List your company
ResearchFormerly WizWand

Summarization on TL;DR (Winrate)

91.8WinrateSignCert-PO
2.67225.81148.9572.089Mar 5, 2026Mar 18, 2026Apr 1, 2026Apr 15, 2026Apr 28, 2026May 12, 2026May 26, 2026
Updated 5mo ago

Evaluation results

MethodLinks
SignCert-POBackbone=Qwen2.5, Model Scale=3B, Gold Reward Model=GPT-4.1 Nano2026.0491.8
Dr.GRPOBackbone=Qwen2.5, Model Scale=3B, Gold Reward Model=GPT-4.1 Nano2026.0490.2
AdvPOBackbone=Qwen2.5, Model Scale=3B, Gold Reward Model=GPT-4.1 Nano2026.0490
CW-DPOAlignment Algorithm=DPO, Model Pair=Qwen2.5-0.5B → Qwen2.5-14B2026.0386.5
WS-DPOAlignment Algorithm=rDPO, Model Pair=Qwen2.5-0.5B → Qwen2.5-14B2026.0385.9
CW-IPOAlignment Algorithm=IPO, Model Pair=Qwen2.5-0.5B → Qwen2.5-14B2026.0384.6
CW-rDPOAlignment Algorithm=rDPO, Model Pair=Qwen2.5-0.5B → Qwen2.5-14B2026.0383.9
WS-DPOAlignment Algorithm=DPO, Model Pair=Qwen2.5-0.5B → Qwen2.5-14B2026.0381.8
UWOBackbone=Qwen2.5, Model Scale=3B, Gold Reward Model=GPT-4.1 Nano2026.0481.8
HumanAlignment Algorithm=DPO, Model Pair=Qwen2.5-0.5B → Qwen2.5-14B2026.0380.8
HumanAlignment Algorithm=IPO, Model Pair=Qwen2.5-0.5B → Qwen2.5-14B2026.0380.3
BSPOBackbone=Qwen2.5, Model Scale=1.5B, Gold Reward Model=Skywork-Reward-Llama-3.1-8B-v0.22026.0478.3
HumanAlignment Algorithm=rDPO, Model Pair=Qwen2.5-0.5B → Qwen2.5-14B2026.0377.6
WS-DPOAlignment Algorithm=IPO, Model Pair=Qwen2.5-0.5B → Qwen2.5-14B2026.0374.6
MultiMetricModel=LLaMA, Baseline Comparison=MPO2026.0574.6
SignCert-POBackbone=Pythia, Model Scale=2.8B, Gold Reward Model=GPT-4.1 Nano2026.0473.8
AdvPOBackbone=Pythia, Model Scale=2.8B, Gold Reward Model=GPT-4.1 Nano2026.0469.1
Dr.GRPOBackbone=Pythia, Model Scale=2.8B, Gold Reward Model=GPT-4.1 Nano2026.0468
Dr.GRPOBackbone=Qwen2.5, Model Scale=1.5B, Gold Reward Model=Skywork-Reward-Llama-3.1-8B-v0.22026.0466.9
SignCert-POBackbone=Qwen2.5, Model Scale=1.5B, Gold Reward Model=Skywork-Reward-Llama-3.1-8B-v0.22026.0466
TUR-DPOEvaluation Protocol=LLM judge2026.0464.8
TUR-DPOEvaluation Protocol=Human eval2026.0464.1
PPOEvaluation Protocol=Human eval2026.0463.7
CW-IPOAlignment Algorithm=IPO, Model Pair=OPT-125M → OPT-13B2026.0363.2
BSPOBackbone=Qwen2.5, Model Scale=3B, Gold Reward Model=GPT-4.1 Nano2026.0463.1
IPOEvaluation Protocol=LLM judge2026.0461.9
MultiMetricModel=GPT-J, Baseline Comparison=MPO2026.0561.6
DPOEvaluation Protocol=LLM judge2026.0461.2
CW-rDPOAlignment Algorithm=rDPO, Model Pair=OPT-125M → OPT-13B2026.0361
DPOEvaluation Protocol=Human eval2026.0460.5
SignCert-POBackbone=Pythia, Model Scale=1B, Gold Reward Model=Skywork-Reward-Llama-3.1-8B-v0.22026.0460
UWOBackbone=Qwen2.5, Model Scale=1.5B, Gold Reward Model=Skywork-Reward-Llama-3.1-8B-v0.22026.0459.6
HumanAlignment Algorithm=IPO, Model Pair=OPT-125M → OPT-13B2026.0359.5
MultiMetricModel=DeepSeek, Baseline Comparison=MPO2026.0558.6
AdvPOBackbone=Qwen2.5, Model Scale=1.5B, Gold Reward Model=Skywork-Reward-Llama-3.1-8B-v0.22026.0458.1
HumanAlignment Algorithm=rDPO, Model Pair=OPT-125M → OPT-13B2026.0356.6
HumanAlignment Algorithm=DPO, Model Pair=OPT-125M → OPT-13B2026.0355.7
MultiMetricModel=BART, Baseline Comparison=MPO2026.0554.8
WS-DPOAlignment Algorithm=rDPO, Model Pair=OPT-125M → OPT-13B2026.0354.5
WS-DPOAlignment Algorithm=IPO, Model Pair=OPT-125M → OPT-13B2026.0353.4
WS-DPOAlignment Algorithm=DPO, Model Pair=OPT-125M → OPT-13B2026.0352.8
CW-DPOAlignment Algorithm=DPO, Model Pair=OPT-125M → OPT-13B2026.0352.8
BSPOBackbone=Pythia, Model Scale=2.8B, Gold Reward Model=GPT-4.1 Nano2026.0450
MultiMetricModel=BART, Baseline Comparison=SFT2026.0547.2
AdvPOBackbone=Pythia, Model Scale=1B, Gold Reward Model=Skywork-Reward-Llama-3.1-8B-v0.22026.0447
MultiMetricModel=GPT-J, Baseline Comparison=SFT2026.0546.8
MultiMetricModel=LLaMA, Baseline Comparison=SFT2026.0543.4
MultiMetricModel=GPT-J, Baseline Comparison=RLHF2026.0542.8
MultiMetricModel=DeepSeek, Baseline Comparison=SFT2026.0540.8
BSPOBackbone=Pythia, Model Scale=1B, Gold Reward Model=Skywork-Reward-Llama-3.1-8B-v0.22026.0440.4
MultiMetricModel=BART, Baseline Comparison=RLHF2026.0540.4
MultiMetricModel=LLaMA, Baseline Comparison=RLHF2026.0539.2
UWOBackbone=Pythia, Model Scale=2.8B, Gold Reward Model=GPT-4.1 Nano2026.0438.1
SFTBackbone=Qwen2.5, Model Scale=3B, Gold Reward Model=GPT-4.1 Nano2026.0435.6
SFTBackbone=Qwen2.5, Model Scale=1.5B, Gold Reward Model=Skywork-Reward-Llama-3.1-8B-v0.22026.0427.3
SFTBackbone=Pythia, Model Scale=2.8B, Gold Reward Model=GPT-4.1 Nano2026.0425.2
Dr.GRPOBackbone=Pythia, Model Scale=1B, Gold Reward Model=Skywork-Reward-Llama-3.1-8B-v0.22026.0421
SFTBackbone=Pythia, Model Scale=1B, Gold Reward Model=Skywork-Reward-Llama-3.1-8B-v0.22026.0415.7
UWOBackbone=Pythia, Model Scale=1B, Gold Reward Model=Skywork-Reward-Llama-3.1-8B-v0.22026.046.1