List your company

Question Answering on OBQA (test)

92.7AccuracyEnsemble
27.783244.636661.4978.3434May 4, 2023Nov 11, 2023May 21, 2024Nov 29, 2024Jun 9, 2025Dec 18, 2025Jun 28, 2026

Evaluation results

MethodLinks
EnsembleParams (M)=11.3052026.0692.74.960.27
MLEParams (M)=3.7682026.0692.256.40.38
MC-DropoutParams (M)=3.7682026.0691.86.60.36
MAPParams (M)=3.7682026.0691.726.610.39
BLoBParams (M)=5.4032026.0691.722.160.21
LaplaceParams (M)=3.7682026.0691.681.90.23
BaRAParams (M)=5.6082026.0691.252.20.21
FVAE-LoRAParams (M)=33.532026.0691.212.080.23
ScalaBLParams (M)=3.7692026.0690.92.430.23
C-LoRAParams (M)=4.2752026.0689.95.230.31
OracleModel Size=Large, Expertise Distribution=Dist. 32026.0471.6——
OracleModel Size=Large2026.0471.6——
OracleModel Size=Large, Noise Type=Uniform2026.0471.6——
REALMModel Size=Large, Expertise Distribution=Dist. 2, Noise Type=Systematic2026.0471——
REALMModel Size=Large, Expertise Distribution=Dist. 2, Noise Type=Uniform2026.0470.68——
REALMModel Size=Large, Noise Type=Uniform, Expertise Distribution=Dist. 32026.0470.12——
REALMModel Size=Large, Noise Type=Systematic, Expertise Distribution=Dist. 32026.0470.12——
REALMModel Size=Large, Expertise Distribution=Dist. 3, Noise Type=Uniform2026.0470.12——
REALMModel Size=Large, Noise Type=Asymmetric, Expertise Distribution=Dist. 32026.0469.8——
REALMModel Size=Large, Expertise Distribution=Dist. 1, Noise Type=Uniform2026.0469.68——
noisyModel Size=Large, Expertise Distribution=Dist. 2, Noise Type=Systematic2026.0469.32——
NOISYModel Size=Large, Expertise Distribution=Dist. 2, Noise Type=Uniform2026.0468.8——
NOISYModel Size=Large, Noise Type=Uniform, Expertise Distribution=Dist. 32026.0465.68——
NOISYModel Size=Large, Expertise Distribution=Dist. 3, Noise Type=Uniform2026.0465.68——
NOISYModel Size=Large, Noise Type=Systematic, Expertise Distribution=Dist. 32026.0465.4——
NOISYModel Size=Large, Noise Type=Asymmetric, Expertise Distribution=Dist. 32026.0464.8——
NOISYModel Size=Large, Expertise Distribution=Dist. 1, Noise Type=Uniform2026.0461.36——
LLAMA 1Size=65B2023.0760.2——
LLAMA 2Size=70B2023.0760.2——
OracleModel Size=Base, Expertise Distribution=Dist. 32026.0459.6——
OracleModel Size=Base2026.0459.6——
OracleModel Size=Base, Noise Type=Uniform2026.0459.6——
REALMModel Size=Base, Expertise Distribution=Dist. 2, Noise Type=Uniform2026.0458.68——
LLAMA 1Size=33B2023.0758.6——
LLAMA 2Size=7B2023.0758.6——
REALMModel Size=Base, Noise Type=Systematic, Expertise Distribution=Dist. 32026.0458.6——
LLAMA 2Size=34B2023.0758.2——
REALMModel Size=Base, Noise Type=Uniform, Expertise Distribution=Dist. 32026.0457.96——
REALMModel Size=Base, Expertise Distribution=Dist. 3, Noise Type=Uniform2026.0457.96——
REALMModel Size=Base, Expertise Distribution=Dist. 1, Noise Type=Uniform2026.0457.88——
REALMModel Size=Base, Noise Type=Asymmetric, Expertise Distribution=Dist. 32026.0457.72——
LLAMA 1Size=7B2023.0757.2——
LLAMA 2Size=13B2023.0757——
NOISYModel Size=Base, Expertise Distribution=Dist. 2, Noise Type=Uniform2026.0456.84——
FalconSize=40B2023.0756.6——
LLAMA 1Size=13B2023.0756.4——
NOISYModel Size=Base, Noise Type=Asymmetric, Expertise Distribution=Dist. 32026.0455.12——
NOISYModel Size=Base, Noise Type=Uniform, Expertise Distribution=Dist. 32026.0455——
NOISYModel Size=Base, Expertise Distribution=Dist. 3, Noise Type=Uniform2026.0455——
NOISYModel Size=Base, Noise Type=Systematic, Expertise Distribution=Dist. 32026.0454.36——
MPTSize=30B2023.0752——
FalconSize=7B2023.0751.6——
MPTSize=7B2023.0751.4——
NOISYModel Size=Base, Expertise Distribution=Dist. 1, Noise Type=Uniform2026.0449.12——
Token DropModel=OLMoE-Instruct, Capacity Factor=2.02025.0347.8——
Expanded DropModel=Mixtral-8x7B-Instruct, Capacity Factor=1.52025.0347.8——
BaselineModel=OLMoE-Instruct, Capacity Factor=2.02025.0347.6——
BaselineModel=Mixtral-8x7B-Instruct, Capacity Factor=1.52025.0347.4——
Expanded DropModel=OLMoE-Instruct, Capacity Factor=2.02025.0347.2——
Expert DropModel=Mixtral-8x7B-Instruct, Capacity Factor=1.5, Drop Strategy=skips one out of eight experts2025.0346.8——
Token DropModel=Mixtral-8x7B-Instruct, Capacity Factor=1.52025.0346.4——
EntailerBackbone=T5-large, Fine-tuning=none, Fact corpus=provided with dataset2023.0545.8——
BaselineModel=DeepSeek-V2-Lite-Chat, Capacity Factor=2.02025.0345.4——
Expanded DropModel=DeepSeek-V2-Lite-Chat, Capacity Factor=2.02025.0345.4——
Token DropModel=DeepSeek-V2-Lite-Chat, Capacity Factor=2.02025.0345.2——
Expert DropModel=OLMoE-Instruct, Capacity Factor=2.0, Drop Strategy=bottom 10% of lowest-load experts2025.0344.6——
REALMModel Size=Small, Expertise Distribution=Dist. 2, Noise Type=Uniform2026.0444.28——
REALMModel Size=Small, Expertise Distribution=Dist. 2, Noise Type=Uniform2026.0444.28——
OracleModel Size=Small, Expertise Distribution=Dist. 32026.0444.2——
OracleModel Size=Small2026.0444.2——
OracleModel Size=Small, Noise Type=Uniform2026.0444.2——
Expanded DropModel=Qwen1.5-MoE-Chat, Capacity Factor=1.52025.0343.4——
REALMModel Size=Small, Noise Type=Systematic, Expertise Distribution=Dist. 32026.0442.56——
REALMModel Size=Small, Expertise Distribution=Dist. 3, Noise Type=Systematic2026.0442.56——
REALMModel Size=Small, Noise Type=Uniform, Expertise Distribution=Dist. 32026.0442.52——
REALMModel Size=Small, Expertise Distribution=Dist. 3, Noise Type=Uniform2026.0442.52——
REALMModel Size=Small, Noise Type=Asymmetric, Expertise Distribution=Dist. 32026.0442.48——
BaselineModel=Qwen1.5-MoE-Chat, Capacity Factor=1.52025.0342.4——
Expert DropModel=DeepSeek-V2-Lite-Chat, Capacity Factor=2.0, Drop Strategy=bottom 10% of lowest-load experts2025.0341.8——
Expert DropModel=Qwen1.5-MoE-Chat, Capacity Factor=1.5, Drop Strategy=bottom 10% of lowest-load experts2025.0341.4——
Token DropModel=Qwen1.5-MoE-Chat, Capacity Factor=1.52025.0340.4——
noisyModel Size=Small, Expertise Distribution=Dist. 2, Noise Type=Uniform2026.0439.52——
NOISYModel Size=Small, Expertise Distribution=Dist. 2, Noise Type=Uniform2026.0439.52——
REALMModel Size=Small, Expertise Distribution=Dist. 1, Noise Type=Uniform2026.0439.44——
NOISYModel Size=Small, Noise Type=Uniform, Expertise Distribution=Dist. 32026.0435.8——
NOISYModel Size=Small, Expertise Distribution=Dist. 3, Noise Type=Uniform2026.0435.8——
NOISYModel Size=Small, Noise Type=Asymmetric, Expertise Distribution=Dist. 32026.0435.76——
NOISYModel Size=Small, Noise Type=Systematic, Expertise Distribution=Dist. 32026.0435——
noisyModel Size=Small, Expertise Distribution=Dist. 3, Noise Type=Systematic2026.0435——
NOISYModel Size=Small, Expertise Distribution=Dist. 1, Noise Type=Uniform2026.0430.28——