Skip to content

Reference: codex_ml.cli metrics

Synopsis

python -m codex_ml.cli metrics <subcommand> [options]

Subcommands

  • ingest — read NDJSON and produce a tidy CSV (and optionally Parquet).
  • summary — print quick statistics from NDJSON metrics.

Options (ingest)

  • --input PATH (required): NDJSON input file.
  • --out-csv PATH (required): CSV output path.
  • --out-parquet PATH (optional): Parquet output path (requires pandas).
  • --run-id STR (optional): Label written into the run_id column.
  • --schema PATH (optional): JSON Schema to validate each record.
  • --to-sqlite PATH (optional): Load rows into a SQLite table.
  • --to-duckdb PATH (optional): Load rows into a DuckDB table.
  • --table STR (optional): Destination table name (default: metrics).
  • --mode {replace,append,fail} (optional): DuckDB write semantics (default: replace).
  • --chunk-size INT (optional): SQLite executemany() batch size (default: 5000).
  • --create-index (flag): Build (run_id, key, epoch) index in SQLite.
  • --allow-unsafe-table-name (flag): Bypass conservative identifier validation.

Options (summary)

  • --input PATH (required): NDJSON input file.