List your company

Accuracy on Multi-Hop Question Answering on Bamboogle (test)

70.1AccuracyGRPO w/ PaW
8.7424.6740.656.53Jun 1, 2026Jun 6, 2026Jun 12, 2026Jun 18, 2026Jun 24, 2026Jun 30, 2026Jul 6, 2026

Evaluation results

MethodLinks
GRPO w/ PaWType=RL Training, Backbone=Qwen2.5-7B-Instruct2026.0670.1
GRPOType=RL Training, Backbone=Qwen2.5-7B-Instruct2026.0669.6
GIGPO w/ PaWType=RL Training, Backbone=Qwen2.5-7B-Instruct2026.0669.5
STAPOBackbone=Qwen2.5-7B-Instruct, Type=Process RL2026.0769.4
GIGPOType=RL Training, Backbone=Qwen2.5-7B-Instruct2026.0668.9
GiGPOBackbone=Qwen2.5-7B-Instruct, Type=Process RL2026.0768.9
GRPO w/ PaWType=RL Training, Backbone=Qwen2.5-3B-Instruct2026.0665.2
GIGPO w/ PaWType=RL Training, Backbone=Qwen2.5-3B-Instruct2026.0664.9
STAPOBackbone=Qwen2.5-3B-Instruct, Type=Process RL2026.0764.9
GRPOType=RL Training, Backbone=Qwen2.5-3B-Instruct2026.0664.1
GiGPOBackbone=Qwen2.5-3B-Instruct, Type=Process RL2026.0764.1
GIGPOType=RL Training, Backbone=Qwen2.5-3B-Instruct2026.0660
StepSearchBackbone=Qwen2.5-7B-Instruct, Type=Process RL2026.0740
Search-R1Type=RL Training, Backbone=Qwen2.5-7B-Instruct2026.0636.8
Search-R1Backbone=Qwen2.5-7B-Instruct, Type=Outcome RL2026.0736.8
StepSearchBackbone=Qwen2.5-3B-Instruct, Type=Process RL2026.0734.4
R1-InstructBackbone=Qwen2.5-7B-Instruct, Type=Outcome RL2026.0729.3
ZeroSearchType=RL Training, Backbone=Qwen2.5-7B-Instruct2026.0627.8
ZeroSearchBackbone=Qwen2.5-7B-Instruct, Type=Outcome RL2026.0727.8
Search-R1Type=RL Training, Backbone=Qwen2.5-3B-Instruct2026.0626.4
Search-R1Backbone=Qwen2.5-3B-Instruct, Type=Outcome RL2026.0726.4
R1-InstructBackbone=Qwen2.5-3B-Instruct, Type=Outcome RL2026.0719.2
ZeroSearchType=RL Training, Backbone=Qwen2.5-3B-Instruct2026.0611.1
ZeroSearchBackbone=Qwen2.5-3B-Instruct, Type=Outcome RL2026.0711.1