Skip to content

defines

Classes

ModelArgs

Model configuration for the conversion process.

Attributes:

  • path (Path) –

    Path to the model file.

  • subgraph (int) –

    Subgraph index to process.

  • type (str | None) –

    Type of the model (e.g., tflite, litert)

  • name (str) –

    Name of the model.

  • description (str | None) –

    Optional description of the model.

  • version (str | None) –

    Version of the model.

ModuleArgs

Module configuration for the conversion process.

Attributes:

  • path (Path) –

    Output path for the generated module. Use a directory for an unpacked module, a .zip suffix for a generic zip archive, or a .pack suffix for an Open-CMSIS-Pack archive (requires module.type=cmsis_pack). Defaults to "output.zip".

  • type (ModuleType) –

    Type of the module (e.g., neuralspot, zephyr, cmake, nsx, cmsis_pack). Defaults to neuralspot.

  • name (ValidModuleNameType) –

    Name of the module. Defaults to "helia_aot_nn".

  • prefix (ValidCVarNameType) –

    Prefix added to sources for unique namespace. Defaults to "aot".

  • schedule (ScheduleMode) –

    Operator dispatch shape emitted into <prefix>_model.c. table (default) keeps the runtime function-pointer table and per-node callback seam; static emits straight-line direct calls. Defaults to table.

TestArgs

Test configuration for the conversion process.

Attributes:

  • enabled (bool) –

    Include test case. Defaults to False.

  • tolerance (float) –

    Test tolerance (integers are coerced). Defaults to 1.0.

  • skip_verification (bool) –

    Skip runtime output verification.

  • golden_data (Path | None) –

    Optional path to golden input/output npz file.

  • num_iterations (int) –

    Number of times to run the same stimulus through the model. Defaults to 1.

  • state_feedback (list[tuple[int, int]]) –

    (output_index, input_index) pairs copied back between iterations so recurrent state carries across invocations. Inputs named here are seeded from the stimulus only on the first iteration. Defaults to no feedback.

MemoryArgs

Memory configuration for the conversion process.

Attributes:

  • planner (MemoryPlannerType) –

    Memory planner strategy. Defaults to greedy.

  • constraints (list[MemoryConstraint] | None) –

    Memory constraints (ordered by preference). Defaults to None.

  • tensors (list[AttributeRuleset]) –

    Per-tensor attribute rulesets.

  • allocate_arenas (bool) –

    If true, generated module declares its own arena buffers.

  • auto_hydrate_constants (bool) –

    Deprecated — retained for backwards compatibility but no longer changes generated runtime behavior. <prefix>_model_init always invokes <prefix>_hydrate_constants between <prefix>_context_init and the operator init loop, so kernels that read constants in their _init hook always observe hydrated arenas. Override the weak <prefix>_hydrate_constants symbol for DMA / async pre-stage / model-swap behavior; model_init calls the override at the same fixed point.

PlatformArgs

Platform configuration for the conversion process.

Attributes:

  • name (str) –

    Target platform name. Defaults to "apollo510_evb".

  • cpu (str | None) –

    CPU core type (e.g., cortex-m55). Optional for built-ins.

  • speeds (list[int]) –

    Supported clock speeds in MHz.

  • memories (dict[MemoryType, int]) –

    Memory sizes in bytes.

  • capabilities (list[SocCapability]) –

    Available SoC capabilities.

  • preferred_memory_order (list[MemoryType]) –

    Placement preference for planner.

  • min_alignment (int | None) –

    Minimum alignment in bytes.

DocumentationArgs

Documentation configuration for the generated module.

Attributes:

  • html (bool) –

    Generate html documentation site (experimental).

ConvertArgs

Configuration for the conversion process.

Attributes:

  • path (Path) –

    Path to yaml configuration.

  • model (ModelArgs) –

    Model configuration.

  • module (ModuleArgs) –

    Module configuration.

  • test (TestArgs) –

    Test configuration.

  • transforms (list[TransformSpec]) –

    Transforms configuration.

  • memory (MemoryArgs) –

    Memory configuration.

  • platform (PlatformArgs) –

    Target platform configuration.

  • verbose (int) –

    Verbosity level (0-3).

  • log_file (Path | None) –

    Optional log file path.

Functions

from_yaml classmethod 
from_yaml(path: str | Path) -> ConvertArgs

Load a conversion config from a YAML file.

The canonical programmatic entry point mirroring the CLI's --path (without the CLI's flag-override merge):

config = ConvertArgs.from_yaml("my-model.yaml")
AotConverter(config=config).convert()

Parameters:

  • path
    (str | Path) –

    Path to a YAML mapping of conversion options.

Returns:

  • ConvertArgs –

    The validated configuration; unknown nested keys warn as usual.

Raises:

Functions