Multitask Language Understanding on MMLU (Drift and Speedup)
0.1Drift (pp)Qwen3-8B
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Qwen3-8BQuantization=INT4, Hardware=NPU, Framework=MINCE2026.06 | 0.1 | 1.7 | |
| Phi-4-mini-InstQuantization=INT4, Hardware=NPU, Framework=MINCE2026.06 | 0.72 | 2 | |
| Llama-3.1-8B-InstQuantization=INT4, Hardware=NPU, Framework=MINCE2026.06 | 1.48 | 3.4 |