List your company
ResearchFormerly WizWand

Large Language Model Evaluation on MMLU, GSM8K, GPQA, HumanEval, TruthfulQA, and IFEVAL

70.7MMLUGRPO
57.49260.92164.3567.779May 26, 2025
Updated 8mo ago

Evaluation results

MethodLinks
GRPOevaluation_setup=Average across three base models (Llama-3.2-3B-Instruct, Llama-3.1-8B-Instruct, Mistral-7B-Instruct-v0.3)2025.0570.775.72864.359.9—62.1
TI-DPOevaluation_setup=Average across three base models (Llama-3.2-3B-Instruct, Llama-3.1-8B-Instruct, Mistral-7B-Instruct-v0.3)2025.057073266762—62.3
TIS-DPOevaluation_setup=Average across three base models (Llama-3.2-3B-Instruct, Llama-3.1-8B-Instruct, Mistral-7B-Instruct-v0.3)2025.0569.370.524.565.562.5—61.1
TPOevaluation_setup=Average across three base models (Llama-3.2-3B-Instruct, Llama-3.1-8B-Instruct, Mistral-7B-Instruct-v0.3)2025.0568.372.727.763.759—60.7
GRPOBackbone=Mistral-7B-v0.32025.056852951645650
CPOevaluation_setup=Average across three base models (Llama-3.2-3B-Instruct, Llama-3.1-8B-Instruct, Mistral-7B-Instruct-v0.3)2025.0567.370.72662.858.3—59.4
KTOevaluation_setup=Average across three base models (Llama-3.2-3B-Instruct, Llama-3.1-8B-Instruct, Mistral-7B-Instruct-v0.3)2025.0566.370.325.36257.7—58.7
cDPOevaluation_setup=Average across three base models (Llama-3.2-3B-Instruct, Llama-3.1-8B-Instruct, Mistral-7B-Instruct-v0.3)2025.0566.170.125.161.957.6—58.5
TI-DPOBackbone=Mistral-7B-v0.32025.056647753665949.7
DPOevaluation_setup=Average across three base models (Llama-3.2-3B-Instruct, Llama-3.1-8B-Instruct, Mistral-7B-Instruct-v0.3)2025.0565.369.3246156.7—57.7
TPOBackbone=Mistral-7B-v0.32025.056548850625347.7
TDPOevaluation_setup=Average across three base models (Llama-3.2-3B-Instruct, Llama-3.1-8B-Instruct, Mistral-7B-Instruct-v0.3)2025.056568.223.560.356.3—57
CPOBackbone=Mistral-7B-v0.32025.0564467.54961.55146.5
SFTevaluation_setup=Average across three base models (Llama-3.2-3B-Instruct, Llama-3.1-8B-Instruct, Mistral-7B-Instruct-v0.3)2025.05646822.759.355.5—56.7
Logic-RLevaluation_setup=Average across three base models (Llama-3.2-3B-Instruct, Llama-3.1-8B-Instruct, Mistral-7B-Instruct-v0.3)2025.0563.873.823.76155.6—57.9
SimPOevaluation_setup=Average across three base models (Llama-3.2-3B-Instruct, Llama-3.1-8B-Instruct, Mistral-7B-Instruct-v0.3)2025.0563.564.721.858.254.2—54.5
KTOBackbone=Mistral-7B-v0.32025.056345748615045.7
IPOevaluation_setup=Average across three base models (Llama-3.2-3B-Instruct, Llama-3.1-8B-Instruct, Mistral-7B-Instruct-v0.3)2025.056365.320.357.352.7—54.2
DPOBackbone=Mistral-7B-v0.32025.056244647605044.8
TDPOBackbone=Mistral-7B-v0.32025.0561435.546604843.9
SFTBackbone=Mistral-7B-v0.32025.05604254559.55444.2
IPOBackbone=Mistral-7B-v0.32025.055940343564741.3
SIMPOBackbone=Mistral-7B-v0.32025.055838442574540.7