List your company
ResearchFormerly WizWand

Single-Hop Question Answering on NQ (EM)

51.7Exact Match (EM)Tree-GRPO
5.31617.35829.441.442Sep 25, 2025Nov 3, 2025Dec 12, 2025Jan 20, 2026Feb 28, 2026Apr 8, 2026May 18, 2026
Updated 5mo ago

Evaluation results

MethodLinks
Tree-GRPOBackbone=Qwen2.5-14b2025.0951.7
GRPOBackbone=Qwen2.5-14b2025.0951.3
GSPOBackbone=Qwen2.5-14b2025.0950.7
MR-Search-BaseTraining Pipeline=Pure RL, External Teacher=None, Backbone=Qwen2.5-7B2026.0550.2
SD-Search-BaseTraining Pipeline=Pure RL, External Teacher=None, Backbone=Qwen2.5-7B2026.0550
SD-Search-InstructTraining Pipeline=Pure RL, External Teacher=None, Backbone=Qwen2.5-7B2026.0549.5
AutoRefineBackbone=Qwen2.5-7B-Instruct2026.0348.4
AutoRefine-BaseTraining Pipeline=Pure RL, External Teacher=None, Backbone=Qwen2.5-7B2026.0548.4
Tree-GRPOBackbone=Qwen2.5-7b2025.0948.1
SKILLGRAPHEvaluation Protocol=In-domain2026.0548
Tree-GRPOBackbone=Llama3.2-3b2025.0947.7
GSPOBackbone=Qwen2.5-7b2025.0947
Search-R1-BaseTraining Pipeline=Pure RL, External Teacher=None, Backbone=Qwen2.5-7B2026.0546.9
Tree-GRPOBackbone=Qwen2.5-3b2025.0946.8
AutoRefineBackbone=Qwen2.5-3B-Instruct2026.0346.7
SkillRLEvaluation Protocol=In-domain2026.0545.9
GRPOBackbone=Qwen2.5-7b2025.0945.8
GRPOBackbone=Llama3.2-3b2025.0945.5
Thinker-InstructTraining Pipeline=Synthesize → SFT, External Teacher=Qwen2.5-72B, Backbone=Qwen2.5-7B2026.0545
GRPOBackbone=Qwen2.5-3b2025.0944.4
Tree-GRPOBackbone=Qwen2.5-1.5b2025.0943.6
ZeroSearchEvaluation Protocol=In-domain2026.0543.6
EvolveREvaluation Protocol=In-domain2026.0543.5
GSPOBackbone=Qwen2.5-3b2025.0943
GSPOBackbone=Llama3.2-3b2025.0941.2
GRPOBackbone=Qwen2.5-1.5b2025.0939.4
Search-R1Backbone=Qwen2.5-7B-Instruct2026.0339.3
Search-R1Evaluation Protocol=In-domain2026.0539.3
Search-R1-InstructTraining Pipeline=Pure RL, External Teacher=None, Backbone=Qwen2.5-7B2026.0539.3
EvalActBackbone=Qwen2.5-3B-Instruct2026.0338.5
EvalActBackbone=Qwen2.5-7B-Instruct2026.0338.5
GSPOBackbone=Qwen2.5-1.5b2025.0936.8
ReActBackbone=Qwen2.5-14b2025.0936.1
Naïve RAGBackbone=Qwen2.5-7B-Instruct2026.0334.9
Naïve RAGBackbone=Qwen2.5-3B-Instruct2026.0334.8
Search-o1Backbone=Qwen2.5-14b2025.0934.7
Search-R1Backbone=Qwen2.5-3B-Instruct2026.0334.1
ReActBackbone=Qwen2.5-7b2025.0930.6
RAGEvaluation Protocol=In-domain2026.0527.4
Search-o1Backbone=Llama3.2-3b2025.0924.2
ReActBackbone=Llama3.2-3b2025.0923.9
Search-o1Backbone=Qwen2.5-3B-Instruct2026.0323.8
Search-o1Backbone=Qwen2.5-7b2025.0923.8
IRCoTBackbone=Qwen2.5-7B-Instruct2026.0322.4
ReActBackbone=Qwen2.5-3b2025.0921.1
Direct InferenceBackbone=Qwen2.5-14b2025.0919.8
Search-o1Evaluation Protocol=In-domain2026.0519.4
Direct InferenceBackbone=Llama3.2-3b2025.0916.2
Search-o1Backbone=Qwen2.5-7B-Instruct2026.0315.1
Search-o1Backbone=Qwen2.5-3b2025.0915.1
Direct GenerationBackbone=Qwen2.5-7B-Instruct2026.0313.4
Direct InferenceBackbone=Qwen2.5-7b2025.0913.4
CoTEvaluation Protocol=In-domain2026.0512.8
Qwen2.5-7B-InstructEvaluation Protocol=In-domain2026.0511.6
IRCoTBackbone=Qwen2.5-3B-Instruct2026.0311.1
Direct GenerationBackbone=Qwen2.5-3B-Instruct2026.0310.6
Direct InferenceBackbone=Qwen2.5-3b2025.0910.6
Search-o1Backbone=Qwen2.5-1.5b2025.0910.2
ReActBackbone=Qwen2.5-1.5b2025.099.5
Direct InferenceBackbone=Qwen2.5-1.5b2025.097.1