Multi-task Language Understanding on MMLU MINCE Calibration n=1500
83.2AccuracyQwen3.5-35B-A3B
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Qwen3.5-35B-A3BPrecision=BF162026.06 | 83.2 | 0.09 | |
| Qwen3-8BPrecision=BF162026.06 | 71.93 | 0.08 | |
| Llama-3.1-8B-InstPrecision=BF162026.06 | 68.87 | 0.09 | |
| Phi-4-mini-InstPrecision=BF162026.06 | 68.27 | 0.37 | |
| Qwen2.5-3B-InstPrecision=BF162026.06 | 66.47 | 1.2 | |
| Llama-3.1-8B (base)Precision=BF162026.06 | 65 | 0.39 | |
| Mistral-7B-InstPrecision=BF162026.06 | 61.73 | 0.86 |