List your company
ResearchFormerly WizWand

Multi-turn decision making on ALFWorld (val-unseen)

94.1Success RateLoRA baseline
14.0234.8155.676.39Mar 18, 2026
Updated 7mo ago

Evaluation results

MethodLinks
LoRA baselineBackbone=Qwen 2.5-7B, Paradigm=Supervised Fine-tuned2026.0394.1
Rule-based ExpertBackbone=Qwen 2.5-7B, Paradigm=Supervised Fine-tuned2026.0389.6
ITPRBackbone=Qwen 2.5-7B, Paradigm=Supervised Fine-tuned2026.0385.1
SANDBackbone=Qwen 2.5-7B, Paradigm=Supervised Fine-tuned2026.0385
ExpRAG baselineBackbone=Qwen 2.5-7B, Paradigm=Training-Free Memory-Augmented2026.0383.6
SFT + ReActBackbone=Qwen 2.5-7B, Paradigm=Supervised Fine-tuned2026.0380.7
ETO + ReActBackbone=Qwen 2.5-7B, Paradigm=Supervised Fine-tuned2026.0379.9
Self-ReflectionBackbone=Qwen 2.5-7B, Paradigm=Supervised Fine-tuned2026.0371.1
IWMBackbone=Qwen 2.5-7B, Paradigm=Supervised Fine-tuned2026.0370.3
NAT + ReActBackbone=Qwen 2.5-7B, Paradigm=Supervised Fine-tuned2026.0366.4
ReflexionBackbone=Qwen 2.5-7B, Paradigm=Training-Free Memory-Augmented2026.0342.7
Memory BankBackbone=Qwen 2.5-7B, Paradigm=Training-Free Memory-Augmented2026.0340.3
AgeMem-noRLBackbone=Qwen 2.5-7B, Paradigm=Training-Free Memory-Augmented2026.0337.9
ITPIBackbone=Qwen 2.5-7B, Paradigm=Prompting2026.0335.7
A-MEMBackbone=Qwen 2.5-7B, Paradigm=Training-Free Memory-Augmented2026.0334.7
Mem0Backbone=Qwen 2.5-7B, Paradigm=Training-Free Memory-Augmented2026.0333.6
Zero-shotBackbone=Qwen 2.5-7B, Paradigm=Prompting2026.0329.9
ReActBackbone=Qwen 2.5-7B, Paradigm=Prompting2026.0317.1