Hi Hailo Team,
I trained a custom YOLOv8 object detection model, exported it to ONNX, and successfully compiled it to a HEF using the Hailo AI Software Suite Docker.
Host Environment
-
Hailo AI Software Suite: 2026-04
-
Hailo Dataflow Compiler (DFC): 5.3.0
hailo --version
Hailo Dataflow Compiler v5.3.0
Target Device
-
Hailo-15H Camera
-
Software Version: 1.6.1
-
HailoRT Version: 4.20.1
hailortcli --version
HailoRT-CLI version 4.20.1
The model compiles successfully:
Compilation complete
Saved HEF to: /local/workspace/microhard_detect.hef
However, after copying the HEF to the Hailo-15H device:
hailortcli parse-hef microhard_detect.hef
I get:
Unsupported hef version 3
Failed parsing HEF file
HAILO_INVALID_HEF(26)
At the same time, the stock model already present on the device parses correctly:
hailortcli parse-hef yolov8s.hef
which confirms the runtime is working.
My assumption is that the HEF generated by DFC 5.3.0 is not compatible with the device runtime HailoRT 4.20.1.
Could you please advise:
-
Is this a known DFC/HailoRT version mismatch?
-
What is the recommended compiler version for Hailo-15 software 1.6.1 / HailoRT 4.20.1?
-
Is there a way to obtain the matching DFC 4.20.x toolchain?
-
Alternatively, should the Hailo-15H camera software be upgraded to a newer version that supports HEFs generated by DFC 5.3.0?
Thanks in advance.