List your company

Harmlessness Evaluation on LINGUASAFE (test)

100Tie RateSHARD
83.04887.44991.8596.251Jun 14, 2026

Evaluation results

MethodLinks
SHARDModel=Qwen3.5 9B2026.0610000
SHARD w/TeacherModel=Qwen3.5 27B2026.0610000
HHTUNEModel=Qwen3.5 9B2026.0610000
SHARDModel=Qwen3.5 27B2026.0699.800.2
SHARD w/TeacherModel=Qwen3.5 9B2026.0699.80.20
HHTUNEModel=Qwen3.5 27B2026.0699.800.2
RATIONALModel=Qwen3.5 9B2026.0699.80.20
RATIONALModel=Qwen3.5 27B2026.0699.800.2
SHARDModel=Qwen3.5 27B2026.0699.8——
Benign IntentModel=Qwen3.5 27B2026.0699.8——
SHARDModel=Qwen3.5 9B2026.0699.8——
Benign IntentModel=Qwen3.5 9B2026.0699.8——
Benign IntentModel=Phi-42026.0699.7——
SHARDModel=Phi-42026.0699.5——
SHARD w/TeacherModel=Phi-42026.0699.40.30.3
RATIONALModel=Llama 3.1 8B2026.0699.40.60
SHARDModel=Phi-42026.069910
HHTUNEModel=Phi-42026.06990.80.2
RATIONALModel=Phi-42026.0698.51.50
Benign IntentModel=Llama 3.1 8B2026.0698.3——
Benign IntentModel=Mistral 24B2026.0698.3——
SHARD w/TeacherModel=Llama 3.1 8B2026.0698.201.8
HHTUNEModel=Llama 3.1 8B2026.0698.21.80
SHARDModel=Llama 3.1 8B2026.0697.5——
SHARDModel=Mistral 24B2026.0697.3——
Benign IntentModel=Llama 3.3 70B2026.0697.2——
RATIONALModel=Mistral 24B2026.0696.72.70.5
SHARDModel=Llama 3.1 8B2026.0696.43.60
HHTUNEModel=Mistral 24B2026.0696.430.5
SHARDModel=Llama 3.3 70B2026.0696.1——
SHARDModel=Mistral 7B2026.0695.70.63.7
RATIONALModel=Mistral 7B2026.0695.71.92.5
HHTUNEModel=Mistral 7B2026.0694.42.53.1
SHARD w/TeacherModel=Mistral 24B2026.0693.26.60.2
Benign IntentModel=Mistral 7B2026.0692.4——
SHARDModel=Mistral 24B2026.069190
SHARDModel=Mistral 7B2026.0690.6——
SHARD w/TeacherModel=Mistral 7B2026.0683.77.19.2