Inference issue with Reshape node on EfficientNet-Lite0 model

Required prerequisites

What commit version of aidge do you use

aidge_core: commit 5e064e918e34fcb97958897ae8c8bfd5ee08bcdd on 2026, march 2nd (all modules have same versions)

Problem description

Trying inference with aidge on network EfficientNet-Lite0, loaded from onnx file, fp32 data type. Inference produces warnings that mention an issue with Reshape operator. Inference results are all nan values.

Expected behaviour : Inference should produce valid numerival values instead of nan.

Is this a regression ? Yes probably

Please provide logs in the form of a code block

[NOTICE] - Reshape_Op: ignoring non-empty Shape attribute because input#1 takes precedence
[WARNING] - Reshape_Op: unable to forwardDims() because output dims are data dependent on input#1
[NOTICE] - Reshape_Op: ignoring non-empty Shape attribute because input#1 takes precedence
[WARNING] - Reshape_Op: unable to forwardDims() because output dims are data dependent on input#1
[NOTICE] - Reshape_Op: ignoring non-empty Shape attribute because input#1 takes precedence
[WARNING] - Reshape_Op: unable to forwardDims() because output dims are data dependent on input#1
[WARNING] - Unable to forward dimensions (circular dependency and/or wrong dimensions and/or data
[WARNING]   dependent dimension?). Unable to compute output dims for nodes
[WARNING]   ["efficientnet-lite0_model_head_dense_BiasAdd_Gemm__89 (Gemm)",
[WARNING]   "efficientnet-lite0_model_head_Squeeze1 (Reshape)"].
[NOTICE] - Reshape_Op: ignoring non-empty Shape attribute because input#1 takes precedence

Reproducible example code

Given the size of the reproducer, here is a link to a public archive (18 MB, tar.gz) :

https://cloud.univ-grenoble-alpes.fr/s/DsQdJ9L8epwwC2n

To reproduce :

decompress the archive
source your aidge env
make run

In case it is useful : the original model is the fp32 tflite checkpoint from TensorFlow repo :

https://github.com/tensorflow/tpu/blob/master/models/official/efficientnet/lite/README.md

I converted it to onnx with tf2onnx python module. A visual inspection of the generated onnx file reveals no issue at least about the Reshape operator.