Agent workload

[lhr-30]

In the paper, you mentioned selecting AutoGen to execute the agent workload; however, the specific agent(s) utilized were not clearly identified. Could you kindly provide some examples?

[Txxx926]

For a common naive RAG, we utilize the RetrieveUserProxyAgent (for retrieval tasks) and the AssistantAgent (for LLM-related tasks). Both interact with locally deployed models. For certain undefined agents (e.g. reranking), we customize them as instructed in AutoGen's doc.

[lhr-30]

But can you provide the examples for the Search engine-empowered generation? It would be helpful if you can provide that :)

[Txxx926]

Sure, I would provide an example of search engine-empowered generation with Ayo. However, I currently do not plan to include the related implementation for AutoGen in the current repository. That said, integrating it with AutoGen should be straightforward, as the workflow would be similar to Ayo's unoptimized version.