List your company

Question Answering on NarrativeQA (test)

76.2ROUGE-LCriSPO
0.07219.83639.659.364Dec 8, 2020Nov 5, 2021Oct 4, 2022Sep 2, 2023Jul 30, 2024Jun 28, 2025May 27, 2026

Evaluation results

MethodLinks
CriSPOModel=Claude Sonnet, Shots=0-shot2024.1076.2—————————
CriSPOLLM=Claude Sonnet, shot_count=0-shot2024.1076.2—————————
CriSPOModel=Claude Instant, Shots=5-shot2024.1076.1—————————
CriSPOLLM=Claude Instant, shot_count=5-shot2024.1076.1—————————
CriSPOModel=Claude Sonnet, Shots=5-shot2024.1075.2—————————
CriSPOLLM=Claude Sonnet, shot_count=5-shot2024.1075.2—————————
CriSPOModel=Claude Instant, Shots=0-shot2024.1075.1—————————
CriSPOLLM=Claude Instant, shot_count=0-shot2024.1075.1—————————
ManualModel=Claude Instant, Shots=5-shot2024.1067—————————
ManualLLM=Claude Instant, shot_count=5-shot2024.1067—————————
ManualModel=Claude Sonnet, Shots=5-shot2024.1066.7—————————
ManualLLM=Claude Sonnet, shot_count=5-shot2024.1066.7—————————
ManualModel=Claude Instant, Shots=0-shot2024.1064.2—————————
ManualLLM=Claude Instant, shot_count=0-shot2024.1064.2—————————
ManualModel=Claude Sonnet, Shots=0-shot2024.1064—————————
ManualLLM=Claude Sonnet, shot_count=0-shot2024.1064—————————
MemGenBackbone=Qwen3-8B2026.0163.94———————50.88—
Reasoner-onlyBackbone=Qwen3-8B2026.0163.81———————51—
PECAN2024.1060.2————61.1311——
OPROModel=Claude Instant2024.1059.4—————————
OPROLLM=Claude Instant2024.1059.4—————————
OPROModel=Claude Sonnet2024.1058.6—————————
OPROLLM=Claude Sonnet2024.1058.6—————————
A-MemBackbone=SmolLM3-3B2026.0155.99———————19.88—
In-context (Oracle)Backbone=QWEN2.5-7B, Setting=Oracle access to ground-truth context2026.0555.03————————55.99
DragonTop-k=72024.1054.2————55.137.51.21——
MemGenBackbone=Qwen2.5-1.5B-Instruct2026.0154.2———————42.63—
DragonTop-k=52024.1052.9————53.826.90.87——
BM25Top-k=72024.1052.9————53.737.51.21——
Llama-3.1-8B2024.1052.6————53.73,361.9108.45——
BM25Top-k=52024.1051.8————52.726.70.86——
RAPTOR-CTStrategy=collapsed tree, Top-k=202024.1051.2————5235.11.13——
A-MemBackbone=Qwen2.5-1.5B-Instruct2026.0149.85———————15.62—
LongLLMLingua2024.1049.5————50.51,789.457.72——
RAPTOR-CTStrategy=collapsed tree, Top-k=52024.1047.8————48.617.90.58——
MemGenBackbone=SmolLM3-3B2026.0147.37———————49.38—
RAGBackbone=Qwen3-8B2026.0147.29———————14.87—
RAGBackbone=QWEN2.5-7B, k=32026.0546.61————————47.36
A-MemBackbone=Qwen3-8B2026.0146.04———————15.88—
RAGBackbone=SmolLM3-3B2026.0140.12———————6.12—
RAPTOR-TTStrategy=tree traversal2024.1039.8————40.620.30.65——
In-context (Oracle)Backbone=QWEN2.5-0.5B, Setting=Oracle access to ground-truth context2026.0539.64————————40.65
SBERTTop-k=72024.1038.8————39.537.51.21——
LATENTGRAPHMEMBackbone=Qwen3-8B2026.0137.6———————51—
Reasoner-onlyBackbone=SmolLM3-3B2026.0137.43———————40.5—
Ours (Oracle Setting)Backbone=QWEN2.5-7B, Setting=Oracle access to adapter2026.0536.78————————37.55
SBERTTop-k=52024.1035.8————36.526.80.86——
LATENTGRAPHMEMBackbone=SmolLM3-3B2026.0135.38———————47—
RAGBackbone=QWEN2.5-7B, k=52026.0534.87————————35.57
LATENTGRAPHMEMBackbone=Qwen2.5-1.5B-Instruct2026.0132.25———————45.5—
Reader-to-Retriever DistillationInitialization=BM25, Iteration=12020.123235.37.511.1——————
Retriever + Reader2026.043235.37.511.1——————
DPR + FiD2020.1230.8346.910.8——————
Reader-to-Retriever DistillationInitialization=BM25, Iteration=02020.1230.333.76.510.4——————
Ours (Oracle Setting)Backbone=QWEN2.5-0.5B, Setting=Oracle access to adapter2026.0530.22————————31.23
Reasoner-onlyBackbone=Qwen2.5-1.5B-Instruct2026.0128.13———————41.5—
Ours (Retrieval)Backbone=QWEN2.5-7B, k=52026.0527.86————————28.64
Ours (Retrieval)Backbone=QWEN2.5-7B, k=32026.0527.68————————28.46
DPRLanguage Model=GPT-4o-mini2026.0426.521.51.314.2——————
RAGBackbone=Qwen2.5-1.5B-Instruct2026.0126.47———————4.5—
DTCRSLanguage Model=GPT-4o-mini2026.042621.11.314.4——————
RAPTORLanguage Model=GPT-4o-mini2026.042521.21.114.1——————
Ours (Retrieval)Backbone=QWEN2.5-0.5B, k=32026.0523.57————————24.54
Ours (Retrieval)Backbone=QWEN2.5-0.5B, k=52026.0523.46————————24.44
Recursively Summarizing Books2026.0421.622.34.210.6——————
RAGBackbone=QWEN2.5-0.5B, k=32026.0520.21————————21.12
RAGBackbone=QWEN2.5-0.5B, k=52026.0515.79————————16.66
BM25 + BERT2026.0415.514.51.45——————
Base ModelBackbone=QWEN2.5-7B2026.0514.4————————14.95
Kočiskỳ et al.2020.121419.12.14.4——————
Base ModelBackbone=QWEN2.5-0.5B2026.0512.26————————13.08
PREMemBackbone=Qwen2.5-1.5B-Instruct2026.0111.64———————10.37—
MeMWalker2024.109.8————11.2353.811.41——
InfiniteICLBackbone=QWEN2.5-0.5B2026.059.35————————10.1
PREMemBackbone=SmolLM3-3B2026.019.11———————11.22—
THEANINEBackbone=SmolLM3-3B2026.017.84———————10.87—
BiDAF2026.046.25.70.33.7——————
TempLoraCDBackbone=QWEN2.5-0.5B2026.056.14————————6.65
Mem0Backbone=Qwen2.5-1.5B-Instruct2026.015.23———————1.12—
THEANINEBackbone=Qwen3-8B2026.015.13———————6.63—
TempLoraBackbone=QWEN2.5-0.5B2026.054.18————————4.67
TempLoraBackbone=QWEN2.5-7B2026.054.11————————4.83
THEANINEBackbone=Qwen2.5-1.5B-Instruct2026.013.84———————2.25—
TempLoraCDBackbone=QWEN2.5-7B2026.053.78————————4.09
InfiniteICLBackbone=QWEN2.5-7B2026.053.77————————4.24
Mem0Backbone=Qwen3-8B2026.013.12———————0.5—
Mem0Backbone=SmolLM3-3B2026.013.04———————0.25—
PREMemBackbone=Qwen3-8B2026.013———————2.57—
AutoComCompression Ratio=4x2026.03—————14.41————
BART-LS2022.09————26.2—————
BM25Executive Model=Qwen2.5-32B-Instruct2026.05————————10.24—
BM25Executive Model=Gemini-3-Flash2026.05————————14.33—
CartridgesExecutive Model=Qwen2.5-32B-Instruct2026.05————————3.75—
HippoRAG2Executive Model=Qwen2.5-32B-Instruct2026.05————————21.39—
HippoRAG2Executive Model=Gemini-3-Flash2026.05————————23.21—
ICAECompression Ratio=4x2026.03—————15.76————
LLOCOCompression Ratio=4x2026.03—————15.88————
LongT5-3BModel size=3B2022.09————29.3—————
LongT5-baseModel size=base2022.09————23—————
LongT5-largeModel size=large2022.09————27.2—————