List your company

Question Answering on 2Wiki (test)

61.8EM AccuracyICRL
-2.47214.21430.947.586Oct 1, 2025Nov 14, 2025Dec 29, 2025Feb 12, 2026Mar 28, 2026May 12, 2026Jun 26, 2026

Evaluation results

MethodLinks
ICRLModel=Qwen2.5-14B2026.0361.8——
INTRAGenerator=T5Gemma22026.0549.2—53.2
INTRARetrieval method=INTRA, Generator=T5Gemma2026.0549.2——
Thinker-InstructTraining Pipeline=Synthesize → SFT, External Teacher=Qwen2.5-72B, Retrieval Strategy=w/ Multi-Hop Retrieval, Backbone=Qwen2.5-3B2026.0546.9——
Qwen3-Emb-4B + RerankerGenerator=T5Gemma22026.0546.8—50.8
Qwen3-Emb-4B + RerankerRetrieval method=Qwen3-Emb-4B + Reranker, Generator=T5Gemma2026.0546.8——
BGEGenerator=T5Gemma22026.0546.1—49.6
BGERetrieval method=BGE, Generator=T5Gemma2026.0546.1——
Hybrid RAGGenerator=T5Gemma22026.0546—49.9
Qwen3-Emb-4BGenerator=T5Gemma22026.0546—50.1
Hybrid RAGRetrieval method=Hybrid RAG, Generator=T5Gemma2026.0546——
Qwen3-Emb-4BRetrieval method=Qwen3-Emb-4B, Generator=T5Gemma2026.0546——
Qwen3-Emb-0.6BGenerator=T5Gemma22026.0545.7—49.8
Qwen3-Emb-0.6BRetrieval method=Qwen3-Emb-0.6B, Generator=T5Gemma2026.0545.7——
SD-Search-BaseTraining Pipeline=Pure RL, External Teacher=–, Retrieval Strategy=w/ Multi-Hop Retrieval, Backbone=Qwen2.5-3B2026.0542——
SD-Search-InstructTraining Pipeline=Pure RL, External Teacher=–, Retrieval Strategy=w/ Multi-Hop Retrieval, Backbone=Qwen2.5-3B2026.0541.9——
BM25Generator=T5Gemma22026.0541.7—45.2
BM25Retrieval method=BM25, Generator=T5Gemma2026.0541.7——
MaxSimGenerator=T5Gemma22026.0541.6—45.8
MaxSimRetrieval method=MaxSim, Generator=T5Gemma2026.0541.6——
MR-Search-BaseTraining Pipeline=Pure RL, External Teacher=–, Retrieval Strategy=w/ Multi-Hop Retrieval, Backbone=Qwen2.5-3B2026.0540.1——
AutoRefine-BaseTraining Pipeline=Pure RL, External Teacher=–, Retrieval Strategy=w/ Multi-Hop Retrieval, Backbone=Qwen2.5-3B2026.0539.3——
TF-IDFGenerator=T5Gemma22026.0539—42.7
TF-IDFRetrieval method=TF-IDF, Generator=T5Gemma2026.0539——
AutoRefine-InstructTraining Pipeline=Pure RL, External Teacher=–, Retrieval Strategy=w/ Multi-Hop Retrieval, Backbone=Qwen2.5-3B2026.0538——
R2-Searcher2026.0636.7—42.6
StepSearch-BaseTraining Pipeline=Pure RL, External Teacher=GPT-4o, Retrieval Strategy=w/ Multi-Hop Retrieval, Backbone=Qwen2.5-3B2026.0533.9——
AutoRefine2026.0633.4—38.8
StepSearch-InstructTraining Pipeline=Pure RL, External Teacher=GPT-4o, Retrieval Strategy=w/ Multi-Hop Retrieval, Backbone=Qwen2.5-3B2026.0532——
Search-R1-InstructTraining Pipeline=Pure RL, External Teacher=–, Retrieval Strategy=w/ Multi-Hop Retrieval, Backbone=Qwen2.5-3B2026.0531——
Search-R12026.0630.1—35.7
DirectModel=Qwen2.5-14B2026.0328.2——
R1-InstructTraining Pipeline=Pure RL, External Teacher=–, Retrieval Strategy=w/o Retrieval, Backbone=Qwen2.5-3B2026.0527.5——
Search-R1-BaseTraining Pipeline=Pure RL, External Teacher=–, Retrieval Strategy=w/ Multi-Hop Retrieval, Backbone=Qwen2.5-3B2026.0527.4——
ReSearch2026.0627.4—33.6
ReSearch-BaseTraining Pipeline=Pure RL, External Teacher=–, Retrieval Strategy=w/ Multi-Hop Retrieval, Backbone=Qwen2.5-3B2026.0527.2——
ReSearch-InstructTraining Pipeline=Pure RL, External Teacher=–, Retrieval Strategy=w/ Multi-Hop Retrieval, Backbone=Qwen2.5-3B2026.0527.2——
R1-BaseTraining Pipeline=Pure RL, External Teacher=–, Retrieval Strategy=w/o Retrieval, Backbone=Qwen2.5-3B2026.0526.8——
CoTModel=Qwen2.5-14B2026.0325.8——
SFTTraining Pipeline=SFT, External Teacher=–, Retrieval Strategy=w/o Retrieval, Backbone=Qwen2.5-3B2026.0524.8——
Direct GenerationTraining Pipeline=Zero-Shot, External Teacher=–, Retrieval Strategy=w/o Retrieval, Backbone=Qwen2.5-3B2026.0524.4——
Naive RAGTraining Pipeline=Zero-Shot, External Teacher=–, Retrieval Strategy=w/ Single-Hop Retrieval, Backbone=Qwen2.5-3B2026.0522.6——
Search-o1Training Pipeline=Prompting, External Teacher=–, Retrieval Strategy=w/ Multi-Hop Retrieval, Backbone=Qwen2.5-3B2026.0521.8——
ZeroSearch2026.0621.3—25.9
Direct Inference2026.0621.1—24.9
COT2026.0619.4—22.9
IRCoTTraining Pipeline=Prompting, External Teacher=–, Retrieval Strategy=w/ Multi-Hop Retrieval, Backbone=Qwen2.5-3B2026.0517.1——
RAG2026.067.88—11
Search-o12026.067.74—12.8
IRCOT2026.067.15—13.6
AFHmode=Absolute2025.104.718.5—
MASH w/ OTC-ST2025.104.632.5—
MASH w/ OTC2025.104.131.3—
MASH w/ EXP2025.103.832.2—
Promptingshots=5-shot2025.103.610.9—
DPO2025.103.331.6—
AFHmode=Multisample2025.102.629.2—
OTC2025.100.7524.1—
Ternary2025.100——