A conversion exits nonzero on failure. Capture both stdout and stderr when investigating it: conversion progress and error diagnostics go to stderr, while the final Results output goes to stdout, and some input failures retain Python tracebacks. Verbosity controls logging detail, not whether the conversion succeeded. See troubleshooting and the error reference.
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). (default: output.zip)
--module.type
neuralspot | zephyr | cmake | nsx | cmsis_pack
neuralspot
Module type (default: neuralspot)
--module.name
str
helia_aot_nn
Module name (default: helia_aot_nn)
--module.prefix
str
aot
Prefix added to sources for unique namespace (default: aot)
--module.schedule
table | static
table
Operator dispatch shape in the generated model: ‘table’ (default) keeps the runtime function-pointer table and per-node callback seam; ‘static’ emits straight-line direct calls (no table, no loop, no indirect dispatch). (default: table)
If true, the module will use internal, statically allocated arenas. If false, the caller must bind every region via <prefix>_bind_arena() / <prefix>_bind_arenas() before model_init. (default: True)
Deprecated — retained for backwards compatibility but no longer changes generated runtime behavior. model_init always invokes hydrate_constants between context_init and the operator init loop. Override the weak hydrate_constants symbol for custom hydration mechanisms (DMA / async pre-stage / model swap). (default: True)
If true, write a machine-readable residency report (<prefix>_residency.json) alongside the emitted module. Mirrors the verbose log summary as JSON for tooling that needs to introspect arena layout, staged-vs-cold residency, and per-tensor placement post-planning. (default: False)
Allow auto knob values that change numerics relative to the knob’s default (e.g. fast accumulation) (default: False)
--optimization.accumulation
auto | precise | fast
auto
FP16 accumulation: precise (standard layout, the library’s default accumulation), fast (packed FP16 weights, one FP16 chain per output) or auto (default: AUTO)
--optimization.kernel
auto | specialized | generic
auto
Int8 convolution and depthwise entry: specialized (shape-specialized ns-cmsis-nn direct entries where they apply), generic (the general entries, fewer kernels linked) or auto; both are exact (default: AUTO)