List your company

Search-Augmented Multi-Turn QA on HotpotQA (EM/F1)

40.2Exact Match (EM)IGPO (gold-sup.)
31.46433.7323638.268May 6, 2026

Evaluation results

MethodLinks
IGPO (gold-sup.)Backbone=Qwen3-8B2026.0540.249.9
EMPOBackbone=Qwen3-8B2026.0539.244.5
GRPO (gold-sup.)Backbone=Qwen3-8B2026.0539.148.4
TTRLBackbone=Qwen3-8B2026.0538.543.1
IGPO (gold-sup.)Backbone=Qwen3-4B2026.0535.945.1
SIOPBackbone=Qwen3-8B2026.0534.845.3
SIOPBackbone=Qwen3-4B2026.0534.444.3
GRPO (gold-sup.)Backbone=Qwen3-4B2026.053344.1
EMPOBackbone=Qwen3-4B2026.053341.4
TTRLBackbone=Qwen3-4B2026.0531.840.1