Skip to content
heliaPROFILER
Reference
HELIA

results

What a profiling run produces: the run result, its per-layer and PMU detail, and the firmware and run metadata that say how it was measured.

Every name on this page is imported from helia_profiler.

API tier: stable

Generated from the src/helia_profiler tree fe269596788d6ab1604cf433559e96b5f48c1d13.

Machine-readable model

  • LayerResultclassProfiling result for a single model layer (averaged across iterations).
  • PresetResultclassResults for a single PMU counter preset (e.g.
  • FirmwareMetaclassMetadata reported by the profiler firmware at startup.
  • PmuResultclassComplete PMU profiling result across all presets.
  • RunMetadataclassAccumulated run metadata — enriched by stages, consumed by reports.
  • ProfileResultclassComplete profiling result — the public return type of hpx.profile().
class

LayerResult

Python

Profiling result for a single model layer (averaged across iterations).

src/helia_profiler/results/models.py:42

LayerResult(
    id: int | str,
    op: str,
    counters: dict[str, float] = dict(),
    cycles: float | None = None,
    overflow: bool = False,
    source_index: int | None = None,
) -> None

dataclass

Profiling result for a single model layer (averaged across iterations).

source_index is the ORIGINAL tflite operator index this measured layer corresponds to, when the op label carries one (#218). id is the firmware’s execution position, which on AOT engines is NOT the original index — helia-aot skips ops without renumbering.

API tier: stable

class

PresetResult

Python

Results for a single PMU counter preset (e.g.

src/helia_profiler/results/models.py:66

PresetResult(
    name: str,
    header: list[str] = list(),
    iterations: list[list[LayerResult]] = list(),
    layers: list[LayerResult] = list(),
) -> None

dataclass

Results for a single PMU counter preset (e.g. basic_cpu).

API tier: stable

class

FirmwareMeta

Python

Metadata reported by the profiler firmware at startup.

src/helia_profiler/results/models.py:97

FirmwareMeta(
    model_size: int | None = None,
    arena_size: int | None = None,
    allocated_arena: int | None = None,
    input_size: int | None = None,
    output_size: int | None = None,
    num_tensors: int | None = None,
    num_inputs: int | None = None,
    num_outputs: int | None = None,
    num_presets: int | None = None,
    system_clock_hz: int | None = None,
    profiled_infer_count: int | None = None,
    profiled_infer_total_us: int | None = None,
    profiled_infer_avg_us: int | None = None,
    clean_infer_count: int | None = None,
    clean_infer_total_cycles: int | None = None,
    clean_infer_avg_cycles: int | None = None,
    clean_infer_avg_us: int | None = None,
    clean_stalled_iters: int | None = None,
    clean_partial_iters: int | None = None,
    clean_ref_cycles: int | None = None,
    clean_dwt_rate_cyc: int | None = None,
    clean_dwt_rate_us: int | None = None,
    clean_attach_wait_us: int | None = None,
    psram: PsramInfo | None = None,
    presets: tuple[str, ...] = (),
    measured_clock_hz: int | None = None,
) -> None

dataclass

Metadata reported by the profiler firmware at startup.

All fields are optional because older firmware versions may not report every field.

API tier: stable

attribute

reported_model_bytes

Python

modelsize when it is a usable byte count, else None.

src/helia_profiler/results/models.py:164

reported_model_bytes: int | None

model_size when it is a usable byte count, else None.

The wire parser keeps an unparseable HPX_MODEL_SIZE as the raw string it received, so a corrupted or foreign line reaches consumers as text. Anything that needs a number asks here instead of assuming; the raw value stays on model_size so diagnostics can still quote what the device actually said (#281).

class

PmuResult

Python

Complete PMU profiling result across all presets.

src/helia_profiler/results/models.py:179

PmuResult(
    meta: FirmwareMeta,
    presets: dict[str, PresetResult] = dict(),
    layers: list[LayerResult] = list(),
    overflow_detected: bool = False,
    groups: dict[str, list[LayerResult]] = dict(),
) -> None

dataclass

Complete PMU profiling result across all presets.

API tier: stable

class

RunMetadata

Python

Accumulated run metadata — enriched by stages, consumed by reports.

src/helia_profiler/results/models.py:285

RunMetadata(
    hpx_version: str = '',
    run_id: str = '',
    timestamp: str = '',
    config_snapshot: dict[str, Any] = dict(),
    platform: PlatformInfo | None = None,
    model: ModelInfo | None = None,
    toolchain: ToolchainInfo | None = None,
    build_images: tuple[BuildImage, ...] = (),
    engine: EngineInfo | None = None,
    firmware: FirmwareMeta | None = None,
    memory_plan: 'MemoryPlan | None' = None,
    timing: TimingInfo | None = None,
    compatibility: CompatibilityResolution | None = None,
    dependencies: 'DependencyProvenance | None' = None,
) -> None

dataclass

Accumulated run metadata — enriched by stages, consumed by reports.

API tier: stable

class

ProfileResult

Python

Complete profiling result — the public return type of hpx.profile().

src/helia_profiler/results/models.py:595

ProfileResult(
    pmu: PmuResult,
    power: PowerResult | None = None,
    power_observation: PowerObservation | None = None,
    power_terminal: PowerTerminalRecord | None = None,
    on_device_power: OnDevicePowerSummary | None = None,
    metadata: RunMetadata = RunMetadata(),
    report_paths: list[Path] = list(),
) -> None

dataclass

Complete profiling result — the public return type of hpx.profile().

This is the one object a programmatic user needs. It carries everything: PMU data, optional power data, run metadata, and report file paths.

API tier: stable

Machine-readable: results.json · results.md, complete and lossless.