liveBook · Manning

Versioning

As explained in LLMs in Production

Versioning is the practice of recording and managing distinct revisions of software, data, models, or configuration.

Versioning keeps model revisions available for replacement and production experiments such as A/B or shadow tests. In a registry, link versions to relevant metadata and control who can promote or demote them. section 3.4.3

Versioning also supports reproducible software setup: because llama.cpp lacks its own versioning, checking out a tested Git commit fixes the repository state. section 11.1.3

$ git checkout 306d34be7ad19e768975409fc80791a274ea0230

Pin the llama.cpp checkout to this tested commit rather than relying on the repository’s changing current state. section 11.1.3

How LLMs in Production covers it

How deep it goes
Connects model revisions to replacement and production tests, but gives no detailed model-versioning workflow.
Examples
Uses production model tests and a pinned llama.cpp Git commit as examples of versioning.
What you can do after
Operate model deployments by retaining revisions for replacement and experiments, and reproduce a llama.cpp setup by checking out its tested commit.
How it connects
Links model versioning with model registries, experiment tracking, A/B testing, and shadow testing.
Cover of LLMs in Production
liveBookmanning.com
Read LLMs in Production on liveBook Get the book at manning.com

Also explained in

Compare how 25 books explain versioning →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of LLMs in Production. The section links point to the parts of the book it draws on.