List your company
ResearchFormerly WizWand

Instruction Following on AlpacaFarm

59.2Win RateGR
10.94423.4723648.528Feb 20, 2026Feb 27, 2026Mar 6, 2026Mar 13, 2026Mar 20, 2026Mar 27, 2026Apr 3, 2026
Updated 7mo ago

Evaluation results

MethodLinks
GRModel Size=3B, Judge Model=GPT4.1-Nano2026.0259.2
KL RegModel Size=3B, Judge Model=GPT4.1-Nano2026.0252.8
SignCert-POBackbone=Qwen2.5, Model Scale=3B, Gold Reward Model=GPT-4.1 Nano2026.0452.3
AdvPOBackbone=Qwen2.5, Model Scale=3B, Gold Reward Model=GPT-4.1 Nano2026.0450
Reference ResetsModel Size=3B, Judge Model=GPT4.1-Nano2026.0249.2
SignCert-POBackbone=Qwen2.5, Model Scale=1.5B, Gold Reward Model=Skywork-Reward-Llama-3.1-8B-v0.22026.0447.9
SFTBackbone=Qwen2.5, Model Scale=3B, Gold Reward Model=GPT-4.1 Nano2026.0447.9
Dr.GRPOBackbone=Qwen2.5, Model Scale=3B, Gold Reward Model=GPT-4.1 Nano2026.0446.9
UWOBackbone=Qwen2.5, Model Scale=3B, Gold Reward Model=GPT-4.1 Nano2026.0446.9
No RegModel Size=3B, Judge Model=GPT4.1-Nano2026.0244.2
BSPOBackbone=Qwen2.5, Model Scale=3B, Gold Reward Model=GPT-4.1 Nano2026.0443.8
UWOBackbone=Qwen2.5, Model Scale=1.5B, Gold Reward Model=Skywork-Reward-Llama-3.1-8B-v0.22026.0438.3
SFTBackbone=Qwen2.5, Model Scale=1.5B, Gold Reward Model=Skywork-Reward-Llama-3.1-8B-v0.22026.0437.8
AdvPOBackbone=Qwen2.5, Model Scale=1.5B, Gold Reward Model=Skywork-Reward-Llama-3.1-8B-v0.22026.0437.5
Dr.GRPOBackbone=Qwen2.5, Model Scale=1.5B, Gold Reward Model=Skywork-Reward-Llama-3.1-8B-v0.22026.0432.6
BSPOBackbone=Qwen2.5, Model Scale=1.5B, Gold Reward Model=Skywork-Reward-Llama-3.1-8B-v0.22026.0431.3
GRModel Size=1.5B, Judge Model=GPT4.1-Nano2026.0229.2
KL RegModel Size=1.5B, Judge Model=GPT4.1-Nano2026.0227.6
Reference ResetsModel Size=1.5B, Judge Model=GPT4.1-Nano2026.0227.1
SFT ModelModel Size=3B, Judge Model=GPT4.1-Nano2026.0226.3
No RegModel Size=1.5B, Judge Model=GPT4.1-Nano2026.0221.7
SFT ModelModel Size=1.5B, Judge Model=GPT4.1-Nano2026.0220.6
GRModel Size=0.5B, Judge Model=GPT4.1-Nano2026.0218.5
Reference ResetsModel Size=0.5B, Judge Model=GPT4.1-Nano2026.0217.4
KL RegModel Size=0.5B, Judge Model=GPT4.1-Nano2026.0216.9
SFT ModelModel Size=0.5B, Judge Model=GPT4.1-Nano2026.0212.8
No RegModel Size=0.5B, Judge Model=GPT4.1-Nano2026.0212.8