API Surface
heliaRT's public API is intentionally minimal and fully compatible with upstream LiteRT. This page documents the heliaRT-specific additions.
heliaRT Version Header
#include "tensorflow/lite/micro/heliart_version.h"
// Returns the heliaRT release version string
const char* version = HELIART_VERSION; // e.g. "v1.12.2"
This header provides the HELIART_VERSION macro, which is managed by release-please and updated automatically on every release.
Upstream LiteRT API
heliaRT does not modify or extend the upstream LiteRT for Micro API. The following core types work identically:
| Type | Header | Purpose |
|---|---|---|
tflite::MicroInterpreter |
micro_interpreter.h |
Run inference |
tflite::Model |
model.h |
Load a .tflite flatbuffer |
tflite::MicroMutableOpResolver<N> |
micro_mutable_op_resolver.h |
Register operators |
tflite::MicroAllocator |
micro_allocator.h |
Arena memory management |
tflite::MicroProfiler |
micro_profiler.h |
Per-layer profiling |
For detailed API documentation, refer to the upstream LiteRT for Micro API docs.
Build-Time Configuration
Compile Definitions
| Macro | Set by | Meaning |
|---|---|---|
CMSIS_NN |
HELIA backend | NN library exposes CMSIS-NN API surface |
HELIA |
HELIA backend | Build uses OPTIMIZED_KERNEL_DIR=helia |
CONV_KERNEL_OPTIMIZED_FOR_SPEED |
Per-kernel knob | Conv2D optimized for latency |
FC_KERNEL_OPTIMIZED_FOR_SIZE |
Per-kernel knob | FullyConnected optimized for code size |
Zephyr Kconfig Options
| Option | Type | Description |
|---|---|---|
CONFIG_HELIA_RT |
bool | Enable heliaRT module |
CONFIG_HELIA_RT_BACKEND_REFERENCE |
choice | Use Reference kernels |
CONFIG_HELIA_RT_BACKEND_CMSIS_NN |
choice | Use open CMSIS-NN kernels |
CONFIG_HELIA_RT_BACKEND_HELIA |
choice | Use Ambiq HELIA kernels |
Next Steps
- Operator Coverage — which operators have optimized kernels
- Upgrading from LiteRT — migration guide