List your company

Downstream Task Evaluation on LMEval Harness (MMLU, ARC-C, HellaSwag, TruthfulQA, Winogrande, GSM8K)

65.8MMLUInitial Model
57.27259.48661.763.914Feb 12, 2026

Evaluation results

MethodLinks
Initial ModelBackbone=Llama-3, Model Type=Instruct, Parameter Scale=8B2026.0265.856.45936.177.374.861.6
DPOBackbone=Llama-3, Model Type=Instruct, Parameter Scale=8B2026.0265.661.360.642.476.473.563.3
HyPOBackbone=Llama-3, Model Type=Instruct, Parameter Scale=8B2026.026564.458.745.373.370.362.8
SimPOBackbone=Llama-3, Model Type=Instruct, Parameter Scale=8B2026.0264.759.85643.872.571.361.4
Initial ModelBackbone=Llama-3, Model Type=Base, Parameter Scale=8B2026.0263.756.961.431.577.450.356.9
HyPOBackbone=Llama-3, Model Type=Base, Parameter Scale=8B2026.0263.665.464.342.878.949.760.8
DPOBackbone=Llama-3, Model Type=Base, Parameter Scale=8B2026.0263.361.664.737.97854.760
SimPOBackbone=Llama-3, Model Type=Base, Parameter Scale=8B2026.0262.163.464.53877.250.459.3
HyPOBackbone=Mistral, Model Type=Instruct, Parameter Scale=7B2026.0259.666.867.45477.640.661
Initial ModelBackbone=Mistral, Model Type=Instruct, Parameter Scale=7B2026.0259.258.766.152.678.144.159.8
Initial ModelBackbone=Mistral, Model Type=Base, Parameter Scale=7B2026.025954.260.82877.734.652.4
DPOBackbone=Mistral, Model Type=Instruct, Parameter Scale=7B2026.0258.860.967.651.278.341.659.7
HyPOBackbone=Mistral, Model Type=Base, Parameter Scale=7B2026.0258.560.263.539.178.931.855.3
SimPOBackbone=Mistral, Model Type=Base, Parameter Scale=7B2026.0258.458.663.633.178.13554.5
SimPOBackbone=Mistral, Model Type=Instruct, Parameter Scale=7B2026.0258.162.567.55577.439.860.1
DPOBackbone=Mistral, Model Type=Base, Parameter Scale=7B2026.0257.658.764.440.377.431.555