Harmful prompt classification on WildGuardMix (val)
98.34F1 ScoreTPC (5th order)
Evaluation Results
| Method | Links | |
|---|---|---|
| TPC (5th order)Backbone LLM=gemma-3-27B-it, Layer=322025.09 | 98.34 | |
| MLPBackbone LLM=gemma-3-27B-it, Layer=322025.09 | 98.31 | |
| EE-MLP (5th exit)Backbone LLM=gemma-3-27B-it, Layer=322025.09 | 98.22 | |
| MLPBackbone LLM=gpt-oss-20b, Layer=162025.09 | 98.21 | |
| TPC (5th order)Backbone LLM=gpt-oss-20b, Layer=162025.09 | 98.13 | |
| Bilinear probeBackbone LLM=gemma-3-27B-it, Layer=322025.09 | 98.1 | |
| EE-MLP (5th exit)Backbone LLM=gpt-oss-20b, Layer=162025.09 | 98.08 | |
| Linear probeBackbone LLM=gemma-3-27B-it, Layer=322025.09 | 97.83 | |
| TPC (5th order)Backbone LLM=Qwen3-30b-A3B-Base, Layer=322025.09 | 97.62 | |
| MLPBackbone LLM=Qwen3-30b-A3B-Base, Layer=322025.09 | 97.57 | |
| EE-MLP (5th exit)Backbone LLM=Qwen3-30b-A3B-Base, Layer=322025.09 | 97.52 | |
| Bilinear probeBackbone LLM=gpt-oss-20b, Layer=162025.09 | 97.5 | |
| Linear probeBackbone LLM=gpt-oss-20b, Layer=162025.09 | 97.37 | |
| TPC (5th order)Backbone LLM=Llama-3.2-3B, Layer=162025.09 | 97.18 | |
| MLPBackbone LLM=Llama-3.2-3B, Layer=162025.09 | 97.12 | |
| Bilinear probeBackbone LLM=Qwen3-30b-A3B-Base, Layer=322025.09 | 97.1 | |
| EE-MLP (5th exit)Backbone LLM=Llama-3.2-3B, Layer=162025.09 | 96.92 | |
| Bilinear probeBackbone LLM=Llama-3.2-3B, Layer=162025.09 | 96.7 | |
| Linear probeBackbone LLM=Qwen3-30b-A3B-Base, Layer=322025.09 | 95.77 | |
| Linear probeBackbone LLM=Llama-3.2-3B, Layer=162025.09 | 95.1 |