List your company

Constrained Path Planning on Textualized Gridworld Short targets (test)

56Success RateSAE-opt-anch
10.34422.19734.0545.903May 21, 2026

Evaluation results

MethodLinks
SAE-opt-anchModel=Qwen3-4B-Instruct2026.055623.4
SAE-optModel=Qwen3-4B-Instruct2026.0555.521
SAE-SSVModel=Qwen3-4B-Instruct2026.0554.824.7
Dense-optModel=Qwen3-4B-Instruct2026.055426.7
CAAModel=Qwen3-4B-Instruct2026.0553.938.8
SAE-optModel=LLaMA3.1-8B-Instruct2026.0540.443
Dense-optModel=LLaMA3.1-8B-Instruct2026.0539.948
CAAModel=LLaMA3.1-8B-Instruct2026.0539.838.8
SAE-SSVModel=LLaMA3.1-8B-Instruct2026.0539.546.5
SAE-opt-anchModel=LLaMA3.1-8B-Instruct2026.0539.547.7
ITIModel=LLaMA3.1-8B-Instruct2026.0539.267.5
SAE-optModel=Phi-3-Mini-4K-Instruct2026.0538.954.2
ITIModel=Phi-3-Mini-4K-Instruct2026.0538.839.9
CAAModel=Phi-3-Mini-4K-Instruct2026.0538.636.4
Dense-optModel=Phi-3-Mini-4K-Instruct2026.053854.7
SAE-opt-anchModel=Phi-3-Mini-4K-Instruct2026.0537.456.1
SAE-SSVModel=Phi-3-Mini-4K-Instruct2026.0535.855
ITIModel=Qwen3-4B-Instruct2026.0532.957.8
ICLModel=Qwen3-4B-Instruct2026.0529.357.1
ICLModel=LLaMA3.1-8B-Instruct2026.0526.952.1
ICLModel=Phi-3-Mini-4K-Instruct2026.0512.181.3