← All posts


2 posts
One ModelDeployment, two serving stacks
A Modelplane cluster can now serve models with NVIDIA's Dynamo components instead of the stack Modelplane composes itself. It's a per-cluster platform choice, and the API an ML team writes doesn't move.

Why Day 0 for Nemotron 3.5 Lightning wasn't a scramble
NVIDIA released Nemotron-3.5-Lightning this morning. It was running on Modelplane by the afternoon, without a line of new Modelplane code, because day-zero model support is built into the design, not a scramble by the team.