General Reasoning on MMLU (test)
75.46AccuracyWeDLM-8B
Evaluation results
| Method | Links | |
|---|---|---|
| WeDLM-8BModel Type=WeDLM (Ours), Base Model=Qwen3-8B, Few-shot setting=5-shot2025.12 | 75.46 | |
| Qwen3-8BModel Type=AR Baseline, Few-shot setting=5-shot2025.12 | 74.03 | |
| WeDLM-7BModel Type=WeDLM (Ours), Base Model=Qwen2.5-7B, Few-shot setting=5-shot2025.12 | 71.93 | |
| Qwen2.5-7BModel Type=AR Baseline, Few-shot setting=5-shot2025.12 | 71.62 | |
| Dream-7BModel Type=DLLM Baseline, Few-shot setting=5-shot2025.12 | 70.64 | |
| NRT-WS (-log p)Backbone=Llama-3.1-8B-Base2026.02 | 66.7 | |
| LLaDA-8BModel Type=DLLM Baseline, Few-shot setting=5-shot2025.12 | 64.61 | |
| RLPR*Backbone=Llama-3.1-8B-Base2026.02 | 58.7 | |
| Gemma3-27B-ITParameters=27B2026.03 | 56 | |
| Qwen3-8BParameters=8B2026.03 | 51.3 | |
| VEPO-7B (Full)Parameters=7B, Configuration=Full2026.03 | 48.5 | |
| VEPO-7B w/o CPTParameters=7B, CPT=excluded2026.03 | 47.6 | |
| VEPO-7B-SFTParameters=7B, Stage=SFT2026.03 | 47.2 | |
| Apertus-8B-Inst.Parameters=8B2026.03 | 45.2 | |
| Tower-Plus-9BParameters=9B2026.03 | 42.1 | |
| Qwen2.5-7B-Inst.Parameters=7B2026.03 | 40.9 |