EasyDeepLearn

How do you control ML serving cost?

medium

Answer

  • (1) Right-size infrastructure: autoscale on utilization.
  • (2) Batch requests where latency allows.
  • (3) Distill to smaller model.
  • (4) Cache popular predictions.
  • (5) Two-stage: cheap filter + expensive rerank only on top-K.
  • (6) Spot / preemptible instances for batch.
  • (7) Compress model (quant / prune) to fit on cheaper HW.
  • (8) Move batch scoring off critical path.
  • (9) Serverless (Lambda / Cloud Run) for spiky traffic.
  • (10) Multi-tenant serving: co-locate models on same GPU.
  • Monitor $/prediction.
Check yourself — multiple choice
  • Random
  • Autoscale + batch + distill + cache + two-stage + spot instances + compress + move batch off critical + serverless spiky + multi-tenant GPU; monitor $/prediction
  • Just scale up
  • Not real

Cost: autoscale + batch + distill + cache + two-stage + spot + compress.

#deployment#infrastructure

Practise MLOps & Data Quality

215 interview questions in this topic.

Related questions