Unable to Force-load TensorFlowLiteSelectTfOps.framework, created with Selective Build, in iOS about tensorflow HOT 9 OPEN

I would appreciate any working solution to this, as long i can run my model on Select Ops based on a TensorFlowLiteSelectTFOps library selectively built based on the tflite model itself. Thanks a lot in advance! :)

from tensorflow.

Hi @tanpengshi, it looks like the system is unable to properly link the google protobuf library ... can you ensure you are following all the directions here correctly: https://www.tensorflow.org/lite/guide/build_ios#using_local_swift_or_objective-c_apis

particularly I see that your path does not point to the TF root directory

pod 'TensorFlowLiteSwift', :path => '../../local-podspecs/TensorFlowLiteSwift.podspec'

Where as the source says:

pod 'TensorFlowLiteSwift', :path => '<your_tensorflow_root_dir>'

Let me know if that works. Thanks.

from tensorflow.

I have tried the above and I am still getting the same error. In addition I did :

  pod 'TensorFlowLiteSwift'
  pod 'TensorFlowLiteSelectTfOps', '~> 0.0.1-nightly'

Then replace i replaced the TensorFlowLiteSelectTfOps.framework with my selectively build TensorFlowLiteSelectTfOps.framework, and I still get the same error when building! I suspect the error comes inherently from the selectively built framework. Are you able to replicate the error I get?

Thank you!

from tensorflow.

Hi @tanpengshi, are you willing to share your model(s) that are used to make your selective build? I'm wondering if any particular ops are causing the issue. Perhaps you can just make a dummy model which includes at least one of all the ops you are using and share that.

from tensorflow.

tflite_CNN_LSTM_dummy.zip

Hi @pkgoogle, here is the dummy model that contains all the ops!

from tensorflow.

Hi @tanpengshi, I attempted to reproduce it with your model... I'm actually failing on the build the select framework step -- considering you were able to build the Select framework, can you share the steps that you use to build it?

I did this:

git switch r2.14 # to match your TF version
./configure
<Said yes to configure for iOS>
<copied your model file over>
bash tensorflow/lite/ios/build_frameworks.sh --input_models=tflite_CNN_LSTM_dummy.tflite --target_archs=x86_64,armv7,arm64

my output:

INFO: Analyzed target //tensorflow/lite/ios/tmp:TensorFlowLiteSelectTfOps_framework (473 packages loaded, 39899 targets configured).
INFO: Found 1 target...
ERROR: /Users/xxxxxxxxx/git/tensorflow/tensorflow/BUILD:1153:21: declared output 'tensorflow/libtensorflow_framework.2.dylib' was not created by genrule. This is probably because the genrule actually didn't create this output, or because the output was a directory and the genrule was run remotely (note that only the contents of declared file outputs are copied from genrules run remotely)
ERROR: /Users/xxxxxxxxx/git/tensorflow/tensorflow/BUILD:1153:21: Executing genrule //tensorflow:libtensorflow_framework.2.dylib_sym [for tool] failed: not all outputs were created or valid
realpath: illegal option -- -
usage: realpath [-q] [path ...]
Target //tensorflow/lite/ios/tmp:TensorFlowLiteSelectTfOps_framework failed to build
Use --verbose_failures to see the command lines of failed build steps.
INFO: Elapsed time: 2868.450s, Critical Path: 470.36s
INFO: 7940 processes: 1075 internal, 6865 local.
FAILED: Build did NOT complete successfully

from tensorflow.

Hi @pkgoogle I made a mistake, the tensorflow version should be r2.16 instead of r.2.14. When I used r2.16, I can successfully build the model, while r2.14 produced similar error.

from tensorflow.

Hmm I'm now currently running into a different issue... which python version are you using? I'm using an M1 and Xcode = 15.0.1. Any other potential environmental difference you can possibly think of?

ERROR: /Users/xxxxxxxx/git/tensorflow/tensorflow/python/BUILD:650:24: Linking tensorflow/python/_pywrap_tensorflow_internal.so [for tool] failed: (Exit 1): cc_wrapper.sh failed: error executing command (from target //tensorflow/python:_pywrap_tensorflow_internal.so) external/local_config_cc/cc_wrapper.sh @bazel-out/darwin_arm64-opt-exec-50AE0418-ST-c5da4e9d584e/bin/tensorflow/python/_pywrap_tensorflow_internal.so-2.params
ld: building exports trie: duplicate symbol '__ZN40TableStruct_tsl_2fprotobuf_2fdnn_2eproto7offsetsE'
clang: error: linker command failed with exit code 1 (use -v to see invocation)
Error in child process '/usr/bin/xcrun'. 1
Target //tensorflow/lite/ios/tmp:TensorFlowLiteSelectTfOps_framework failed to build
Use --verbose_failures to see the command lines of failed build steps.
INFO: Elapsed time: 4574.330s, Critical Path: 373.41s
INFO: 14670 processes: 407 internal, 14263 local.

from tensorflow.

My apologies again, if you use TensorFlow r2.16, you will run into the issue later on in Xcode:

The TensorFlow version should instead be r2.9, with the .frameworks bazel installed with Python 3.9 installed with Anaconda with M3 and Xcode 15.3

from tensorflow.