Deploy a Model by Dropping in a Folder (Model Repository)

Why this matters

The thing that turns model serving from artisanal to industrial is a standardized layout: a model repository where every model is a directory with a known structure — config, versions, weights — that the server discovers and loads automatically. Once deployment is 'drop a correctly shaped folder into the repository,' you get reproducibility, code review, CI, and rollback for free, because deploying a model becomes a file operation under version control. This is the same shift that container images brought to apps. Understanding the model-repository convention is the foundation for everything in Triton, and it's why this course keeps coming back to config.pbtxt and versioned directories.

Demo

The demo lays out a Triton-style model repository as a directory tree, showing that each model is self-describing: a config plus numbered version folders holding the actual weights.

Try it yourself

1. Add a third model directory (a safety classifier) with its own config.pbtxt and a 1/ version folder, matching the convention exactly.
2. Give the embeddings model a version 3 folder and reason about how the server decides which version is 'latest'.
3. Identify which single file in each model directory is the contract the server reads first, and why it must exist before any weights are loaded.
4. Explain how putting this tree under git turns a model deploy into a reviewable, revertible pull request.

References

• Triton — model repository layout
• Triton — model configuration (config.pbtxt)

Chat about this lesson