CLI Reference¶

Frozen inference¶

python -m src.ecg_sqi_inference predict
  --model {12lead-conformer,singlelead-conformer,12lead-rbfsvm,singlelead-rbfsvm}
  --input PATH
  --fs HZ
  --out PATH
  [--device {cpu,cuda}]

Option	Contract
`--model`	Required stable model identifier from the catalogue
`--input`	One supported ECG file or a recursively scanned directory
`--fs`	Positive source sampling frequency in hertz
`--out`	Output directory; created if absent
`--device`	Conformer device request; defaults to CPU

Additional commands:

python -m src.ecg_sqi_inference verify-bundles
python -m src.ecg_sqi_inference export-inference-bundle --out PATH

Classical SQI pipeline¶

python -m src.sqi_pipeline.cli
  [--profile {baseline,paper_aligned}]
  [--artifacts_dir PATH]
  [--seed INT]
  [--only STEP,STEP]
  [--fresh] [--force] [--verbose]

--fresh deletes generated artifacts below the selected output root before running. --force asks each stage to ignore its normal reuse checks. --only executes named stages from the selected profile; dependencies must already exist when an upstream stage is omitted.

Transformer pipeline¶

python -m src.transformer_pipeline.cli [GLOBAL OPTIONS] COMMAND

Command	Purpose	Mutating execution gate
`audit`	Validate protocol and frozen contracts	None
`plot`	Render current diagnostic figures	None
`report`	Build current generated report outputs	None
`extract-but`	Extract BUT data	`--run`
`clean-smoke`	Run public-data smoke reconstruction	`--run`
`build-v116`	Construct v116 protocol	`--run`
`split`	Materialise frozen split	`--run`
`train --model E31`	Train selected model	`--run`
`pipeline --train E31`	Orchestrate full path	`--run`

The explicit --run gates prevent an accidental expensive build from a help or audit command.