List your company
ResearchFormerly WizWand

LLM Alignment Evaluation on AlpacaEval 2

51.9LC Win RateSDPO
9.98820.86931.7542.631Aug 25, 2025Oct 9, 2025Nov 24, 2025Jan 9, 2026Feb 24, 2026Apr 11, 2026May 27, 2026
Updated 5mo ago

Evaluation results

MethodLinks
SDPOBase Model=Qwen3-8B2026.0251.9———
SDPOBase Model=Olmo3-7B-DPO2026.0251.8———
Olmo3-7B-DPOModel Size=7B, Training=DPO2026.0250.4———
HyPOModel Family=Llama-3, Model State=Instruct, Model Parameters (B)=8B2026.0249.546.2——
Qwen3-8BModel Size=8B2026.0249.3———
AdaDPOConfiguration Selection=Best across 16 (lr, beta) combinations2026.0548.346.11.81,908
AAOBackbone=Llama3.1-8B-Instruct2025.1148.1144.89——
SimPOBackbone=Llama3.1-8B-Instruct2025.1147.543.64——
TIS-DPOBackbone=Llama3.1-8B-Instruct2025.1147.0443.99——
RTOBackbone=Llama3.1-8B-Instruct2025.1146.8441.98——
TR-DPOModel Family=Llama-3, Model State=Instruct, Model Parameters (B)=8B2026.0246.742.7——
RainbowPOModel Family=Llama-3, Model State=Instruct, Model Parameters (B)=8B2026.0246.743.5——
TDPOBackbone=Llama3.1-8B-Instruct2025.1146.5643.21——
DPOConfiguration Selection=Best across 16 (lr, beta) combinations2026.0546.444.51.81,933
SDPOBase Model=Qwen3-4B2026.0246.1———
SimPOModel Family=Llama-3, Model State=Instruct, Model Parameters (B)=8B2026.024643.1——
DPOBackbone=Llama3.1-8B-Instruct2025.1145.8744.34——
IPOBackbone=Llama3.1-8B-Instruct2025.1145.4344.56——
FocalPOModel Family=Llama-3, Model State=Instruct, Model Parameters (B)=8B2026.0245.143.6——
KTOBackbone=Llama3.1-8B-Instruct2025.1143.8642——
DPOModel Family=Llama-3, Model State=Instruct, Model Parameters (B)=8B2026.0240.941.3——
KTOModel Family=Llama-3, Model State=Instruct, Model Parameters (B)=8B2026.0240.539——
HyPOModel Family=Mistral, Model State=Instruct, Model Parameters (B)=7B2026.0238.947.9——
SimPOModel Family=Mistral, Model State=Instruct, Model Parameters (B)=7B2026.0238.440——
CPOModel Family=Llama-3, Model State=Instruct, Model Parameters (B)=8B2026.0238.140.4——
Qwen3-4BModel Size=4B2026.0237.9———
SLiC-HFModel Family=Llama-3, Model State=Instruct, Model Parameters (B)=8B2026.0236.736.8——
AAOBackbone=Mistral-7B-Instruct2025.1136.6638.54——
TR-DPOModel Family=Mistral, Model State=Instruct, Model Parameters (B)=7B2026.0236.533.7——
DPO-RandomBackbone=Llama3.1-8B-Instruct2025.1136.3635.1——
SimPOBackbone=Mistral-7B-Instruct2025.1135.9334.48——
FocalPOModel Family=Mistral, Model State=Instruct, Model Parameters (B)=7B2026.0235.935——
RainbowPOModel Family=Mistral, Model State=Instruct, Model Parameters (B)=7B2026.0235.733.9——
TDPOBackbone=Mistral-7B-Instruct2025.1135.3430.55——
SDPOBase Model=Olmo3-7B-SFT2026.0235.2———
DPOModel Family=Mistral, Model State=Instruct, Model Parameters (B)=7B2026.0235.131.4——
KTOModel Family=Mistral, Model State=Instruct, Model Parameters (B)=7B2026.023531.3——
CPOModel Family=Mistral, Model State=Instruct, Model Parameters (B)=7B2026.0234.939.9——
HyPOModel Family=Llama-3, Model State=Base, Model Parameters (B)=8B2026.0234.733.6——
KTOBackbone=Mistral-7B-Instruct2025.1134.6530.9——
TIS-DPOBackbone=Mistral-7B-Instruct2025.1134.5433.51——
RTOBackbone=Mistral-7B-Instruct2025.1134.4932.52——
Olmo3-7B-SFTModel Size=7B, Training=SFT2026.0234.3———
HyPOModel Family=Mistral, Model State=Base, Model Parameters (B)=7B2026.0232.829.6——
DPOBackbone=Mistral-7B-Instruct2025.1132.6228.7——
SLiC-HFModel Family=Mistral, Model State=Instruct, Model Parameters (B)=7B2026.0232.431.2——
TR-DPOModel Family=Llama-3, Model State=Base, Model Parameters (B)=8B2026.0231.830.2——
IPOBackbone=Mistral-7B-Instruct2025.1131.4630.31——
SimPOModel Family=Llama-3, Model State=Base, Model Parameters (B)=8B2026.0230.726.2——
RainbowPOModel Family=Llama-3, Model State=Base, Model Parameters (B)=8B2026.0230.327.1——
DPO-RandomBackbone=Mistral-7B-Instruct2025.1130.1428.45——
SFTConfiguration Selection=Best across 16 (lr, beta) combinations2026.0528.528.81.61,976
RainbowPOModel Family=Mistral, Model State=Base, Model Parameters (B)=7B2026.0228.426.7——
AAOBackbone=Llama3.1-8B-Base2025.1128.3640.23——
SimPOBackbone=Llama3.1-8B-Base2025.1127.4533.03——
SimPOModel Family=Mistral, Model State=Base, Model Parameters (B)=7B2026.0227.325.4——
FocalPOModel Family=Llama-3, Model State=Base, Model Parameters (B)=8B2026.0227.225.4——
TIS-DPOBackbone=Llama3.1-8B-Base2025.1126.8431.47——
FocalPOModel Family=Mistral, Model State=Base, Model Parameters (B)=7B2026.0225.819.7——
KTOBackbone=Llama3.1-8B-Base2025.1125.7924.79——
AAOBackbone=Mistral-7B-Base2025.1125.0123.45——
TR-DPOModel Family=Mistral, Model State=Base, Model Parameters (B)=7B2026.0224.921.4——
RTOBackbone=Llama3.1-8B-Base2025.1124.4325.84——
SimPOBackbone=Mistral-7B-Base2025.1124.422.8——
IPOBackbone=Llama3.1-8B-Base2025.1124.3824.9——
DPOModel Family=Llama-3, Model State=Base, Model Parameters (B)=8B2026.0224.321.9——
KTOModel Family=Llama-3, Model State=Base, Model Parameters (B)=8B2026.0223.620.3——
TDPOBackbone=Llama3.1-8B-Base2025.1123.3426.45——
PSFT → DPOBackbone=Qwen3-4B-Base, Training Dataset=UltraFeedback, Post-training=DPO2025.0823.2920.13——
DPOModel Family=Mistral, Model State=Base, Model Parameters (B)=7B2026.0222.618.5——
DPOBackbone=Llama3.1-8B-Base2025.1122.4531.3——
CPOModel Family=Llama-3, Model State=Base, Model Parameters (B)=8B2026.0222.324.6——
TIS-DPOBackbone=Mistral-7B-Base2025.1121.5718.84——
TDPOBackbone=Mistral-7B-Base2025.1121.4718.56——
DPOBackbone=Mistral-7B-Base2025.1120.4517.55——
RTOBackbone=Mistral-7B-Base2025.1120.416.77——
SLiC-HFModel Family=Llama-3, Model State=Base, Model Parameters (B)=8B2026.0219.815.9——
PSFTprolong → DPOBackbone=Qwen3-4B-Base, Training Dataset=UltraFeedback, Post-training=DPO2025.0819.2615.17——
DPO-RandomBackbone=Llama3.1-8B-Base2025.1119.0733.97——
DPO-RandomBackbone=Mistral-7B-Base2025.1117.7111.22——
IPOBackbone=Mistral-7B-Base2025.1117.1312.13——
SFT → DPOBackbone=Qwen3-4B-Base, Training Dataset=UltraFeedback, Post-training=DPO2025.0816.9613.4——
KTOBackbone=Mistral-7B-Base2025.1116.499.32——
CPOModel Family=Mistral, Model State=Base, Model Parameters (B)=7B2026.0213.111.6——
KTOModel Family=Mistral, Model State=Base, Model Parameters (B)=7B2026.0212.99.3——
SFTBackbone=Qwen3-4B-Base, Training Dataset=UltraFeedback2025.0812.248.52——
PSFTprolongBackbone=Qwen3-4B-Base, Training Dataset=UltraFeedback2025.0811.958.37——
PSFTBackbone=Qwen3-4B-Base, Training Dataset=UltraFeedback2025.0811.797.49——
SLiC-HFModel Family=Mistral, Model State=Base, Model Parameters (B)=7B2026.0211.69.1——