List your company
ResearchFormerly WizWand

LLM Alignment Evaluation on AlpacaEval 2.0 (test)

30.35LC Win RateOTPO
2.66529.852617.0424.2274May 18, 2025Jun 30, 2025Aug 12, 2025Sep 24, 2025Nov 6, 2025Dec 19, 2025Feb 1, 2026
Updated 8mo ago

Evaluation results

MethodLinks
OTPOBackbone=Mistral-7B-Instruct-v0.2, Training Dataset=HelpSteer22025.0530.3528.21,839
OTPOImplicit Token Weighting=true2025.0529.6429.541,991
BCOImplicit Token Weighting=false2025.0529.2328.761,955
IPOBase Model=Qwen-2-Instruct-7B2025.0529.0325.68411
RSOImplicit Token Weighting=false2025.0528.9328.81,961
R-DPOImplicit Token Weighting=false2025.0528.6628.341,940
CDPOImplicit Token Weighting=false2025.0528.6228.521,954
AOTImplicit Token Weighting=false2025.0528.5527.831,924
LDDPOImplicit Token Weighting=true2025.0528.5528.541,956
SPPOImplicit Token Weighting=false2025.0528.4828.131,946
LR-DPOImplicit Token Weighting=false2025.0528.2322.531,654
SGDPOBase Model=Llama-3.1-Instruct-8B2025.0528.2228.96444
LDDPOBackbone=Mistral-7B-Instruct-v0.2, Training Dataset=HelpSteer22025.0528.2225.841,815
BCOBase Model=Llama-3.1-Instruct-8B2025.0528.0329.32435
DPOImplicit Token Weighting=false2025.0527.9127.451,945
NCAImplicit Token Weighting=false2025.0527.7227.821,951
DPOBase Model=Llama-3.1-Instruct-8B2025.0527.5328.35438
SamPOBase Model=Llama-3.1-Instruct-8B2025.0527.4527.69443
RPOImplicit Token Weighting=false2025.0527.4327.411,963
SamPOBackbone=Mistral-7B-Instruct-v0.2, Training Dataset=HelpSteer22025.0527.3125.311,844
DPOBackbone=Mistral-7B-Instruct-v0.2, Training Dataset=HelpSteer22025.0527.2524.781,775
EXOImplicit Token Weighting=false2025.0527.2226.821,937
IPOBase Model=Llama-3.1-Instruct-8B2025.0527.0725.82459
SamPOImplicit Token Weighting=true2025.0526.9527.161,991
SFTBase Model=Llama-3.1-Instruct-8B2025.0526.8427.77459
SimPOImplicit Token Weighting=true2025.0526.7727.251,984
NCABase Model=Llama-3.1-Instruct-8B2025.0526.3327.77441
IPOImplicit Token Weighting=false2025.0525.0226.062,021
SamPOBase Model=Qwen-2-Instruct-7B2025.0524.5726.6426
DPOBase Model=Qwen-2-Instruct-7B2025.0524.2624.5414
SGDPOBase Model=Qwen-2-Instruct-7B2025.0523.8924.86419
BCOBase Model=Qwen-2-Instruct-7B2025.0523.7623.95411
SimPOBackbone=Mistral-7B-Instruct-v0.2, Training Dataset=HelpSteer22025.0523.56221,906
InitialImplicit Token Weighting=false2025.0522.9222.571,899
NCABase Model=Qwen-2-Instruct-7B2025.0521.9421.75409
SFTBase Model=Qwen-2-Instruct-7B2025.0520.9822.2418
OTPOBackbone=Qwen-2.5-3B-Instruct, Training Dataset=HelpSteer22025.0519.7122.062,074
SamPOBackbone=Qwen-2.5-3B-Instruct, Training Dataset=HelpSteer22025.0518.2620.752,105
InitialBackbone=Mistral-7B-Instruct-v0.2, Training Dataset=HelpSteer22025.0517.7415.051,630
LDDPOBackbone=Qwen-2.5-3B-Instruct, Training Dataset=HelpSteer22025.0517.3220.152,107
DPOBackbone=Qwen-2.5-3B-Instruct, Training Dataset=HelpSteer22025.0516.9319.622,100
InitialBackbone=Qwen-2.5-3B-Instruct, Training Dataset=HelpSteer22025.0516.8220.12,145
SimPOBackbone=Qwen-2.5-3B-Instruct, Training Dataset=HelpSteer22025.0516.7519.472,112
MixDPOBase model=LLaMA-3-8B, Preference dataset=Ultrafeedback2026.0214.4236.65—
vanilla DPOBase model=LLaMA-3-8B, Preference dataset=Ultrafeedback2026.029.3716.77—
CHESBase model=LLaMA-3-8B, Preference dataset=Ultrafeedback, Selected proportion=90%2026.028.1313.91—
CHESBase model=LLaMA-3-8B, Preference dataset=Ultrafeedback, Selected proportion=50%2026.026.1611.93—
Cal-DPOBase model=LLaMA-3-8B, Preference dataset=Ultrafeedback2026.024.567.45—
DPOPBase model=LLaMA-3-8B, Preference dataset=Ultrafeedback2026.024.534.04—
DPO + NLLBase model=LLaMA-3-8B, Preference dataset=Ultrafeedback2026.024.258.45—
Base-SFTBase model=LLaMA-3-8B, Preference dataset=Ultrafeedback2026.023.7310.19—