Skip to content

[Feature Request]: Allow model aggregation in AutoGen #3040

Closed as not planned
@qingyun-wu

Description

@qingyun-wu

Is your feature request related to a problem? Please describe.

An Agent in AutoGen can take an llm_config with multiple models. Currently AutoGen tries the models one by one and uses the first one that works from the provided model list. It would be great if AutoGen also supports model aggregation.

Describe the solution you'd like

Mixture of Agents (MoA) is an approach that leverages the collective strengths of multiple LLMs to enhance performance, achieving state-of-the-art results. This approach could be realized as one way of model aggregation with AutoGen.

EcoAssistant could be another mode of model aggregration.

Additional context

@JieyuZ2 what do you think? We are preparing a DeepLearning.AI course and are considering demonstrating EcoAssisant in one of the lessons. We need to figure out an elegant way to support EcoAssistant.

Metadata

Metadata

Assignees

No one assigned

    Labels

    0.2Issues which are related to the pre 0.4 codebaseneeds-triage

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions