List your company
ResearchFormerly WizWand

Reasoning on BIG-Bench Hard MIX-14K

69.9AccuracyDDRO
-1.3417.15535.6554.145Apr 6, 2026
Updated 6mo ago

Evaluation results

MethodLinks
DDROModel=Llama-8B, Evaluation Protocol=3-shot, Training Dataset=MIX-14K2026.0469.9
RDROModel=Llama-8B, Evaluation Protocol=3-shot, Training Dataset=MIX-14K2026.0469.4
KTOModel=Llama-8B, Evaluation Protocol=3-shot, Training Dataset=MIX-14K2026.0466.8
RDROModel=Llama-3B, Evaluation Protocol=3-shot, Training Dataset=MIX-14K2026.0455.3
DDROModel=Llama-3B, Evaluation Protocol=3-shot, Training Dataset=MIX-14K2026.0455.2
KTOModel=Llama-3B, Evaluation Protocol=3-shot, Training Dataset=MIX-14K2026.0453.8
RDROModel=Qwen-1.5B, Evaluation Protocol=3-shot, Training Dataset=MIX-14K2026.0428.4
DDROModel=Qwen-1.5B, Evaluation Protocol=3-shot, Training Dataset=MIX-14K2026.0427.5
KTOModel=Qwen-1.5B, Evaluation Protocol=3-shot, Training Dataset=MIX-14K2026.0424.3
KTOModel=Qwen-3B, Evaluation Protocol=3-shot, Training Dataset=MIX-14K2026.041.7
DDROModel=Qwen-3B, Evaluation Protocol=3-shot, Training Dataset=MIX-14K2026.041.5
RDROModel=Qwen-3B, Evaluation Protocol=3-shot, Training Dataset=MIX-14K2026.041.4