List your company

Multi-step interaction on 20Q

32.1WinrateNLAC
11.61216.93122.2527.569Dec 4, 2025

Evaluation results

MethodLinks
NLACParadigm=Fine-tuning, Base Model=QwQ-32B2025.1232.1
NLRLParadigm=Fine-tuning, Base Model=QwQ-32B2025.1231.8
NLQLParadigm=Fine-tuning, Base Model=QwQ-32B, Notes=ablation2025.1231.4
ReActParadigm=Prompting, Base Model=GPT52025.1230.2
Self-DistillationParadigm=Fine-tuning, Base Model=QwQ-32B, Notes=ablation2025.1226.8
NLACParadigm=Fine-tuning, Base Model=Qwen2.5-7B-Instruct2025.1226
NLRLParadigm=Fine-tuning, Base Model=Qwen2.5-7B-Instruct2025.1225.8
GRPOParadigm=Fine-tuning, Base Model=QwQ-32B2025.1225.6
NLQLParadigm=Fine-tuning, Base Model=Qwen2.5-7B-Instruct, Notes=ablation2025.1224.2
PPOParadigm=Fine-tuning, Base Model=QwQ-32B2025.1224
RFTParadigm=Fine-tuning, Base Model=QwQ-32B2025.1222
GRPOParadigm=Fine-tuning, Base Model=Qwen2.5-7B-Instruct2025.1218.4
PPOParadigm=Fine-tuning, Base Model=Qwen2.5-7B-Instruct2025.1217.2
RFTParadigm=Fine-tuning, Base Model=Qwen2.5-7B-Instruct2025.1212.6
Self-DistillationParadigm=Fine-tuning, Base Model=Qwen2.5-7B-Instruct, Notes=ablation2025.1212.4