CLI overview¶

The modeldiffr command.

usage: modeldiffr [-h] [--version] {audit,suites,tasks,power,refusal-eval,verify,fingerprint} ...

Audit what changed between a base LLM and a derivative model.

Command	Purpose
`audit`	Compare a base model with a derivative model
`verify`	Check that a report was not edited and matches the real model files
`fingerprint`	Print content hashes for one model
`suites`	List available suites
`tasks`	List tasks with their pinned datasets
`power`	Smallest change a task size can detect
`refusal-eval`	Measure a refusal detector on labeled responses

Global options¶

`--version`¶

flag

Print modeldiffr 0.3.0 and exit.

`-h, --help`¶

flag

Show help for the command or subcommand.

Entry points¶

All three are equivalent:

modeldiffr audit base derivative
python -m modeldiffr audit base derivative

from modeldiffr.cli import main

exit_code = main(["audit", "base", "derivative"])

main returns the exit code instead of exiting, which makes it easy to call from tests. See exit codes.

Output streams¶

Stream	Content
stderr	Progress lines prefixed with `[modeldiffr]`, and errors
stdout	Metric summary table, notes, and report paths