Understaing DFC information when compiling YOLOv8

Hi, I managed to compile YOLOv8m from ONNX to HEF. The command I used was
hailomz compile --ckpt yolov8m.onnx --calib my-tiny-dataset --yaml yolov8m.yaml --end-node-names YOLOV8M-FEATURE-MAPS-NODES
The tiny dataset I used for calibration had only 20 images.

Although an HEF model was generated, I saw some apparent red flags that I don’t quite stand, as shown below

[info] Solving the allocation (Mapping), time per context: 59m 59s
Context:0/3 Iteration 4: Trying parallel mapping...  
          cluster_0  cluster_1  cluster_2  cluster_3  cluster_4  cluster_5  cluster_6  cluster_7  prepost 
 worker0  V          X          V          V          V          V          V          V          V       
 worker1  V          V          V          V          V          V          V          V          V       
 worker2  *          *          *          *          *          *          *          *          V       
 worker3  V          V          V          V          V          V          V          V          V       
Context:1/3 Iteration 4: Trying parallel mapping...  
          cluster_0  cluster_1  cluster_2  cluster_3  cluster_4  cluster_5  cluster_6  cluster_7  prepost 
 worker0  V          *          V          V          V          V          V          V          V       
 worker1  *          *          *          *          *          *          *          *          V       
 worker2  *          *          *          *          *          *          *          *          V       
 worker3  *          *          *          *          *          *          *          *          V       
Context:2/3 Iteration 4: Trying parallel mapping...  
          cluster_0  cluster_1  cluster_2  cluster_3  cluster_4  cluster_5  cluster_6  cluster_7  prepost 
 worker0  V          *          V          V          V          V          V          V          V       
 worker1  *          *          *          *          *          *          *          *          V       
 worker2  V          V          V          V          V          V          V          V          V       
 worker3  *          *          *          *          *          *          *          *          V       
Context:3/3 Iteration 0: Trying parallel splits...   
          cluster_0  cluster_1  cluster_2  cluster_3  cluster_4  cluster_5  cluster_6  cluster_7  prepost 
 worker0                                                                                                  
 worker1  V          V          V          V          V          V          V          V          V       
 worker2  V          V          V          V          V          V          V          V          V       
 worker3  *          *          *          *          *          *          *          *          V       

There are my questions:

  1. What is the document that discusses this map? I skimmed the DFC User Guide release 3.28.0, but did not see a map like above.
  2. What is this map trying to tell me? Judging by the red color, I guess there is something wrong with the model. I have not run the model on the Hailo-8 HW yet.

Thanks

Hi @rsun,

Your model seems successfully compiled (:
There’s nothing to worry about there, this is just an info log for the allocation part of the compilation.

Thanks for the confirmation. Still it’d be helpful to understand the info log, especially regarding how the resource allocation would affect the inference performance.

@rsun The info log shows the 4 workers trying to find an allocation for the device’s clusters.

For information about the model’s resource allocation, I’d recommend running the profiler with runtime data.