[error] Reshape is needed for layers: ..., but adding a reshape has failed

I encountered an error when converting an ONNX file, which is a monocular depth estimation model, into a HAR file and attempting to run ClientRunner.compile().

Below is the error message. Under what conditions does this error occur, and are there any known workarounds?

Error message:

[info] To achieve optimal performance, set the compiler_optimization_level to "max" by adding performance_param(compiler_optimization_level=max) to the model script. Note that this may increase compilation time.
[info] Original names list for layer depth_model_fixed_for_hailo/conv_feature_splitter1_2 is too long (>10), Trimming it to 10
[info] Original names list for layer depth_model_fixed_for_hailo/conv_feature_splitter2_2 is too long (>10), Trimming it to 10
[info] Original names list for layer depth_model_fixed_for_hailo/conv_feature_splitter3_2 is too long (>10), Trimming it to 10
[info] Original names list for layer depth_model_fixed_for_hailo/conv_feature_splitter4_2 is too long (>10), Trimming it to 10
[info] Original names list for layer depth_model_fixed_for_hailo/conv_feature_splitter5_2 is too long (>10), Trimming it to 10
[info] Original names list for layer depth_model_fixed_for_hailo/conv_feature_splitter6_2 is too long (>10), Trimming it to 10
[info] Original names list for layer depth_model_fixed_for_hailo/conv_feature_splitter7_2 is too long (>10), Trimming it to 10
[info] Original names list for layer depth_model_fixed_for_hailo/conv_feature_splitter8_2 is too long (>10), Trimming it to 10
[info] Original names list for layer depth_model_fixed_for_hailo/conv_feature_splitter9_2 is too long (>10), Trimming it to 10
[info] Original names list for layer depth_model_fixed_for_hailo/conv_feature_splitter10_2 is too long (>10), Trimming it to 10
[info] Original names list for layer depth_model_fixed_for_hailo/conv_feature_splitter11_2 is too long (>10), Trimming it to 10
[info] Original names list for layer depth_model_fixed_for_hailo/conv_feature_splitter12_2 is too long (>10), Trimming it to 10
[info] Loading network parameters
[info] Starting Hailo allocation and compilation flow
[warning] Failed to add spatial reshapes
[error] Mapping Failed (allocation time: 6s)
Reshape is needed for layers: ew_add19, precision_change0, reduce_mean1_layer_normalization53, ew_sub1_layer_normalization53, 

(...snip, a large number of node names)

, but adding a reshape has failed.

I appreciate any guidance you can provide.
Best regards.

Facing the same issue while converting depth anything, any update on this? Many thanks!

@ChunHung Did you successfully compile depth anything models?

No i have not yet figure out how hailo trying to do a layer norm, do you use their AI suite to compile? Secondly, did anyone try to use alls to solve this?

Hey @ChunHung, yes, I tried both AI SW Suites—DFC v29 and v30—but encountered the same compilation issue. It would be helpful if the authors provided additional information on converting VIT and BEiT transformers to the Hailo inference format.