CHECK_SUCCESS failed with status=6: NMS score threshold set but no NMS output in model

General Discussion
, , ,
1

Hi everyone,

I am trying to run a custom model based on YOLO (version 2026/latest) on my Raspberry Pi 5 with Hailo-8L, but I am hitting a wall with an NMS error.

Environment:

  • Hardware: Raspberry Pi 5 26 TFLOPS

  • Accelerator: Hailo-8

  • Software: Hailo RT / Model Zoo

    The Error: When I try to run the inference, I get the following log:

    Plaintext

    NMS score threshold is set, but there is no NMS output in this model.
CHECK_SUCCESS failed with status=6

    What I’ve tried:

    1. Compiled the model using the Hailo Model Compiler.

    2. Followed various video guides and the official Hailo GitHub repository documentation.

    3. The model seems to load, but it fails right at the start of the pipeline.

      Question: It seems the model was compiled without the NMS integrated into the .hef file, or the application is expecting an NMS layer that isn’t defined in the post-processing configuration. How can I ensure the NMS is correctly included during the conversion from ONNX to HEF for this specific YOLO version? Do I need to modify the Allals (CLI) script to include nms_postprocess?

      Thanks in advance!

2

Hi @Jose_Fuertes,

This error typically means that your HEF was compiled without the NMS postprocess layer baked in, while the application pipeline you’re running expects it to be there. You might try re-compiling your model by adding the nms_postprocess command to your model script (ALLS file) - for example, adding a line like nms_postprocess(meta_arch=yolov8, engine=cpu, nms_scores_th=0.2, nms_iou_th=0.6) (adjust meta_arch to match your specific YOLO variant). You can look at the Hailo Model Zoo ALLS files for reference, e.g. hailo_model_zoo/hailo_model_zoo/cfg/alls/generic/yolov8s.alls at master · hailo-ai/hailo_model_zoo · GitHub - the key line being the nms_postprocess(...) call. Once you recompile the HEF with that command included and redeploy it, the runtime should find the NMS output layer it expects and the error should go away.

Thanks,
Michael.