Skip to main content

cli

CLI Reference

llem provides four commands and a version flag.

llem run [CONFIG] [OPTIONS] # run an experiment or study
llem config [OPTIONS] # show environment and configuration status
llem doctor [OPTIONS] # verify Docker images match the host toolchain
llem report-gaps [OPTIONS] # propose invariants corpus entries from runtime feedback
llem --version # print version and exit

llem run

Run an LLM efficiency experiment

Arguments:

ArgumentTypeRequiredDescription
configpathnoPath to the experiment or study YAML config

Options:

FlagShortTypeDefaultDescription
--output-ostrOutput directory for results
--dry-runflagfalseValidate config and estimate VRAM without running
--quiet-qflagfalseSuppress progress bars
--verbose-vint0Increase verbosity (-v=INFO, -vv=DEBUG)
--skip-preflightflagfalseSkip Docker pre-flight checks (GPU visibility, CUDA/driver compatibility)
--resumeflagfalseResume most recent interrupted study
--resume-dirpathResume a specific study directory
--no-lockflagfalseDisable GPU lock files (advanced)

llem config

Show environment and configuration status

Options:

FlagShortTypeDefaultDescription
--verbose-vint0Increase verbosity (-v=INFO, -vv=DEBUG)

llem doctor

Verify Docker images match the host ExperimentConfig schema

llem report-gaps

Propose invariants corpus entries from runtime observations

Options:

FlagShortTypeDefaultDescription
--sourcestrruntime-warningsFeedback source to scan. Only 'runtime-warnings' is wired in this release.
--study-dirpathStudy directory to scan. Repeat the flag to pass multiple.
--enginestrFilter: only propose rules for this engine (transformers/vllm/tensorrt).
--outpathOutput path for proposed YAML fragments (one YAML document per gap, separated by '---').
--include-exceptionsflagfalseAlso propose rules from runtime exceptions. Disabled by default; exceptions ship through a different review path.
--verbose-vint0Increase verbosity (-v=INFO, -vv=DEBUG)

llem study

Write and prepare study files (llem run stays the executor).