Issue with hailomz compile - Missing Key parser in YAML File

General
1

Hello Hailo Community,

I’m a student from South Korea currently studying AI and embedded systems. As part of my project, I fine-tuned a YOLOv8n model for object detection and am now trying to convert it into the Hailo Executable Format (HEF) using the Hailo Model Zoo. My goal is to evaluate its performance on Hailo hardware.

Here’s what I have done so far:

  1. I fine-tuned the YOLOv8n model and exported it in ONNX format.
  2. I prepared a calibration dataset for quantization.
  3. I created a model_config.yaml file based on the official documentation, specifying the input and output shapes, quantization mode, and parser settings.
  4. I attempted to compile the model using the hailomz compile command with the following configuration:
    hailomz compile yolov8n
    –ckpt /workspace/yolov8n.onnx
    –calib-path /workspace/calibration_data/data/images
    –yaml /workspace/model_config.yaml
    –hw-arch hailo8l

However, I encountered the following error:
omegaconf.errors.ConfigAttributeError: Missing key parser
full_key: parser
object_type: dict

Below is the model_config.yaml file I am currently using:
model:
name: yolov8n
framework: onnx
input_shapes:
- [1, 3, 640, 640]
input_names:
- images
output_shapes:
- [1, 5, 8400]
output_names:
- output0

quantization:
mode: int8
calib_data_path: /workspace/calibration_data/data/images

network:
network_name: yolov8n
parser:
start_node_shapes:
- [1, 3, 640, 640]
end_node_shapes:
- [1, 5, 8400]
nodes:
- images
- output0

My Questions:

  1. Am I missing any required keys or parameters in my model_config.yaml file? If so, could you point out the missing parts?
  2. Could this issue be related to unsupported operations in my ONNX model, and if so, how can I identify and resolve them?
  3. Is there a specific way to debug this problem further, or should I take a different approach?

I am using:

  • Hailo Dataflow Compiler v3.30.0
  • HailoRT v4.20.0
  • Hailo Model Zoo v2.14.0

Your guidance would be greatly appreciated. Thank you for your time and support!

2

Hey @qkrtlgud0620,

Welcome to the Hailo Community!

I noticed you’re trying to go directly from ONNX to HEF, but there are a couple of important intermediate steps in our workflow:

Summary of the Workflow

  1. Parse → ONNX to HAR (Validate the input model)
  2. Optimize → HAR to Optimized/Quantized HAR
  3. Compile → Optimized HAR to HEF (Final deployment format)

Each step is crucial for proper model conversion and optimization. Let me know if you’d like help with implementing any of these steps!

Best regards,
Omria