List your company

Click-in-target on WinDOM in-distribution 2,000-example subsample (held-out)

82.1Click-in-target AccuracySFT
69.51672.78376.0579.317Jun 24, 2026

Evaluation results

MethodLinks
SFTBase model=Qwen3.5-2B, Cold-start regime=SFT, RL regime=None2026.0682.1
SFT +RL LateBase model=Qwen3.5-2B, Cold-start regime=SFT, RL regime=Late2026.0680.8
SFD-4B +RL LateBase model=Qwen3.5-2B, Cold-start regime=SFD-4B, RL regime=Late2026.0679.7
SFD-EMA +RL LateBase model=Qwen3.5-2B, Cold-start regime=SFD-EMA, RL regime=Late2026.0679
SFT +RL EarlyBase model=Qwen3.5-2B, Cold-start regime=SFT, RL regime=Early2026.0677.2
SFD-EMABase model=Qwen3.5-2B, Cold-start regime=SFD-EMA, RL regime=None2026.0676.8
SFD-4BBase model=Qwen3.5-2B, Cold-start regime=SFD-4B, RL regime=None2026.0676.5
SFD-4B +RL EarlyBase model=Qwen3.5-2B, Cold-start regime=SFD-4B, RL regime=Early2026.0676.5
SFD-EMA +RL EarlyBase model=Qwen3.5-2B, Cold-start regime=SFD-EMA, RL regime=Early2026.0675.4
Qwen3.5-4B (base)Base model=Qwen3.5-4B, Cold-start regime=None, RL regime=None2026.0671.2
Qwen3.5-2B (base)Base model=Qwen3.5-2B, Cold-start regime=None, RL regime=None2026.0670