List your company
ResearchFormerly WizWand

Question Answering on TabMWP (Accuracy)

88.04Accuracyzero-shot
1.678424.099246.5268.9408Apr 17, 2026Apr 28, 2026May 10, 2026May 22, 2026Jun 3, 2026Jun 15, 2026Jun 27, 2026
Updated 4mo ago

Evaluation results

MethodLinks
zero-shotSetting=Tuned LLM (LoRA)2026.0688.04
GRABSetting=Tuned LLM (LoRA)2026.0687.78
GRABSetting=Frozen LLM2026.0687.01
Prompt tuningSetting=Frozen LLM2026.0684.75
TAMOSetting=Frozen LLM2026.0684.49
TableLLamaSetting=Others2026.0680.05
GPT-5.4-miniSetting=Others2026.0672.06
NWCADModel=Ministral-3-8B2026.0454.2
AdaCADModel=Ministral-3-8B2026.0453.8
CoCoAModel=Ministral-3-8B2026.0453.2
CADModel=Ministral-3-8B2026.0452.2
NWCADModel=Llama-3.1-70B2026.0450
AdaCADModel=Llama-3.1-70B2026.0449.12
CoCoAModel=Llama-3.1-70B2026.0445.88
NWCADModel=Llama-3.1-8B2026.0436
zero-shotSetting=Inference-only2026.0634.75
AdaCADModel=Llama-3.1-8B2026.0434.4
With-contextModel=Ministral-3-8B2026.0431.8
CoCoAModel=Llama-3.1-8B2026.0430.2
CADModel=Llama-3.1-70B2026.0430
With-contextModel=Llama-3.1-70B2026.0428.38
CADModel=Llama-3.1-8B2026.0427.2
With-contextModel=Llama-3.1-8B2026.0418.4
BaselineModel=Llama-3.1-70B2026.046.62
BaselineModel=Llama-3.1-8B2026.045.2
BaselineModel=Ministral-3-8B2026.045