List your company

Search-augmented multi-turn Question Answering on Bamboogle (EM, F1)

48.8Exact Match (EM)GRPO (gold-sup.)
39.64842.02444.446.776May 6, 2026

Evaluation results

MethodLinks
GRPO (gold-sup.)Backbone=Qwen3-8B2026.0548.859.9
SIOPBackbone=Qwen3-8B2026.0548.861.2
EMPOBackbone=Qwen3-8B2026.0547.258
GRPO (gold-sup.)Backbone=Qwen3-4B2026.0546.453.4
IGPO (gold-sup.)Backbone=Qwen3-4B2026.0546.455
EMPOBackbone=Qwen3-4B2026.0545.651
IGPO (gold-sup.)Backbone=Qwen3-8B2026.0545.658.3
TTRLBackbone=Qwen3-8B2026.054453.9
SIOPBackbone=Qwen3-4B2026.0540.852.8
TTRLBackbone=Qwen3-4B2026.054052