7c91a20The short answer
Git tracks the code.
What tracks the build?
Git is essential. Keep it.
A GGUF build also depends on files and runtime details that may never enter the repository. ModelBake records those details and compares them with the build you accepted.
01 One common failure
Same commit.
Different GGUF.
sha256 changedsha256 changedreview again02 What each one records
Two jobs.
Use both.
BUILD DETAILGITMODELBAKE
Repository commitRecordedReferenced
Checkpoint bytesOnly if committedHashed
Converter bytes and revisionOnly if committedRecorded
Exact build commandNot the executed commandRecorded
Python environmentNot the installed stateRecorded
GGUF output bytesOnly if committedHashed
Accepted predecessorManual conventionBound to the record
Build-to-build diffRepository filesBuild inputs and outputs
03 Where it fits
Put the record
back in Git.
ModelBake produces a readable manifest. Keep that manifest beside your code, attach it to a release, or check it in CI.
$ modelbake compare approved/manifest.json candidate/manifest.json
recipe: changed
source digest: same
converter digest: changed
q4 output digest: changed
No promotion decision inferred.Current scope: local Llama safetensors to GGUF builds with llama.cpp on macOS and Linux. It records evidence. It does not certify model quality, safety, licensing, or production readiness.
04 See the whole loop
Try it without
a model.
The tour creates two tiny sample builds and shows the exact diff. Nothing is uploaded.
01 python -m pip install https://modelbake.dev/downloads/modelbake_ai-0.1.0-py3-none-any.whl
02 modelbake tour