List your company

Commonsense Reasoning on HellaSwag (NLL)

0.58NLLBase + TS
0.55280.73640.921.1036May 26, 2026

Evaluation results

MethodLinks
Base + TSBackbone=QWEN2-7B, Temperature Scaling=true2026.050.58
PPT (Distribution)Backbone=QWEN2-7B, Temperature Scaling=false2026.050.6
PPT (Distribution) + TSBackbone=QWEN2-7B, Temperature Scaling=true2026.050.6
PPT (Mean)Backbone=QWEN2-7B, Temperature Scaling=false2026.050.75
BaseBackbone=QWEN2-7B, Temperature Scaling=false2026.050.84
PPT (Distribution) + TSBackbone=LLAMA3-8B, Temperature Scaling=true2026.050.87
PPT (Distribution)Backbone=LLAMA3-8B, Temperature Scaling=false2026.050.88
Base + TSBackbone=LLAMA3-8B, Temperature Scaling=true2026.050.95
PPT (Mean)Backbone=LLAMA3-8B, Temperature Scaling=false2026.050.98
BaseBackbone=LLAMA3-8B, Temperature Scaling=false2026.051.26