List your company

Search-augmented multi-turn Question Answering on Natural Questions (NQ)

28.3Exact Match (EM) AccuracyGRPO (gold-sup.)
24.97225.83626.727.564May 6, 2026

Evaluation results

MethodLinks
GRPO (gold-sup.)Backbone=Qwen3-4B2026.0528.340.8
SIOPBackbone=Qwen3-4B2026.0528.140
IGPO (gold-sup.)Backbone=Qwen3-8B2026.0527.540.7
EMPOBackbone=Qwen3-4B2026.052740.4
GRPO (gold-sup.)Backbone=Qwen3-8B2026.052742.2
SIOPBackbone=Qwen3-8B2026.052740.3
TTRLBackbone=Qwen3-8B2026.0526.640.1
IGPO (gold-sup.)Backbone=Qwen3-4B2026.0526.239.6
TTRLBackbone=Qwen3-4B2026.0526.240.2
EMPOBackbone=Qwen3-8B2026.0525.140.6