Hello community,
We have deployed the NVIDIA BioNeMo container on our public cloud infrastructure (not on DGX Cloud) and we are able to open the Jupyter Notebook in a browser.
Q1: Is there a difference between deploying BioNeMo on a cloud compared to using NVIDIA DGX Cloud, specifically in terms of access to different Graphical User Interfaces? This question arose after watching this video on BioNeMo that show some GUIs which are different from a Jupyter Notebook:
In this video, the relevant section starts at 8 minutes and 40 seconds.:
Q2: Are these GUIs available for users who deploy BioNeMo on their infrastructure or other public clouds, or are they only accessible through NVIDIA DGX or other NVIDIA enterprise solutions?
Dear @KindnessCUDA
Thank you for your questions. The video referenced here is from GTC 2023; please refer to GTC 2024 announcements for the latest updates.
NVIDIA BioNeMo has two aspects:
-
BioNeMo Framework – for pre-training, fine-tuning, and inferencing AI models for Biology, Chemistry, and Drug discovery. The concept here is to use BioNeMo Framework as an AI Foundry or AI Factory for drug discovery, and the Framework container provides access to custom tools, pre-trained model checkpoints, scaling, and inferencing recipes to run relevant tasks for model development and adaptations. You can learn more about this here: Introduction — NVIDIA BioNeMo Framework
-
BioNeMo Cloud API—for using cutting-edge AI models for inferencing. It provides access to more than a dozen models for supporting various steps of drug discovery, from protein language modeling to small molecule generation, protein structure prediction, docking, etc. Please visit https://bionemo.ngc.nvidia.com/ and https://ai.nvidia.com/ to learn more.
Coming to your questions:
- Once the BioNeMo framework container is running on a node, it can be accessed via a JupyterLab session. This is one way to use the container for model pre-training, fine-tuning, etc.
- On the other hand, BioNeMo Cloud API service, which differs from BioNeMo Framework, provides either GUI-based or programmatic API-call-based access. It provides access to the models running and hosted by NVIDIA, and users can use them for inferencing. This web GUI is not linked to the BioNeMo Framework.
Also, please note that NVIDIA NIMs are announced during the GTC 2024 conference. To learn more, please see NVIDIA GTC 2024 Healthcare Sessions.
1 Like
Thank you @nilkanthp!
Q1: In addition to the ‘BioNeMo Framework’ and ‘BioNeMo Cloud API,’ I see another option named ‘BioNeMo microservices’ which can be self-hosted. It is not clear to me what the differences are between ‘NVIDIA Microservices’ and the other two options, or how they are related. Could you please kindly clarify?
Q2: Please correct me if I am wrong: inferencing is possible with both the ‘BioNeMo Framework’ and the ‘BioNeMo Cloud API,’ but the BioNeMo Cloud API offers a more user-friendly GUI, correct? Is there any difference in the output of inferencing for the same model between these two options? For example, might the Cloud API use updated versions of certain components?
Q3: While exploring the page Try NVIDIA NIM APIs, I clicked on the profile picture and noticed the ‘1000 Credits left - Request More’ option. Upon clicking ‘Request More,’ the following statement appeared: “Download and use the API on any Kubernetes cluster for 90 days.”
As it mentions the use of the API on any Kubernetes cluster, does this mean we can self-host and deploy it on our own infrastructure? If so, will we have access to the same GUIs or APIs as those available on DGX Cloud, and will they offer the same capabilities and user-friendliness that the BioNeMo Framework lacks?
To your Q1: NVIDIA NIMs is a new offering that packages the solution in a highly optimized way for inferencing anywhere (i.e., on-prem, cloud, NVIDIA DGX cloud). For more, please check out Try NVIDIA NIM APIs and GTC announcements.
To your Q2: The BioNeMo Cloud-API offering provides inferencing solutions for a set of Gen-AI models hosted on NVIDIA Cloud. Users can use web-UI or API calls to perform inference. BioNeMo Framework, on the other hand, provides a way of building, fine-tuning, and inferencing Gen-AI models for drug discovery. There is an overlap between the models hosted on BioNeMo Cloud API and those offered via BioNeMo Framework.
Regarding Q3: If you’d like to host the NIMs on your local on-prem or Cloud compute nodes, you may apply for NVAIE evaluation (valid for 90 days). Those models are the same as the ones hosted by NVIDIA; however, they will not have GUIs. Once they are up and running on a GPU node, you can use them for inferencing via API calls. To use beyond the evaluation period of 90 days, and to get technical support from NVIDIA, please contact the NVIDIA Sales representative.
1 Like
