Error compiling from custom yolo11 onnx

General
1

Totally cryptic message I guess while building the hef, in the final stages.
Any help?

[info] Solving the allocation (Mapping), time per context: 59m 59s
Context:0/4 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 V
worker1 V V V V V V V V V
worker2 * * * * * * * * V
worker3 * * * * * * * * V
Context:1/4 Iteration 4: Trying parallel mapping…
cluster_0 cluster_1 cluster_2 cluster_3 cluster_4 cluster_5 cluster_6 cluster_7 prepost
worker0 X V V V V V X V V
worker1 V V V V V V V V V
worker2 V V V V V V V V V
worker3 V V V V V V V V V
Context:2/4 Iteration 16: Trying parallel splits…
cluster_0 cluster_1 cluster_2 cluster_3 cluster_4 cluster_5 cluster_6 cluster_7 prepost
worker0
worker1
Context:2/4 Iteration 20: Trying parallel splits…
cluster_0 cluster_1 cluster_2 cluster_3 cluster_4 cluster_5 cluster_6 cluster_7 prepost
worker0
worker1
Context:2/4 Iteration 24: Trying parallel splits…
cluster_0 cluster_1 cluster_2 cluster_3 cluster_4 cluster_5 cluster_6 cluster_7 prepost
worker0
worker1
Context:2/4 Iteration 28: Trying parallel splits…
cluster_0 cluster_1 cluster_2 cluster_3 cluster_4 cluster_5 cluster_6 cluster_7 prepost
worker0
worker1
Context:2/4 Iteration 32: Trying parallel splits…
cluster_0 cluster_1 cluster_2 cluster_3 cluster_4 cluster_5 cluster_6 cluster_7 prepost
worker0
worker1
Context:2/4 Iteration 36: Trying parallel splits…
cluster_0 cluster_1 cluster_2 cluster_3 cluster_4 cluster_5 cluster_6 cluster_7 prepost
worker0
worker1
Context:2/4 Iteration 96: Trying parallel splits…
cluster_0 cluster_1 cluster_2 cluster_3 cluster_4 cluster_5 cluster_6 cluster_7 prepost
worker0
worker1
worker2
worker3
Reverts on pre-mapping validation: 21n: 36: Negative size is: 3
05:19 on split failed: 0a, iteration: 34: Negative size is: 3
Reverts on cluster mapping: 11teration: 39: Negative size is: 3
Reverts on inter-cluster connectivity: 150: Negative size is: 3
Reverts on pre-mapping validation: 66n: 37: Negative size is: 3
Reverts on split failed: 4a, iteration: 38: Negative size is: 3
[info] context_0 (context_0):
Iterations: 4
Reverts on cluster mapping: 0
Reverts on inter-cluster connectivity: 0
Reverts on pre-mapping validation: 0
Reverts on split failed: 0
[info] context_1 (context_1):
Iterations: 4
Reverts on cluster mapping: 1
Reverts on inter-cluster connectivity: 0
Reverts on pre-mapping validation: 0
Reverts on split failed: 0
[info] context_2 (context_2):
Iterations: 100
Reverts on cluster mapping: 11
Reverts on inter-cluster connectivity: 15
Reverts on pre-mapping validation: 66
Reverts on split failed: 4
[info] context_3 (context_3):
Iterations: 0 [ Did not run due to an error ]
Reverts on cluster mapping: 0
Reverts on inter-cluster connectivity: 0
Reverts on pre-mapping validation: 0
Reverts on split failed: 0
[info] context_4 (context_4):
Iterations: 0 [ Did not run due to an error ]
Reverts on cluster mapping: 0
Reverts on inter-cluster connectivity: 0
Reverts on pre-mapping validation: 0
Reverts on split failed: 0
[warning] Failed recording best-effort buffers: l4_portal_from_conv_feature_splitter1_1_dc_to_concat2 was not found in toposort recipe during buffers override record
[error] Mapping Failed (Timeout, allocation time: 1h 10m 9s)
[error] Mapping Failed (Timeout, allocation time: 1h 10m 9s)
Resolver didn’t find possible solution.
worker 0 failed on Splitter shutdown timeout reached (status: 31)
worker 1 failed on Splitter shutdown timeout reached (status: 31)
worker 2 failed on Splitter shutdown timeout reached (status: 31)
worker 3 failed on Splitter shutdown timeout reached (status: 31)

2

Hey @Thor,

This can be related to a lot of issues. Without more info about the ALLS file you’re using, the performance level, and the YAML configuration, these are just general options:

Summary of Solutions

Issue Solution
Model Too Large Reduce input size, use a smaller YOLO variant
Inter-Cluster Mapping Fails Limit max_parallel_workers=2
Concat Layer Issue Break Concat into smaller operations or use feature_splitter
Incorrect Hardware Target Use --hw-arch hailo8l during compilation
Mapping Timeout Set mapping_timeout=7200 and enable_auto_mapping=True
3

Hi @Omria,
Thanks. It is something related to the size of the model, as smaller models with same yaml and alls work, I’ll send you the alls and yaml later.

But now I have a more pressing question
Supposing I generate a rectangular onyx (w: 1536, h:864) using the following:
yolo export model=best.pt format=onnx opset=11 imgsz=(864,1536)
There are two places where input size is defined in the hailo yaml chain:
"image_dims”: in yolov11m_nms_config.json
input_shape: in yolo.yaml

To define a rectangular input shape to match the onnx, do I have to modify one or both?
Are the dimension w,h or h,w ?

Thanks

4

Hey @Thor ,

To properly define a rectangular input shape (864×1536) for your Hailo YAML chain that matches your ONNX model, you need to modify two key settings:

  1. In yolov11m_nms_config.json:

    • Update image_dims to (864, 1536)
    • This affects post-processing operations like NMS and decoding

  2. In yolo.yaml:

    • Set input_shape to (864, 1536)
    • This defines the network’s input tensor dimensions

Hailo Uses (H,W)

If you encounter issues after making these changes, try swapping the dimensions to (1536, 864) in both files.

5

Thank you.
However I did not see any change in the time hailo takes to process an image
Meaning,
a yolo model (square by default) converted to onnx without changing imgsz, then create a hef specifying square images 1536x1536
Vs
a yolo model (square by default) converted to onnx specifying imgsz 864x1536, then create a hef specifying images 864x1536
Hailo takes the same time to process
Why?