List your company

Question Answering on HotpotQA (test) (F1, EM, CR, TP)

65.49Exact Match (EM)ATACompressor
2.372418.758735.14551.5313Dec 8, 2025Dec 17, 2025Dec 27, 2025Jan 5, 2026Jan 15, 2026Jan 24, 2026Feb 3, 2026

Evaluation results

MethodLinks
ATACompressorBackbone=Qwen-2-7B2026.0265.4980.2323.813.63
ATACompressorBackbone=LLaMA-2-7B2026.0262.6578.4424.153.86
QGCBackbone=Qwen-2-7B2026.0251.572.3613.981.83
QGCBackbone=LLaMA-2-7B2026.0245.1268.1814.581.95
500CompressorBackbone=Qwen-2-7B2026.0242.1567.423.163.51
LongLLMLinguaBackbone=Qwen-2-7B2026.0240.0564.64.671.28
Original-ContextBackbone=Qwen-2-7B2026.0239.7359.8811.24
ICAEBackbone=Qwen-2-7B2026.0239.6865.3523.163.6
500CompressorBackbone=LLaMA-2-7B2026.0239.664.323.163.69
ICAEBackbone=LLaMA-2-7B2026.0237.762.0823.163.76
LongLLMLinguaBackbone=LLaMA-2-7B2026.0237.362.84.221.31
Selective-ContextBackbone=Qwen-2-7B2026.0237.0853.73.731.24
Original-ContextBackbone=LLaMA-2-7B2026.0236.253.7111.21
Selective-ContextBackbone=LLaMA-2-7B2026.0236.0551.653.521.21
AutoCompressorBackbone=LLaMA-2-7B2026.023259.6412.232.95
PPO + PRS_shortBackbone=Qwen2.5-3B-Instruct2025.1232———
GRPO + PRS_shortBackbone=Qwen2.5-3B-Instruct2025.1231———
VSPO + PRS_shortBackbone=Qwen2.5-3B-Instruct2025.1231———
SFTBackbone=Qwen2.5-3B-Instruct2025.1218.6———
Closed-bookBackbone=Qwen-2-7B2026.0210.330.58—5.64
UntrainedBackbone=Qwen2.5-3B-Instruct2025.126.22———
Closed-bookBackbone=LLaMA-2-7B2026.024.822.8—6.37