List your company
ResearchFormerly WizWand

Dialogue Agent Interaction on DEMO Non-Collaboration set

8.03Goal Achievement ScoreLlama3.1-8B-Instruct w/ AMPO
7.52047.65277.7857.9173May 4, 2025
Updated 8mo ago

Evaluation results

MethodLinks
Llama3.1-8B-Instruct w/ AMPOInteraction Partner=Claude-3.5-Sonnet, Base Model=Llama3.1-8B-Instruct2025.058.03
Llama3.1-8B-Instruct w/ GRPOInteraction Partner=GPT-4o, Base Model=Llama3.1-8B-Instruct2025.057.96
Llama3.1-8B-Instruct w/ AMPOInteraction Partner=GPT-4o, Base Model=Llama3.1-8B-Instruct2025.057.95
Qwen2.5-7B-Instruct w/ AMPOInteraction Partner=Claude-3.5-Sonnet, Base Model=Qwen2.5-7B-Instruct2025.057.87
Qwen2.5-7B-Instruct w/ GRPOInteraction Partner=Claude-3.5-Sonnet, Base Model=Qwen2.5-7B-Instruct2025.057.84
Llama3.1-8B-Instruct w/ GRPOInteraction Partner=Claude-3.5-Sonnet, Base Model=Llama3.1-8B-Instruct2025.057.81
Qwen2.5-7B-Instruct w/ AMPOInteraction Partner=GPT-4o, Base Model=Qwen2.5-7B-Instruct2025.057.77
Qwen2.5-7B-Instruct w/ GRPOInteraction Partner=GPT-4o, Base Model=Qwen2.5-7B-Instruct2025.057.7
Llama3.1-8B-InstructInteraction Partner=GPT-4o, Base Model=Llama3.1-8B-Instruct2025.057.68
Qwen2.5-7B-InstructInteraction Partner=GPT-4o, Base Model=Qwen2.5-7B-Instruct2025.057.62
Qwen2.5-7B-InstructInteraction Partner=Claude-3.5-Sonnet, Base Model=Qwen2.5-7B-Instruct2025.057.62
Llama3.1-8B-InstructInteraction Partner=Claude-3.5-Sonnet, Base Model=Llama3.1-8B-Instruct2025.057.54