Two Qwen3 models on one DGX Spark: the residency math

Posting my takeaways from AI Engineers Code (NYC) Conference