List your company

Multi-Hop Question Answering on HotpotQA (F1-score, Collapse Point)

63.09F1 ScoreCalibAdv
47.479651.532355.58559.6377Apr 20, 2026

Evaluation results

MethodLinks
CalibAdvBackbone=Qwen2.5-7B-Base2026.0463.09—
LLDBackbone=Qwen2.5-7B-Base2026.0460.63—
CalibAdvBackbone=Qwen2.5-3B-Base2026.0459.19—
CalibAdvBackbone=Llama3.2-3B-Instruct2026.0456.97—
SimpleTIRBackbone=Qwen2.5-3B-Base2026.0456.54310
Search-R1Backbone=Qwen2.5-3B-Base2026.0456.52292
LLDBackbone=Llama3.2-3B-Instruct2026.0455.59—
SimpleTIRBackbone=Qwen2.5-7B-Base2026.0454.4765
MT-GRPOBackbone=Qwen2.5-7B-Base2026.0454—
GiGPOBackbone=Qwen2.5-7B-Base2026.0450.64—
Search-R1Backbone=Qwen2.5-7B-Base2026.0450.1658
SimpleTIRBackbone=Llama3.2-3B-Instruct2026.0449.654
StepSearchBackbone=Qwen2.5-7B-Base2026.0448.84—
Search-R1Backbone=Llama3.2-3B-Instruct2026.0448.0840