Having successfully run a self-hosted NIM on my device, I’m now exploring how to package my own model as a NIM.
So far, all NIM models appear to be official, and I haven’t found any guidelines for creating a NIM-compliant container image. I’d like to confirm if this is possible.
In my opinion, it is technically feasible, based on general image creation methods. However, it needs to be confirmed whether NVIDIA allows users to do this. Additionally, the operational process may require users to figure it out themselves.
Hi @changtimwu, we do not support or guide users in creating their own NIM.
Thank you for your responses. Since NVIDIA actively promotes NIM, many assumed it to be an open AI application standard for developers. NVIDIA even integrates NIM into their RTX AI PC ecosystem: NVIDIA Launches AI Foundation Models for RTX AI PCs | NVIDIA Newsroom.
Some NIMs also come from third-party vendors: breeze-7b-instruct Model by MediaTek | NVIDIA NIM, suggesting there is a way for ISVs to publish models on the NIM store.
I believe NVIDIA should clarify their policy and future plans for NIM. Many excellent models are still missing from the platform, which should be seen as an opportunity.