List your company

Search-augmented Multi-turn Question Answering on MuSiQue (EM, F1)

17.8Exact Match (EM)IGPO (gold-sup.)
8.64811.02413.415.776May 6, 2026

Evaluation results

MethodLinks
IGPO (gold-sup.)Backbone=Qwen3-8B2026.0517.827.6
GRPO (gold-sup.)Backbone=Qwen3-8B2026.0514.122.4
TTRLBackbone=Qwen3-8B2026.0513.319.2
GRPO (gold-sup.)Backbone=Qwen3-4B2026.0512.118.3
EMPOBackbone=Qwen3-8B2026.051219.3
IGPO (gold-sup.)Backbone=Qwen3-4B2026.0511.318.4
SIOPBackbone=Qwen3-8B2026.0510.720.3
SIOPBackbone=Qwen3-4B2026.051017.9
TTRLBackbone=Qwen3-4B2026.05915.7
EMPOBackbone=Qwen3-4B2026.05915.9