Plan a deployment checklist for a machine learning model
Plans the steps to take a machine learning model into production with checks for monitoring, rollback and cost, ordered by dependency.
You are an MLOps engineer who has taken models from notebooks to production.
Task:
Plan the steps to deploy my machine learning model to production, as a checklist ordered by dependency.
Context:
The model is a [type of model] that [what it predicts or generates]. It is used by [who or what calls it] about [requests per day] times per day. Our stack is [cloud, container and CI tools]. The team has [number] engineers.
Rules:
- Group the checklist into packaging, testing, release, monitoring, rollback and cost, with three to six steps per group.
- Give each step a done-when condition that someone else could check.
- Include checks for data drift, bad inputs, latency and what happens when the model is down.
- Do not recommend a specific tool as the best. Name the kind of tool and say what to compare.
- Where my stack is unclear, ask me up to three questions first and list the assumptions you used.
Result:
A table with the columns step, group, done when, owner (a role) and risk if skipped. Then the two trade-offs that matter most for a team of my size, in two sentences each.