Dear Orchestrators,. I hope you all had a fantastic weekend.

16 Sep 2024, 09:40

Dear Orchestrators, I hope you all had a fantastic weekend! 🌴 Over the past few days, we conducted some production load tests with one of our startups to help them onboard to the network (see the [dune dashboard](https://dune.com/rickstaa/livepeer-ai). While we identified a few areas for improvement, we successfully achieved a throughput of approximately 110 requests per minute ⚡. I’ve noticed that several of you have already switched over to SAM2 due to the higher fees 💪. Please note that since we are providing close support to this startup, their requests are being routed through a dedicated AI SPE gateway, so they won’t appear on the public cloud dashboard. I’m looking into creating a secondary dashboard for scenarios like this. Longer-term, I believe building a data aggregator where orchestrators can contribute their data would provide a more comprehensive view of overall network activity. For the next 24 hours, the startup will continue sending requests through SAM2 in the US region, providing opportunities for you to capture additional fees from that pipeline. I’ll be adjusting the base capacity to ensure that others can participate during this transition. The SAM2 model only requires 4GB of memory, so it can be stacked with other models. You can follow [this guide on stacking models](https://livepeer.notion.site/Model-Stacking-Guide-1030a348568780c49294cd71617108fa?pvs=4) on a GPU. Please ensure your setup has enough VRAM available as a buffer and follow all validation steps carefully to avoid performance issues. Since this job is critical, we will need to blacklist orchestrators whose setups are incorrect to ensure network reliability. Thank you for your understanding, and happy computing! 🚀💰