hpx validate

Run hardware-in-the-loop validation suite (MLPerf Tiny models)

function

hpx validate

Python

Run hardware-in-the-loop validation suite (MLPerf Tiny models)

src/helia_profiler/cli/validation_app.py:85

hpx validate [OPTIONS]

Options

hpx validate options

Option	Type	Default	Description
--models	str		Comma-separated model IDs (default: all). See hpx validate --list.
--models-file	Optional[Path]		YAML registry of custom validation models and comparison groups.
--model-paths	str		Comma-separated .tflite paths for an ad hoc comparison.
--comparison-group	str	custom	Shared decision group for models supplied through --model-paths.
--model-arena-size	int	524288	Arena size in bytes for models supplied through --model-paths.
--engines	str		Comma-separated engines: rt,aot,tflm,et,executorch,helia-rt,helia-aot (default: all).
--executorch-backends	str	both	ExecuTorch CMSIS-NN providers: arm, ns, or both (default: both). TFLM always uses ARM CMSIS-NN; heliaRT and heliaAOT always use ns-cmsis-nn.
--ns-cmsis-nn-ref	str		Exact ns-cmsis-nn commit/ref used by heliaRT, heliaAOT, and ExecuTorch/ns.
--power	both | on | off	off	Power matrix: off (default) | on (only Joulescope runs) | both.
--power-boards	str		Comma-separated boards allowed to use power capture (default: all selected boards).
--boards	str		Comma-separated board IDs (default: apollo510_evb).
--toolchains	str		Comma-separated toolchains: gcc,armclang/acfe,atfe (default: board defaults).
--interfaces	str		Comma-separated interfaces/transports: rtt,uart,swo,usb_cdc (default: board defaults).
--memories	str		Comma-separated model placement presets: auto,tcm,sram,mram,psram (default: board defaults).
--suite	smoke | models-rt | models-aot | complete		Preset suite. 'smoke' defaults unset axes to models=kws, engines=helia-rt, toolchains=arm-none-eabi-gcc, interfaces=rtt, memories=auto. 'models-rt' and 'models-aot' default unset axes to all MLPerf Tiny models, Apollo510 + Apollo330mP, gcc + atfe, rtt, auto memory, and the selected engine. 'complete' runs the same axes for helia-rt, helia-aot, TFLM/CMSIS-NN, and both ExecuTorch CMSIS-NN providers. Explicit axis flags always win.
--jlink-serials	str		Comma-separated board=serial entries for multi-board validation.
--power-serials	str		Comma-separated board=Joulescope-serial entries for powered multi-board validation.
--power-gpios	str		Comma-separated board=gate:state:go entries for powered boards without default sync wiring.
--repeat	int	1	Repeat each selected case N times for stress testing (default: 1).
--output-dir	Path	results/validation	Where to write per-case artifacts + summary report (default: ./results/validation).
--timeout	float	900	Per-case timeout in seconds (default: 900).
-k, -k	str		Pytest keyword expression — filter cases by substring match (e.g. 'kws-aot').
--junit-xml	Optional[Path]		Emit JUnit-XML report at this path (for CI consumption).
--list	bool	false	List matching cases and exit without running.
--verbose, -v	int	0

Examples

hpx validate examples

Hardware validation — runs canonical MLPerf Tiny models end-to-end

against a real EVB + J-Link (and optional Joulescope).

Examples:

  hpx validate                         # Apollo510 reliability matrix, power off

  hpx validate --list                  # preview what would run

  hpx validate --models kws,ic         # subset by model

  hpx validate --engines aot           # subset by engine

  hpx validate --power off             # skip Joulescope (default)

  hpx validate --boards apollo3p_evb --repeat 2 --power off

                                       # require two passing iterations per case

  hpx validate -k kws-aot              # pytest keyword filter

  hpx validate --suite smoke           # quick preset: kws / helia-rt / gcc / rtt / auto

  hpx validate --suite models-rt       # 16-case RT sweep: 2 boards x 4 models x 2 toolchains

  hpx validate --suite models-aot      # 16-case AOT sweep: 2 boards x 4 models x 2 toolchains

  hpx validate --suite complete        # RT + AOT + TFLM + ExecuTorch sweep

---

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