run_metrics

run_metrics(
    ground_truth_csv,
    llm_output_csv,
    label,
    output_csv,
    model_id,
    codebook_path,
    report_file,
    columns=None,
    quantization_type=None,
    temperature=None,
    top_p=None,
    prompt_type=None,
    use_examples=None,
    chat_mode=None,
    reasoning=None,
    process_textbox=False,
    process_span=False,
    emissions_file=None,
    experiment_directory=None,
    timestamp=None,
    timing_file=None,
    char_counts_file=None,
    run_id=None,
)

Evaluate one model-output CSV against ground truth and persist the results.

Parameters

Name	Type	Description	Default
ground_truth_csv		Path to the human-labeled reference CSV.	required
llm_output_csv		Path to the model-generated annotation CSV.	required
label		Experiment label written to the metrics CSV, usually the task name.	required
output_csv		Path to the aggregate metrics CSV to create or update.	required
model_id		Stable identifier for the model and prompt configuration.	required
codebook_path		Path to the codebook used for the run.	required
report_file		Path to the per-column report text file.	required
columns		Optional subset of annotation column names to evaluate.	None
quantization_type		Optional quantization metadata string.	None
temperature		Optional temperature metadata value.	None
top_p		Optional top-p metadata value.	None
prompt_type		Optional prompt wrapper name stored as metadata.	None
use_examples		Optional boolean-like metadata flag.	None
chat_mode		Optional chat-history policy stored as metadata.	None
reasoning		Optional Ollama reasoning mode stored as metadata.	None
process_textbox		Whether textbox metrics should be computed.	False
emissions_file		Optional path to emissions.csv.	None
experiment_directory		Optional path to the per-run output directory.	None
timestamp		Optional timestamp string.	None
timing_file		Optional path to timing_data.json.	None
char_counts_file		Optional path to char_counts.json.	None

Returns

Name	Type	Description
		codebook_lab.types.MetricsRunResult for the completed evaluation.