List your company

Multi-hop Question Answering on MuSiQue (dev)

65.073F1 ScoreDense-aware HKVM controller
13.1218826.6091940.096553.58381Jun 5, 2026Jun 7, 2026Jun 9, 2026Jun 12, 2026Jun 14, 2026Jun 16, 2026Jun 19, 2026

Evaluation results

MethodLinks
Dense-aware HKVM controllervariant_type=Dense-aware controller, training_signals=out-of-fold HKVM2026.0665.07364.72264.115
ColBERTv2variant_type=Reference predictor2026.0658.30957.75849.317
ColBERTv22026.0658.30957.75849.317
Controller dense-onlyvariant_type=Source-only controller control, training_signals=out-of-fold HKVM2026.0658.24757.75852.172
Controller HKVM-onlyvariant_type=Source-only controller control, training_signals=out-of-fold HKVM2026.0656.7656.21358.64
Learned HKVM calibrationvariant_type=Reference predictor, training_signals=out-of-fold HKVM2026.0656.75456.21353.303
Contriever2026.0655.56554.90354.199
BM252026.0646.04545.34541.746
Weighted-HG-KV2026.0639.92539.22238.643
Static-HG2026.0637.71436.90536.74
RaRBackbone=Qwen3-4B2026.0636.4626.6—
Weighted-KG-PPR2026.0636.37835.58132.727
KG-PPR2026.0636.33235.5432.478
R1-SearcherBackbone=Qwen3-4B2026.0636.0226—
Co-occurrence-HG2026.0635.63234.71231.734
ARCOBackbone=Qwen3-4B2026.0635.5527.4—
AgentPRMBackbone=Qwen3-4B2026.0634.8125.8—
Search-R1Backbone=Qwen3-4B2026.0633.6624.8—
RLERBackbone=Qwen3-4B2026.0633.3524.4—
CARMOBackbone=Qwen3-4B2026.0632.2622.2—
ARCOBackbone=Llama-3.2-3B2026.0629.721.6—
Base ModelBackbone=Qwen3-4B, Prompting=zero-shot2026.0628.0519.4—
Search-R1Backbone=Llama-3.2-3B2026.0626.9918.8—
R1-SearcherBackbone=Llama-3.2-3B2026.0626.4117.4—
AgentPRMBackbone=Llama-3.2-3B2026.0625.8316.6—
RaRBackbone=Llama-3.2-3B2026.0625.2616—
RLERBackbone=Llama-3.2-3B2026.0623.0214.6—
CARMOBackbone=Llama-3.2-3B2026.0621.4913.4—
Base ModelBackbone=Llama-3.2-3B, Prompting=zero-shot2026.0615.129.4—