Tensorflow lite in Android App

[LiangZhaoSRA]

Issue type

Bug

Have you reproduced the bug with TensorFlow Nightly?

Yes

Source

source

TensorFlow version

2.3

Custom code

Yes

OS platform and distribution

Ubuntu 20.04

Mobile device

Galaxy S24

Python version

3.8

Bazel version

No response

GCC/compiler version

No response

CUDA/cuDNN version

No response

GPU model and memory

No response

Current behavior?

Got the following error when try to run a tflite model on GPU:

W/System.err(30515): Ignoring failed delegate application: java.lang.IllegalArgumentException: Internal error: Failed to apply delegate: Attempting to use a delegate that only supports static-sized tensors with a graph that has dynamic-sized tensors.

When I visualize the model using Netron, there is no dynamic-sized tensor (see attached file).

Standalone code to reproduce the issue

W/System.err(30515): Ignoring failed delegate application: java.lang.IllegalArgumentException: Internal error: Failed to apply delegate: Attempting to use a delegate that only supports static-sized tensors with a graph that has dynamic-sized tensors.

Relevant log output

W/System.err(30515): Ignoring failed delegate application: java.lang.IllegalArgumentException: Internal error: Failed to apply delegate: Attempting to use a delegate that only supports static-sized tensors with a graph that has dynamic-sized tensors.

[sawantkumar]

hi @LiangZhaoSRA ,

If possible , can you share me the .tflite file please

[LiangZhaoSRA]

@sawantkumar Attaches is the tflite file. Thanks.
model.zip

[sawantkumar]

Hi @LiangZhaoSRA ,

I inspected your model using custom code and I found out that your model has dynamic sized tensors . If possible can you provide me the model conversion code.

Dynamic tensors found:
Name: Where_101;StatefulPartitionedCall/Where_101, Shape: [0 1]
Name: Squeeze_83;StatefulPartitionedCall/Squeeze_83, Shape: [0]
Name: GatherV2_83;StatefulPartitionedCall/GatherV2_83;GatherV2_99/axis;StatefulPartitionedCall/GatherV2_99/axis, Shape: [0]
Name: GatherV2_78;StatefulPartitionedCall/GatherV2_78;GatherV2_99/axis;StatefulPartitionedCall/GatherV2_99/axis, Shape: [0]
Name: GatherV2_79;StatefulPartitionedCall/GatherV2_79;GatherV2_99/axis;StatefulPartitionedCall/GatherV2_99/axis, Shape: [0]
Name: GatherV2_71;StatefulPartitionedCall/GatherV2_71;GatherV2_99/axis;StatefulPartitionedCall/GatherV2_99/axis, Shape: [0]
Name: GatherV2_66;StatefulPartitionedCall/GatherV2_66;GatherV2_99/axis;StatefulPartitionedCall/GatherV2_99/axis, Shape: [0]
Name: GatherV2_67;StatefulPartitionedCall/GatherV2_67;GatherV2_99/axis;StatefulPartitionedCall/GatherV2_99/axis, Shape: [0]
Name: GatherV2_59;StatefulPartitionedCall/GatherV2_59;GatherV2_99/axis;StatefulPartitionedCall/GatherV2_99/axis, Shape: [0]
Name: GatherV2_54;StatefulPartitionedCall/GatherV2_54;GatherV2_99/axis;StatefulPartitionedCall/GatherV2_99/axis, Shape: [0]
Name: GatherV2_55;StatefulPartitionedCall/GatherV2_55;GatherV2_99/axis;StatefulPartitionedCall/GatherV2_99/axis, Shape: [0]
Name: GatherV2_12;StatefulPartitionedCall/GatherV2_12;GatherV2_99/axis;StatefulPartitionedCall/GatherV2_99/axis, Shape: [0]
Name: GatherV2_13;StatefulPartitionedCall/GatherV2_13;GatherV2_99/axis;StatefulPartitionedCall/GatherV2_99/axis, Shape: [0]
Name: GatherV2_26;StatefulPartitionedCall/GatherV2_26;GatherV2_99/axis;StatefulPartitionedCall/GatherV2_99/axis, Shape: [0]
Name: GatherV2_27;StatefulPartitionedCall/GatherV2_27;GatherV2_99/axis;StatefulPartitionedCall/GatherV2_99/axis, Shape: [0]
Name: GatherV2_40;StatefulPartitionedCall/GatherV2_40;GatherV2_99/axis;StatefulPartitionedCall/GatherV2_99/axis, Shape: [0]
Name: GatherV2_41;StatefulPartitionedCall/GatherV2_41;GatherV2_99/axis;StatefulPartitionedCall/GatherV2_99/axis, Shape: [0]
Name: GatherV2_5;StatefulPartitionedCall/GatherV2_5;GatherV2_99/axis;StatefulPartitionedCall/GatherV2_99/axis, Shape: [0]
Name: GatherV2;StatefulPartitionedCall/GatherV2;GatherV2_99/axis;StatefulPartitionedCall/GatherV2_99/axis, Shape: [0]
Name: GatherV2_1;StatefulPartitionedCall/GatherV2_1;GatherV2_99/axis;StatefulPartitionedCall/GatherV2_99/axis, Shape: [0]
Name: GatherV2_47;StatefulPartitionedCall/GatherV2_47;GatherV2_99/axis;StatefulPartitionedCall/GatherV2_99/axis, Shape: [0]
Name: GatherV2_42;StatefulPartitionedCall/GatherV2_42;GatherV2_99/axis;StatefulPartitionedCall/GatherV2_99/axis, Shape: [0]
Name: GatherV2_43;StatefulPartitionedCall/GatherV2_43;GatherV2_99/axis;StatefulPartitionedCall/GatherV2_99/axis, Shape: [0]

[LiangZhaoSRA]

Hi Sawant,
I converted the torch model to tflite via torch to onnx, onnx to tensorflow and tensorflow to tflite. Attached are the scripts for model conversion. Please let me know if you have any questions. Thanks.
tr2lite.zip