Generate workload for multiturn conversation

[hasanar1f]

Thanks for sharing this amazing work! I was wondering how can we generate workload for multi-turn conversation? Ideally I would like to regenerate the scenarios explained in 5.2 Arrival Pattern and Multi-Turn Conversations of the paper.

Thanks a lot!

[lixueclaire]

Thanks for sharing this amazing work! I was wondering how can we generate workload for multi-turn conversation? Ideally I would like to regenerate the scenarios explained in 5.2 Arrival Pattern and Multi-Turn Conversations of the paper.

Thanks a lot!

Hi @hasanar1f,

Thank you for your interest in our work and for highlighting this important point! We truly appreciate your feedback. While we did consider the characteristics of multi-turn conversations when generating the workload, I apologize for not including specific details related to multi-turn scenarios in the results. We will definitely take your suggestion into account as we refine our approach moving forward.

[hasanar1f]

Looking forward to that. I would appreciate if we can get more more deeper insights in the future. For example, for multi-turn chatting, what factors impact the inter-turn times of users? Like - longer prompt/response may need more time to write/read. Or they may come from specific type of devices (e.g., desktop or mobile app). And what not! Any plan to share the traces?

Thanks again for the effort.

[lixueclaire]

Looking forward to that. I would appreciate if we can get more more deeper insights in the future. For example, for multi-turn chatting, what factors impact the inter-turn times of users? Like - longer prompt/response may need more time to write/read. Or they may come from specific type of devices (e.g., desktop or mobile app). And what not! Any plan to share the traces?

Thanks again for the effort.

We appreciate your suggestion, and while we are committed to delving deeper into these aspects in future work, we must also consider privacy implications. As such, we are unable to fully share the raw request data due to confidentiality and privacy restrictions. Nonetheless, we will aim to provide aggregated insights and high-level analyses.