1B2026.05 | 37.01 | |
| α-β divergenceDistillation Pair=Llama3.1-8B → Llama3.2-1B2026.05 | 34.9 | |
| JS divergenceDistillation Pair=Llama3.1-8B → Llama3.2-1B2026.05 | 34.42 | |
| Adaptive KLDistillation Pair=Llama3.1-8B → Llama3.2-1B2026.05 | 33.74 | |
| Skew FKLDistillation Pair=Llama3.1-8B → Llama3.2-1B2026.05 | 32.34 | |
| Skew RKLDistillation Pair=Llama3.1-8B → Llama3.2-1B2026.05 | 32.24 | |
| Reverse KLDistillation Pair=Llama3.1-8B → Llama3.2-1B2026.05 | 32.08 | |
| Total VariationDistillation Pair=Llama3.1-8B → Llama3.2-1B2026.05 | 28.99 | |
| Reverse KLDistillation Pair=Gemma3-4B → Gemma3-1B2026.05 | 27.58 | |
| Skew FKLDistillation Pair=Gemma3-4B → Gemma3-1B2026.05 | 26.14 | |
| α-β divergenceDistillation Pair=Gemma3-4B → Gemma3-1B2026.05 | 26.07 | |
| HybKDDistillation Pair=Gemma3-4B → Gemma3-1B2026.05 | 26.07 | |
| JS divergenceDistillation Pair=Gemma3-4B → Gemma3-1B2026.05 | 26.03 | |
| Skew RKLDistillation Pair=Gemma3-4B → Gemma3-1B2026.05 | 25.93 | |
| Total VariationDistillation Pair=Gemma3-4B → Gemma3-1B2026.05 | 25.83 | |
| Adaptive KLDistillation Pair=Gemma3-4B → Gemma3-1B2026.05 | 24.07 |