List your company
ResearchFormerly WizWand

Sequential Decision Making on HotPotQA

4.7Average Steps per EpisodeTeacher (LLaMA-13B)
4.6285.1145.66.086May 20, 2025
Updated 7mo ago

Evaluation results

MethodLinks
Teacher (LLaMA-13B)Backbone=LLaMA, Model Size=13B2025.054.7
Teacher (OPT-13B)Backbone=OPT, Model Size=13B2025.054.8
Structured Agent DistillationBackbone=LLaMA, Model Size=7B2025.054.8
Structured Agent DistillationBackbone=OPT, Model Size=6.7B2025.054.9
Structured Agent DistillationBackbone=OPT, Model Size=2.7B2025.055
Token-levelBackbone=LLaMA, Model Size=7B2025.055.2
Structured Agent DistillationBackbone=OPT, Model Size=1.3B2025.055.3
Token-levelBackbone=OPT, Model Size=6.7B2025.055.3
SeqKDBackbone=LLaMA, Model Size=7B2025.055.5
Token-levelBackbone=OPT, Model Size=2.7B2025.055.6
SeqKDBackbone=OPT, Model Size=6.7B2025.055.6
KDBackbone=OPT, Model Size=6.7B2025.055.7
KDBackbone=LLaMA, Model Size=7B2025.055.7
SeqKDBackbone=OPT, Model Size=2.7B2025.055.9
Token-levelBackbone=OPT, Model Size=1.3B2025.056
KDBackbone=OPT, Model Size=2.7B2025.056.1
SeqKDBackbone=OPT, Model Size=1.3B2025.056.2
KDBackbone=OPT, Model Size=1.3B2025.056.5