List your company
ResearchFormerly WizWand

Question Answering on Natural Questions (Accuracy)

46.39AccuracyRL
19.776426.685733.59540.5043Mar 26, 2024Aug 1, 2024Dec 8, 2024Apr 16, 2025Aug 23, 2025Dec 30, 2025May 8, 2026
Updated 5mo ago

Evaluation results

MethodLinks
RLBackbone=Llama-3.1-8B2026.0546.39
InternLM2-20BModel size=13 ~ 20B, shot=5-shot2024.0344.6
InternLM2-7BModel size=< 7B, shot=5-shot2024.0342.2
Mixtral-8x7B-v0.1Model size=13 ~ 20B, shot=5-shot2024.0339.6
ChatGLM3-6B-BaseModel size=< 7B, shot=5-shot2024.0338.7
Qwen-14BModel size=13 ~ 20B, shot=5-shot2024.0337.1
RLBackbone=OLMo-2-7B2026.0536.91
RFTBackbone=Llama-3.1-8B2026.0536.7
Llama2-13BModel size=13 ~ 20B, shot=5-shot2024.0334.5
RLBackbone=Qwen2.5-7B2026.0534.12
LLAMA2Size=13B, Tokens=2T, Context=4K2024.0431.2
Mistral-7B-v0.1Model size=< 7B, shot=5-shot2024.0331.1
DPOBackbone=Llama-3.1-8B2026.0530.48
InternLM2-20B-BaseModel size=13 ~ 20B, shot=5-shot2024.0330.3
BaseBackbone=Llama-3.1-8B2026.0530.15
SFTBackbone=Llama-3.1-8B2026.0529.45
Llama2-7BModel size=< 7B, shot=5-shot2024.0328.8
InternLM2-7B-BaseModel size=< 7B, shot=5-shot2024.0328.4
Baichuan2-13B-BaseModel size=13 ~ 20B, shot=5-shot2024.0327.5
RFTBackbone=Qwen2.5-7B2026.0527.03
LLAMA2Size=7B, Tokens=2T, Context=4K2024.0425.7
MEGALODONSize=7B, Tokens=2T, Context=32K2024.0425.7
SFTBackbone=OLMo-2-7B2026.0525.39
RFTBackbone=OLMo-2-7B2026.0525.33
Ours (SCD)decoding=SCD2025.0224.3
Baichuan2-7B-BaseModel size=< 7B, shot=5-shot2024.0324
SFTBackbone=Qwen2.5-7B2026.0523.67
MistralSize=7B, Context=16K2024.0423.2
DPOBackbone=OLMo-2-7B2026.0523.06
GemmaSize=8B, Tokens=6T, Context=8K2024.0423
DPOBackbone=Qwen2.5-7B2026.0522.94
BaseBackbone=OLMo-2-7B2026.0522.79
Qwen-7BModel size=< 7B, shot=5-shot2024.0322.7
Greedydecoding=Greedy2025.0222.1
BaseBackbone=Qwen2.5-7B2026.0521.42
MPTSize=7B, Tokens=1T, Context=4K2024.0420.8