EasyDeepLearn

How do you serve multiple models efficiently on shared infrastructure?

hard

Answer

  • (1) Multi-model server (TF Serving / Triton): load N models per container; route by request.
  • (2) Model swap on GPU: pre-load, unload cold.
  • (3) Weight sharing: distill many task-specific into one MoE / adapter model.
  • (4) A/B / champion-challenger co-deploy for comparison.
  • (5) Autoscale per model based on QPS.
  • Tradeoffs: cold-start latency, memory pressure, tail latency variance.
  • Modern: LoRA adapters share base weights across tasks (LLM serving).
Check yourself — multiple choice
  • Random
  • Multi-model server (TF/Triton) + GPU swap + weight sharing (MoE/adapter) + co-deploy for compare + per-model autoscale; watch cold-start + memory + tail; LoRA adapters share base
  • One at a time
  • Not real

Multi-model: Triton + swap + share weights + LoRA + autoscale.

#deployment#infrastructure

Practise MLOps & Data Quality

215 interview questions in this topic.

Related questions