What is "smuffer" means when compile the har file?

General
1

Hi, team

Im trying the latest hailo dfc (3.30.0) and faced this message when i compile an quantized har file;

[2025-02-25 16:31:14.774] [default] [info] Ran from command: "/usr/local/bin/hailomz compile --performance --ckpt glass_simplified.onnx --calib-path calib_path/ --yaml resnet_v1_50.yaml --classes 10"

[2025-02-25 16:31:14.775] [default] [info] Loading network parameters

[2025-02-25 16:31:15.092] [default] [info] Starting Hailo allocation and compilation flow

[2025-02-25 16:31:15.114] [default] [info] Model name: resnet_v1_50

[2025-02-25 16:31:15.648] [default] [info] No successful assignment for conv42, most common errors: Agent infeasible

conv42 failed on kernel validation: Smuffers in input is not supported in conv42

conv42: Minimum needed amount of subclusters was not reached. Needed: 5, but requests only 1 subclusters

conv42: Minimum needed amount of subclusters was not reached. Needed: 5, but requests only 2 subclusters

conv42: Minimum needed amount of subclusters was not reached. Needed: 5, but requests only 3 subclusters

conv42: Minimum needed amount of subclusters was not reached. Needed: 5, but requests only 4 subclusters

[2025-02-25 16:31:15.648] [default] [info] No successful assignment for conv43_dummy_10472, most common errors:

[2025-02-25 16:31:16.709] [default] [info] No successful assignment for conv42, most common errors: Agent infeasible

conv42 failed on kernel validation: Smuffers in input is not supported in conv42

conv42: Minimum needed amount of subclusters was not reached. Needed: 5, but requests only 1 subclusters

conv42: Minimum needed amount of subclusters was not reached. Needed: 5, but requests only 2 subclusters

conv42: Minimum needed amount of subclusters was not reached. Needed: 5, but requests only 3 subclusters

conv42: Minimum needed amount of subclusters was not reached. Needed: 5, but requests only 4 subclusters

[2025-02-25 16:31:16.709] [default] [info] No successful assignment for conv43_dummy_21224, most common errors:

[2025-02-25 16:31:17.294] [default] [info] e[1;41mMapping Failed (allocation time: 1s)e[0m

[2025-02-25 16:31:17.296] [default] [info] Builder fail with status 110: No successful assignments: conv42 errors:

Agent infeasible

conv42 failed on kernel validation: Smuffers in input is not supported in conv42

conv42: Minimum needed amount of subclusters was not reached. Needed: 5, but requests only 1 subclusters

conv42: Minimum needed amount of subclusters was not reached. Needed: 5, but requests only 2 subclusters

conv42: Minimum needed amount of subclusters was not reached. Needed: 5, but requests only 3 subclusters

conv42: Minimum needed amount of subclusters was not reached. Needed: 5, but requests only 4 subclusters

[2025-02-25 16:31:17.301] [default] [info] No successful assignments: conv42 errors:

Agent infeasible

conv42 failed on kernel validation: Smuffers in input is not supported in conv42

conv42: Minimum needed amount of subclusters was not reached. Needed: 5, but requests only 1 subclusters

conv42: Minimum needed amount of subclusters was not reached. Needed: 5, but requests only 2 subclusters

conv42: Minimum needed amount of subclusters was not reached. Needed: 5, but requests only 3 subclusters

conv42: Minimum needed amount of subclusters was not reached. Needed: 5, but requests only 4 subclusters

Does Anyone (Admins) have an Idea with this kind of error and can point me in the right direction?
Thank you!

2

Hey @Gao_Yunpeng ,

Welcome to the Hailo Community!

Those error messages typically indicate resource allocation and compatibility issues with your Hailo model compilation. Here are some steps to troubleshoot:

  1. Input Format Problem
  • The “Smuffers in input” error suggests your input format isn’t compatible with the Hailo compiler
  • Check your input tensor format, padding, and tensor layout
  • Convert to a supported format if needed
  1. Subcluster Allocation
  • The compiler requires at least 5 subclusters for conv42
  • Try increasing resource allocation:
performance_param(compiler_optimization_level="max")

Hope this helps! Let me know if you need more specific guidance.