inference/research
verifiedlow · 0.60updated 2026-10-15

HailoRT (Hailo AI Software Suite runtime)

Hailo-8, Hailo-8L, Hailo-10H, Hailo-15H

Fields

id
engines/hailo-rt
Repo
https://github.com/hailo-ai/hailort
Languages
C++Python
License
proprietary (free-of-charge binaries + open hailort repo)
First release
2021
Design goal
run Dataflow-Compiler-produced .hef binaries on Hailo-8/8L/10H NPUslow-latency edge inference with on-die model memory
Scheduling
hardware-scheduled dataflow
Backends
Hailo-8Hailo-8LHailo-10HHailo-15H
Formats
hef (Hailo Executable Format, INT8)hef INT4 (Hailo-10H LLMs)
Notable features
HEF binaries embed the quantized network + memory map; no general tensor-op interpretergenerative-AI (LLM) path exists only on Hailo-10H
Hardware caveats
Hailo-8/8L cannot run transformer-decode LLMs - vision CNNs onlymodel must be compiled through the licensed Dataflow Compiler (ONNX -> HAR -> optimize -> HEF)
Best for
sub-5W edge vision inferenceRaspberry Pi 5 AI kit acceleration
Avoid for
any generative/LLM workload on Hailo-8
notes
Engine record created 2026-10-15 to back Hailo-8 Model-Zoo benchmark rows. hailort GitHub repo hosts the runtime source; DFC toolchain is proprietary/licensed.

Notes

Engine record created 2026-10-15 to back Hailo-8 Model-Zoo benchmark rows. hailort GitHub repo hosts the runtime source; DFC toolchain is proprietary/licensed.

Cross-references

This record references no other records — it is a leaf. It is cited by 6.