List your company

Multi-hop Question Answering on MuSiQue 300-question hop-stratified (held-out)

75.7Best@3Max-at-K
29.21241.28153.3565.419May 21, 2026

Evaluation results

MethodLinks
Max-at-KBase Model=Qwen3-1.7B, Prompt Strategy=single-answer prompt2026.0575.776.878.380.257.317.5
MaxRLBase Model=Qwen3-1.7B, Prompt Strategy=single-answer prompt2026.0575.175.776.577.655.26.8
VPOBase Model=Qwen3-1.7B, Prompt Strategy=multi-answer prompt2026.0574.27880.983.267.858.7
Random-wBase Model=Qwen3-1.7B, Prompt Strategy=single-answer prompt2026.0571.271.672.172.844.56.1
GRPOBase Model=Qwen3-1.7B, Prompt Strategy=single-answer prompt2026.0571.171.672.172.844.75.4
Multi-RLVRBase Model=Qwen3-1.7B, Prompt Strategy=multi-answer prompt2026.0559.961.662.763.349.881.4
Qwen3-1.7BPrompt Strategy=multi-answer prompt2026.0540.243.947.550.637.985.5
Qwen3-1.7BPrompt Strategy=single-answer prompt2026.053133.937.342.316.163.4