Tool-calling on τ-Bench (test)
43.77TSRQwen 2.5 7B (SFT)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Qwen 2.5 7B (SFT)Defense=None, Stage=SFT, Poison=5%, Threat Model=TM32025.10 | 43.77 | 100 | |
| Qwen 2.5 7B (SFT) + Llama-firewallDefense=Llama-firewall (LF), Stage=SFT, Poison=5%, Threat Model=TM32025.10 | 43.19 | 100 | |
| Qwen 2.5 7B (SFT) + Granite GuardianDefense=Granite Guardian 3.3-8B (GG), Stage=SFT, Poison=5%, Threat Model=TM32025.10 | 42.61 | 100 | |
| Qwen 2.5 7B (Base)Defense=None, Stage=Base, Poison=0%2025.10 | 22.61 | — |