OpenSearch 2.6 will include the first release of the admin UI for the ml-commons plugin. The dashboard plugin will provide admins with a Machine Learning (ML) Model Health dashboard that will give them visibility into where ML models are running in an OpenSearch cluster and whether or not they are responsive.
We have plans to provide more UI support for ml-commons through the ml-commons-dashboards plugin in future releases. This includes a UI for the model management functionality within the ML Model Serving Framework to help MLOps engineers and other ML solution builders to manage and operationalize their models on OpenSearch.
Please share your initial thoughts on our ML Model Health dashboard and help us create a great user experience for managing and integrating ML workloads with OpenSearch!
Using Ubuntu 22.0.4 Hyper-V virtual machine /w 4CPU, 4GB RAM, and 300GB Drive
Running OpenSearch 2.6.x with ML module all-MiniLM-L6-v2
Just started working with OpenSearch and set up a test production environment recently. The ML install from the documentation worked great and was surprised how easy it was using Dev Tools. Less than 1 minute it was running.
Only issue I’m having is that my VM is running out of memory, I’m only send logs from 4 windows Server 2019 VM’s using Logstash. My heap is 50% (i.e., 2 GB). When I learn more to actually set up a non-testing production environment, I do have a Virtual machine waiting for a full install so I think this small issue will be resolved.
I do have a question on machine learning. Im setting up a lab to use GPU acceleration for Windows Hyper-v but think it using something like GPU-P. What card would be good for just a lab settup?
@Gsmitt, as long as you’re using CUDA compatible GPU it should work. Most of my experience with running ML on GPU workloads are on NVIDIA data center GPUs (on AWS). So, I can’t speak to what works well with Hyper-v GPU-P.
If you’re running a dev environment in the cloud or data center environment, the NVIDIA A10G Tensor Core GPUs are generally great for price performance, and the cloud provider specialized hardware (eg. Amazon Inferentia) will often deliver even better price performance (as long as the chip can support all your model’s operators). If it’s all about maximum performance (cost doesn’t matter), the NVIDIA A100s and the H100s are probably your best options.
