LLMOps
The operational practices specific to deploying and managing large language models — prompt versioning, output validation, cost tracking.
Think of it like
MLOps's younger, more chaotic sibling focused on inference rather than training.
Example
Versioning prompts alongside model versions, A/B testing different system prompts, monitoring token costs and latency per model.
How it actually works
LLMOps is a newer discipline because LLM inference is different from training: you're not retraining constantly, you're prompt-engineering and model-picking. Tracking what prompt produced what output, versioning system prompts like code, and catching quality regressions early are the key jobs.
For product teams
LLMOps keeps LLM applications reliable and cost-effective as you iterate on prompts.
For engineers
Treat prompts as versioned code; log outputs for offline evaluation; track token usage per endpoint.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome