List your company

Alignment Robustness Evaluation on CLINC-150 and 20 Newsgroups (averaged)

14.2Alignment Depth DDeep Alignment Training
2.3445.4228.511.578Dec 2, 2025

Evaluation results

MethodLinks
Deep Alignment TrainingBackbone=Qwen2.5-32B2025.1214.20.942.26.912
Deep Alignment TrainingBackbone=Qwen3-4B2025.1213.60.922.47.210
Deep Alignment TrainingBackbone=Qwen2.5-3B2025.1213.10.912.77.69
Deep Alignment TrainingBackbone=Qwen3-1.7B2025.1212.50.893.18.38
Standard TrainingBackbone=Qwen2.5-32B2025.123.20.711189.1—
Standard TrainingBackbone=Qwen3-4B2025.123.10.6911.592.4—
Standard TrainingBackbone=Qwen2.5-3B2025.122.90.6712.195.8—
Standard TrainingBackbone=Qwen3-1.7B2025.122.80.6512.598.2—