CLI Reference

This page lists every option of bsllmner2_extract and bsllmner2_select. Default values are taken from the implementation; see bsllmner2/cli_common.py.

bsllmner2_extract

Runs Named Entity Recognition (NER) over BioSample records. Output: bsllmner2-results/extract/{run_name}.json (see Data Formats).

Common Options

Option	Type	Default	Description
--bs-entries PATH	path	(required)	Input BioSample JSON or JSONL file.
--model STR	string	llama3.1:70b	Ollama model identifier.
--thinking BOOL	bool	false	Enable the model's thinking mode. Accepts true/false/1/0/yes/no/on/off.
--max-entries INT	int	-1	Process only the first N entries. -1 means all.
--ollama-host URL	string	http://localhost:11434	Ollama server URL. Usually inherited from the OLLAMA_HOST environment variable.
--debug	flag	off	Enable DEBUG-level logging.
--run-name STR	string	{model}_{YYYYmmdd_HHMMSS}	Identifier used in output and resume file names.
--resume	flag	off	Resume from an incomplete run. See Resume.
--batch-size INT	int	1024	Entries per batch. Each batch dumps a resume file.
--num-ctx INT	int	4096	Ollama context length (num_ctx).

Extract-Specific Options

Option	Type	Default	Description
--prompt PATH	path	bsllmner2/prompts/extract.yml	YAML prompt file (list of {role, content} messages).
--format PATH	path	(none)	JSON Schema file passed as the Ollama format option. Example: bsllmner2/format/cell_line.schema.json.

bsllmner2_select

Extract + ontology mapping. Output: bsllmner2-results/select/select_{run_name}.json (see Data Formats).

Common Options

Same as bsllmner2_extract.

Select-Specific Options

Option	Type	Default	Description
--select-config PATH	path	(required)	Select configuration JSON. Schema: see Data Formats.
--mapping PATH	path	(none)	Mapping TSV used to compute evaluation metrics against a gold standard. Format: see Data Formats.
--no-reasoning	flag	reasoning on	Disable reasoning in the LLM selection step. When omitted, the LLM emits a reasoning field alongside each picked term ID.

bsllmner2_select reuses bsllmner2_extract's NER stage with a prompt and JSON Schema dynamically built from --select-config. See Select Mode for the Stage 1 prompt construction details.

Resume

--resume continues a prior run that ended with status != "completed". The same --run-name must be supplied so the CLI can locate the existing resume files.

Files written atomically per batch under BSLLMNER2_RESULT_DIR:

File	Written by	Contents
extract/{run_name}_resume.json	bsllmner2_extract, bsllmner2_select	list[ExtractEntry] of completed extract entries.
select/select_{run_name}_resume.json	bsllmner2_select	list[SelectEntry] of completed select entries.

Resume behaviour:

• bsllmner2_extract -- accessions present in the extract resume file are skipped; remaining entries are processed.
• bsllmner2_select -- the CLI cross-checks both resume files:
  • Accessions present in both files are skipped.
  • Orphans (in extract, missing in select) are re-run through the select stage before normal batch processing resumes.
  • If the select resume file references an accession the extract resume file does not contain, the run aborts with ResumeDataError to prevent silent data corruption.

When a run finishes with status == "completed", the corresponding resume files are removed automatically.

Exit Codes

Code	Meaning
0	Run finished with status == "completed" and zero batch errors.
1	Run was aborted or one or more batches reported errors. The result JSON is still written.