List your company

Sequential Decision Making on ALFWorld (val_seen)

15,527Total Wall-Clock Time (s)Selective-rollout gate
15,452.3615,956.1816,46016,963.82May 7, 2026

Evaluation results

MethodLinks
Selective-rollout gateBackbone=Qwen2.5-7B, LoRA=true, Gating Mechanism=Enabled, K=10, dL=0.12, G=8, Training Iterations=60, Prompts per iteration=10, Decoding Strategy=greedy2026.0515,5271060.14539.544
baselineBackbone=Qwen2.5-7B, LoRA=true, Gating Mechanism=Disabled, K=10, dL=0.12, G=8, Training Iterations=60, Prompts per iteration=10, Decoding Strategy=greedy2026.0517,39300.12539.541.5