List your company
ResearchFormerly WizWand

Dialogue on MT-Bench (test)

8.36GPT-4 ScoreLoRA
5.06325.91916.7757.6309Dec 26, 2025Dec 31, 2025Jan 6, 2026Jan 12, 2026Jan 18, 2026Jan 24, 2026Jan 30, 2026
Updated 8mo ago

Evaluation results

MethodLinks
LoRABackbone Model=Deepseek R1-1.5B, Trainable Parameters=1.21M2025.128.36
FAABackbone Model=Deepseek R1-1.5B, Trainable Parameters=1.63M2025.128.35
FFBackbone Model=Deepseek R1-1.5B, Trainable Parameters=1.5B2025.128.34
FourierFTBackbone Model=Deepseek R1-1.5B, Trainable Parameters=0.15M2025.128.33
AdapterHBackbone Model=Deepseek R1-1.5B, Trainable Parameters=1.63M2025.128.32
FFBackbone Model=LLaMA3 8B, Trainable Parameters=8.03B2025.128.17
FFBackbone Model=Qwen2 7B, Trainable Parameters=7.07B2025.127.88
LoRABackbone Model=Qwen2 7B, Trainable Parameters=5.40M2025.127.86
FAABackbone Model=Qwen2 7B, Trainable Parameters=7.30M2025.127.82
FourierFTBackbone Model=Qwen2 7B, Trainable Parameters=0.85M2025.127.81
AdapterHBackbone Model=Qwen2 7B, Trainable Parameters=7.29M2025.127.78
AdapterHBackbone Model=LLaMA3 8B, Trainable Parameters=8.73M2025.127.48
FAABackbone Model=LLaMA3 8B, Trainable Parameters=8.73M2025.127.45
FourierFTBackbone Model=LLaMA3 8B, Trainable Parameters=0.91M2025.127.41
LoRABackbone Model=LLaMA3 8B, Trainable Parameters=6.47M2025.127.4
GRPO + R2MBackbone=LLaMA3-8B-Instruct2026.017.3
RLOO + R2MBackbone=LLaMA3-8B-Instruct2026.017.3
GRPO + Pretrained RMBackbone=LLaMA3-8B-Instruct2026.017.1
RLOOBackbone=LLaMA3-8B-Instruct2026.017.1
RLOO + Pretrained RMBackbone=LLaMA3-8B-Instruct2026.017.1
ReMaxBackbone=LLaMA3-8B-Instruct2026.017
GRPOBackbone=LLaMA3-8B-Instruct2026.017
GRPO + Iterative RMHeadBackbone=LLaMA3-8B-Instruct2026.017
RLOO + Iterative RMHeadBackbone=LLaMA3-8B-Instruct2026.017
SFTBackbone=LLaMA3-8B-Instruct2026.016.9
REINFORCE++Backbone=LLaMA3-8B-Instruct2026.016.8
RLOO + R2MBackbone=Qwen2.5-3B-Instruct2026.016.7
GRPO + R2M w/o TrainBackbone=LLaMA3-8B-Instruct2026.016.7
GRPO + R2MBackbone=Qwen2.5-3B-Instruct2026.016.6
GRPO + Iterative RMHeadBackbone=Qwen2.5-3B-Instruct2026.016.5
RLOO + Pretrained RMBackbone=Qwen2.5-3B-Instruct2026.016.5
RLOO + Iterative RMHeadBackbone=Qwen2.5-3B-Instruct2026.016.5
RLOO + R2M w/o TrainBackbone=LLaMA3-8B-Instruct2026.016.5
SFTBackbone=Qwen2.5-3B-Instruct2026.016.4
ReMaxBackbone=Qwen2.5-3B-Instruct2026.016.4
GRPO + Pretrained RMBackbone=Qwen2.5-3B-Instruct2026.016.4
RLOOBackbone=Qwen2.5-3B-Instruct2026.016.4
REINFORCE++Backbone=Qwen2.5-3B-Instruct2026.016.3
GRPOBackbone=Qwen2.5-3B-Instruct2026.016.3
GRPO + R2M w/o TrainBackbone=Qwen2.5-3B-Instruct2026.016.2
RLOO + R2M w/o TrainBackbone=Qwen2.5-3B-Instruct2026.016.2
FAABackbone Model=LLaMA2 7B, Trainable Parameters=7.27M2025.125.24
AdapterHBackbone Model=LLaMA2 7B, Trainable Parameters=7.27M2025.125.23
LoRABackbone Model=LLaMA2 7B, Trainable Parameters=5.37M2025.125.22
FourierFTBackbone Model=LLaMA2 7B, Trainable Parameters=0.82M2025.125.21
FFBackbone Model=LLaMA2 7B, Trainable Parameters=6.94B2025.125.19