Language Understanding on MMLU (Accuracy and Inference Time)
73.1AccuracyUnwatermarked
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| UnwatermarkedBase Model=Ministral3-8B2026.05 | 73.1 | 1 | |
| UnwatermarkedBase Model=Qwen3-8B2026.05 | 72.6 | 1 | |
| De-markBase Model=Qwen3-8B2026.05 | 72.4 | 9.47 | |
| Watermarked BaselineBase Model=Qwen3-8B2026.05 | 72.3 | 1 | |
| WASHBase Model=Qwen3-8B, n=42026.05 | 72.3 | 1.01 | |
| WASHBase Model=Qwen3-8B, n=52026.05 | 72.3 | 1.01 | |
| ToBlendBase Model=Qwen3-8B2026.05 | 72.2 | 1.01 | |
| WASHBase Model=Qwen3-8B, n=32026.05 | 72.2 | 1.01 | |
| WASHBase Model=Ministral3-8B, n=42026.05 | 71.8 | 1.03 | |
| WASHBase Model=Ministral3-8B, n=52026.05 | 71.8 | 1.03 | |
| De-markBase Model=Ministral3-8B2026.05 | 71.7 | 10.02 | |
| Watermarked BaselineBase Model=Ministral3-8B2026.05 | 71.6 | 1 | |
| ToBlendBase Model=Ministral3-8B2026.05 | 71.6 | 1.04 | |
| WASHBase Model=Ministral3-8B, n=32026.05 | 71.6 | 1.04 | |
| WASHBase Model=Llama3.1-8B, n=52026.05 | 64.6 | 1.06 | |
| WASHBase Model=Llama3.1-8B, n=42026.05 | 64.4 | 1.06 | |
| ToBlendBase Model=Llama3.1-8B2026.05 | 63.1 | 1.06 | |
| WASHBase Model=Llama3.1-8B, n=32026.05 | 63.1 | 1.06 | |
| UnwatermarkedBase Model=Llama3.1-8B2026.05 | 58.3 | 1 | |
| De-markBase Model=Llama3.1-8B2026.05 | 56.5 | — | |
| Watermarked BaselineBase Model=Llama3.1-8B2026.05 | 56 | 1 |