List your company

Multi-hop Question Answering on MuSiQue (Recall@K, NDCG@K)

36.8Recall@1F-GRPO
-1.168.69518.5528.405May 13, 2026

Evaluation results

MethodLinks
F-GRPOBackbone=Qwen3-4B2026.0536.871.371.891.678.576.5
GRPOBackbone=Qwen3-4B2026.0536.4636390.871.369.2
F-GRPOBackbone=Qwen3.5-2B2026.0536.272.773.289.979.277.2
GRPOBackbone=Qwen3.5-2B2026.0536.163.165.29071.470.6
MonoT52026.0534.964.574.986.170.774.6
DuoT52026.0534.362.873.584.869.273.3
LiT52026.0533.654.764.482.761.965.6
RankZephyr2026.0533.561.371.582.367.171.1
Decoupled SFT (rank. only)Backbone=Qwen3-4B2026.0532.36368.379.868.669.7
SFTBackbone=Qwen3-4B2026.0527.954.460.969.559.561.5
Decoupled SFT (full)Backbone=Qwen3-4B2026.05243944.260.644.946.5
Zero-shot (single-step)Backbone=Qwen3-4B2026.0523.448.156.358.451.654.7
BM252026.0522.942.953.45746.751.4
Decoupled SFT (rank. only)Backbone=Qwen3.5-2B2026.0517.837.449.843.73944.8
SFTBackbone=Qwen3.5-2B2026.0513.529.23834.431.235
Decoupled SFT (sel. only)Backbone=Qwen3-4B2026.058.616.318.321.71818.5
Zero-shot (single-step)Backbone=Qwen3.5-2B2026.055.210.612.413.211.512.1
Decoupled SFT (full)Backbone=Qwen3.5-2B2026.055.214.824.513.814.719.5
Random2026.0551524.713.314.919.6
Zero-shot (two-step)Backbone=Qwen3-4B2026.054.18.5910.29.19.2
Zero-shot (two-step)Backbone=Qwen3.5-2B2026.052.24.14.85.44.54.7
Decoupled SFT (sel. only)Backbone=Qwen3.5-2B2026.050.30.60.70.70.60.7