Streaming safety evaluation on QWENGUARD response loc (test)
97.6F1 ScoreLlama3-StreamGuard-3B
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Llama3-StreamGuard-3BModel scale=3B, Backbone family=Llama32026.04 | 97.6 | 92.4 | 2.8 | 4.8 | |
| Llama3-StreamGuard-8BModel scale=8B, Backbone family=Llama32026.04 | 97.5 | 92.6 | 2.5 | 4.9 | |
| Llama3-StreamGuard-1BModel scale=1B, Backbone family=Llama32026.04 | 97.2 | 92.9 | 1.6 | 5.5 | |
| Qwen3Guard-Stream-4B-strictModel scale=4B, Safety threshold=strict, Backbone family=Qwen2026.04 | 96.8 | 91 | 2.7 | 6.3 | |
| Qwen3Guard-Stream-8B-strictModel scale=8B, Safety threshold=strict, Backbone family=Qwen2026.04 | 95.9 | 89.9 | 2.2 | 7.9 | |
| Qwen3Guard-Stream-0.6B-strictModel scale=0.6B, Safety threshold=strict, Backbone family=Qwen2026.04 | 95 | 88.3 | 2.1 | 9.6 | |
| Qwen3Guard-Stream-8B-looseModel scale=8B, Safety threshold=loose, Backbone family=Qwen2026.04 | 95 | 88.7 | 1.7 | 9.6 | |
| Qwen3Guard-Stream-4B-looseModel scale=4B, Safety threshold=loose, Backbone family=Qwen2026.04 | 94.9 | 88.3 | 2 | 9.7 | |
| Qwen3Guard-Stream-0.6B-looseModel scale=0.6B, Safety threshold=loose, Backbone family=Qwen2026.04 | 94.3 | 87.9 | 1.2 | 10.8 |