Skip to main content
Implementation of the m decompose run CLI command. Accepts a task prompt (from a text file or interactive input), calls the multi-step LLM decomposition pipeline to produce a structured list of subtasks each with constraints and inter-subtask dependencies, then validates and topologically reorders the subtasks before writing a JSON result file and a rendered Python script to the specified output directory.

​
Functions

​
FUNC reorder_subtasks 

reorder_subtasks(subtasks: list[DecompSubtasksResult]) -> list[DecompSubtasksResult]
Topologically sort subtasks by their depends_on relationships. Args:
  • subtasks: List of subtask dicts, each with a "tag" and optional "depends_on" field.
Returns:
  • list[DecompSubtasksResult]: The subtasks reordered so that dependencies
  • come before dependents, with numbering prefixes updated.
Raises:
  • ValueError: If a circular dependency is detected.

​
FUNC verify_user_variables 

verify_user_variables(decomp_data: DecompPipelineResult, input_var: list[str] | None) -> DecompPipelineResult
Validate that all required input variables and dependencies exist. Args:
  • decomp_data: The decomposition pipeline result containing subtasks.
  • input_var: User-provided input variable names, or None for none.
Returns:
  • The (possibly reordered) decomposition data.
Raises:
  • ValueError: If a subtask requires an input variable that was not provided, or depends on a subtask tag that does not exist.

​
FUNC run 

run(out_dir: Annotated[Path, typer.Option(help='Path to an existing directory to save the output files.')], out_name: Annotated[str, typer.Option(help='Name for the output files. Defaults to "m_decomp_result".')] = 'm_decomp_result', input_file: Annotated[str | None, typer.Option(help='Path to a text file containing user queries.')] = None, model_id: Annotated[str, typer.Option(help='Model name/id used to run the decomposition pipeline. Defaults to "mistral-small3.2:latest", valid for the "ollama" backend.')] = 'mistral-small3.2:latest', backend: Annotated[DecompBackend, typer.Option(help='Backend used for inference. Options: "ollama", "openai", and "rits".', case_sensitive=False)] = DecompBackend.ollama, backend_req_timeout: Annotated[int, typer.Option(help='Timeout in seconds for backend requests. Defaults to "300".')] = 300, backend_endpoint: Annotated[str | None, typer.Option(help='Backend endpoint / base URL. Required for "openai" and "rits".')] = None, backend_api_key: Annotated[str | None, typer.Option(help='Backend API key. Required for "openai" and "rits".')] = None, version: Annotated[DecompVersion, typer.Option(help='Version of the mellea program generator template to use.', case_sensitive=False)] = DecompVersion.latest, input_var: Annotated[list[str] | None, typer.Option(help='Optional user input variable names. You may pass this option multiple times. Each value must be a valid Python identifier.')] = None, log_mode: Annotated[LogMode, typer.Option(help='Readable logging mode. Options: "demo" or "debug".', case_sensitive=False)] = LogMode.demo) -> None
Runs the m decompose CLI workflow and writes generated outputs. Reads user queries from a file or interactive input, runs the decomposition pipeline for each task job, and writes one JSON file, one rendered Python program, and any generated validation modules under a per-job output directory. Args:
  • out_dir: Existing directory under which per-job output directories are created.
  • out_name: Base name used for the per-job output directory and generated files.
  • input_file: Optional path to a text file containing one or more task prompts. Each non-empty line is processed as a separate task job. When omitted, the command prompts interactively for one task.
  • model_id: Model identifier used for all decomposition pipeline stages.
  • backend: Inference backend used to execute model calls.
  • backend_req_timeout: Request timeout in seconds for backend inference calls.
  • backend_endpoint: Endpoint URL or base URL required by remote backends.
  • backend_api_key: API key required by remote backends.
  • version: Template version used to render the generated Python program. latest resolves to the most recently declared concrete version.
  • input_var: Optional user input variable names to expose in generated prompts and programs. Each name must be a valid non-keyword Python identifier.
  • log_mode: Logging verbosity for CLI and pipeline execution.
Raises:
  • AssertionError: If out_name is invalid, out_dir does not name an existing directory, input_file does not name an existing file, or any declared input_var is not a valid Python identifier.
  • ValueError: If input_file exists but contains no non-empty task lines.
  • Exception: Propagates pipeline, rendering, parsing, or file-writing failures. Any output directories created earlier in the run are removed before the exception is re-raised.

​
Classes

​
CLASS DecompVersion

Available template versions for generated decomposition programs. Newer concrete versions must be declared after older ones so that latest can resolve to the most recently declared template version. Attributes:
  • latest: Sentinel value that resolves to the last declared concrete template version.
  • v1: Version 1 of the decomposition program template.
  • v2: Version 2 of the decomposition program template.