List your company

Question Answering on TriviaQA (Acc., Steps, Speed Up)

79.89AccuracySandwich
48.1756.40564.6472.875Apr 29, 2025Jun 30, 2025Aug 31, 2025Nov 2, 2025Jan 3, 2026Mar 6, 2026May 8, 2026

Evaluation results

MethodLinks
SandwichBackbone=Llama3.1-8B-Instruct2025.0479.89——
InstructionalBackbone=Llama3.1-8B-Instruct2025.0479.44——
Robustness via ReferencingBackbone=Llama3.1-8B-Instruct2025.0479.44——
NoneBackbone=Llama3.1-8B-Instruct2025.0479.11——
ReminderBackbone=Llama3.1-8B-Instruct2025.0478.89——
Robustness via ReferencingBackbone=Qwen2-7B-Instruct2025.0478——
SandwichBackbone=Llama3-8B-Instruct2025.0477.44——
Robustness via ReferencingBackbone=Llama3-8B-Instruct2025.0477.44——
SandwichBackbone=Qwen2-7B-Instruct2025.0477.44——
SpotlightBackbone=Llama3.1-8B-Instruct2025.0477.44——
NoneBackbone=Qwen2-7B-Instruct2025.0477.22——
SpotlightBackbone=Qwen2-7B-Instruct2025.0477.11——
ReminderBackbone=Qwen2-7B-Instruct2025.0476.78——
InstructionalBackbone=Qwen2-7B-Instruct2025.0476.22——
StruQBackbone=Llama3.1-8B-Instruct2025.0476.22——
NoneBackbone=Llama3-8B-Instruct2025.0475.78——
ReminderBackbone=Llama3-8B-Instruct2025.0475.67——
StruQBackbone=Llama3-8B-Instruct2025.0475.56——
StruQBackbone=Qwen2-7B-Instruct2025.0475——
SpotlightBackbone=Llama3-8B-Instruct2025.0474.22——
InstructionalBackbone=Llama3-8B-Instruct2025.0473.89——
RLBackbone=Llama-3.1-8B2026.0569.89——
RLBackbone=OLMo-2-7B2026.0565.85——
RFTBackbone=Llama-3.1-8B2026.0565.21——
DPOBackbone=Llama-3.1-8B2026.0564.4——
BaseBackbone=Llama-3.1-8B2026.0563.54——
SFTBackbone=Llama-3.1-8B2026.0562.35——
RLBackbone=Qwen2.5-7B2026.0559.94——
RFTBackbone=OLMo-2-7B2026.0559.63——
SFTBackbone=OLMo-2-7B2026.0557.67——
DPOBackbone=OLMo-2-7B2026.0555.62——
BaseBackbone=OLMo-2-7B2026.0554.86——
RFTBackbone=Qwen2.5-7B2026.0553.65——
OriginalModel=LLaDA, Decoding Strategy=Step-by-step2026.0352.64256—
DPOBackbone=Qwen2.5-7B2026.0552.43——
KLASSModel=LLaDA, Decoding Strategy=Parallel2026.0352.2177.41.44
Fast-dLLMModel=LLaDA, Decoding Strategy=Parallel2026.0352.12124.42.06
DAPDModel=LLaDA, Decoding Strategy=Parallel2026.0352.0866.23.87
SFTBackbone=Qwen2.5-7B2026.0551.79——
EBModel=LLaDA, Decoding Strategy=Parallel2026.0351.2131.31.95
BaseBackbone=Qwen2.5-7B2026.0549.39——