TensorFlow.js Converter Issue

[TheRealCasmat]

Hey!

I'm able to use tensorflowjs converter properly as used in this notebook from @shmishra99. But when I try and use the converter in my docker container running on linux arm64, it fails and gives me this error:
# tensorflowjs_converter --input_format=tfjs_layers_model --output_format=keras models/tfjs/model.json models/keras/fla_model.h5

WARNING:root:Failure to load the inference.so custom c++ tensorflow ops. This error is likely caused the version of TensorFlow and TensorFlow Decision Forests are not compatible. Full error:/usr/local/lib/python3.10/site-packages/tensorflow_decision_forests/tensorflow/ops/inference/inference.so: cannot open shared object file: No such file or directory
Traceback (most recent call last):
  File "/usr/local/bin/tensorflowjs_converter", line 5, in <module>
    from tensorflowjs.converters.converter import pip_main
  File "/usr/local/lib/python3.10/site-packages/tensorflowjs/__init__.py", line 21, in <module>
    from tensorflowjs import converters
  File "/usr/local/lib/python3.10/site-packages/tensorflowjs/converters/__init__.py", line 21, in <module>
    from tensorflowjs.converters.converter import convert
  File "/usr/local/lib/python3.10/site-packages/tensorflowjs/converters/converter.py", line 38, in <module>
    from tensorflowjs.converters import tf_saved_model_conversion_v2
  File "/usr/local/lib/python3.10/site-packages/tensorflowjs/converters/tf_saved_model_conversion_v2.py", line 28, in <module>
    import tensorflow_decision_forests
  File "/usr/local/lib/python3.10/site-packages/tensorflow_decision_forests/__init__.py", line 64, in <module>
    from tensorflow_decision_forests import keras
  File "/usr/local/lib/python3.10/site-packages/tensorflow_decision_forests/keras/__init__.py", line 53, in <module>
    from tensorflow_decision_forests.keras import core
  File "/usr/local/lib/python3.10/site-packages/tensorflow_decision_forests/keras/core.py", line 62, in <module>
    from tensorflow_decision_forests.keras import core_inference
  File "/usr/local/lib/python3.10/site-packages/tensorflow_decision_forests/keras/core_inference.py", line 36, in <module>
    from tensorflow_decision_forests.tensorflow.ops.inference import api as tf_op
  File "/usr/local/lib/python3.10/site-packages/tensorflow_decision_forests/tensorflow/ops/inference/api.py", line 179, in <module>
    from tensorflow_decision_forests.tensorflow.ops.inference import op
  File "/usr/local/lib/python3.10/site-packages/tensorflow_decision_forests/tensorflow/ops/inference/op.py", line 15, in <module>
    from tensorflow_decision_forests.tensorflow.ops.inference.op_dynamic import *
  File "/usr/local/lib/python3.10/site-packages/tensorflow_decision_forests/tensorflow/ops/inference/op_dynamic.py", line 24, in <module>
    raise e
  File "/usr/local/lib/python3.10/site-packages/tensorflow_decision_forests/tensorflow/ops/inference/op_dynamic.py", line 21, in <module>
    ops = tf.load_op_library(resource_loader.get_path_to_datafile("inference.so"))
  File "/usr/local/lib/python3.10/site-packages/tensorflow/python/framework/load_library.py", line 54, in load_op_library
    lib_handle = py_tf.TF_LoadLibrary(library_filename)
tensorflow.python.framework.errors_impl.NotFoundError: /usr/local/lib/python3.10/site-packages/tensorflow_decision_forests/tensorflow/ops/inference/inference.so: cannot open shared object file: No such file or directory

This is the docker file im running off:

FROM python:3.10.12-slim

WORKDIR /backend

COPY . /backend

RUN apt-get update && apt-get install -y pkg-config gcc g++ libhdf5-dev

RUN pip install --upgrade pip setuptools wheel

RUN pip install --no-cache-dir flask flask-sqlalchemy psycopg2-binary python-dotenv flask-cors pillow tensorflowjs tensorflow-aarch64==2.15.0 tensorflow_decision_forests==1.8.1

RUN apt-get clean && rm -rf /var/lib/apt/lists/*

EXPOSE 5000

ENV FLASK_APP=app.py
ENV FLASK_ENV=development

CMD ["flask", "run", "--host=0.0.0.0", "--debug"]

Thanks for any help!

[shmishra99]

Hi @TheRealCasmat ,

I have created a Dockerfile image using the same base image, python:3.10.12-slim (linux arm64). I am able to convert the Keras model to a TFJS model using tfjs-converter, using the same Colab snippet file that I shared previously.

In your Dockerfile, you can use pip install tensorflow==2.15.0 instead of pip install tensorflow-aarch64==2.15.0. Let me know if you have any specific reason to use tensorflow-aarch64.

Thank You!

[TheRealCasmat]

Hey @shmishra99!

I forgot to mention I wasn't exactly following the notebook you shared. I ran tensorflowjs_converter --input_format=tfjs_layers_model --output_format=keras models/tfjs/model.json models/keras/fla_model.h5 inside the container and got the error. Could you try that using this model on your end:
backend.zip

If that does work, could you share your docker image? I don't know really why i used tensorflow-aarch64... used tensorflow and built again but still getting error. The Dockerfile is the same as I used before but with the '-aarch64' removed. I'm not sure why the inference.so file cannot be opened, It does exist on my end.

---

Also, just a quick question, when using the converter in python, if I wanted to convert keras h5 model that is created using the command above back to a tfjs layers model, would I use tfjs.converters.dispatch_keras_h5_to_tfjs_layers_model_conversion('models/keras/fla-model.h5', 'models/tfjs/')?

Thanks again for your help!

[shmishra99]

Hi @TheRealCasmat ,

I'm building a Docker image using your provided Dockerfile. You could follow the steps I've tried, they might be helpful for you.

Here are the steps I've taken:

1. Created a Dockerfile using your provided file, making sure to include TensorFlow 2.15.0.
2. Ran the build command docker build -t my-container-image . with my app.py file as the dummy server.
   Output logs:
   
3. Run the container using the command: docker run -it my-container-image
4. Enter the container by executing: docker exec -it <container-id> bash
5. Run the converter.py script (as described in the Colab notebook).
   Output:
   

Let me know if these steps are help for you or not.

The command tfjs.converters.dispatch_keras_h5_to_tfjs_layers_model_conversion('models/keras/fla-model.h5', 'models/tfjs/') is the correct approach to convert your Keras H5 model to a TensorFlow.js model.

Thank You!!

[TheRealCasmat]

Hey @shmishra99!

Hmm still doesn’t work for me for some reason. It seems that on my end I’m not able to import tensorflowjs as tfjs and that gives the inference.so error. I created new script in the backend directory that is the same as the last code block in collab

Used:

FROM python:3.10.12-slim

WORKDIR /backend

COPY . /backend

RUN apt-get update && apt-get install -y pkg-config gcc g++ libhdf5-dev

RUN pip install --upgrade pip setuptools wheel

RUN pip install --no-cache-dir flask flask-sqlalchemy psycopg2-binary python-dotenv flask-cors pillow tensorflowjs tensorflow==2.15.0 tensorflow_decision_forests==1.8.1

RUN apt-get clean && rm -rf /var/lib/apt/lists/*

EXPOSE 5000

ENV FLASK_APP=app.py
ENV FLASK_ENV=development

CMD ["flask", "run", "--host=0.0.0.0", "--debug"]

[shmishra99]

Could you share the most recent error messages from the converter script, along with the command you're using to run it?

[TheRealCasmat]

WARNING:root:Failure to load the inference.so custom c++ tensorflow ops. This error is likely caused the version of TensorFlow and TensorFlow Decision Forests are not compatible. Full error:/usr/local/lib/python3.10/site-packages/tensorflow_decision_forests/tensorflow/ops/inference/inference.so: cannot open shared object file: No such file or directory
Traceback (most recent call last):
  File "/backend/r.py", line 2, in <module>
    import tensorflowjs as tfjs
  File "/usr/local/lib/python3.10/site-packages/tensorflowjs/__init__.py", line 21, in <module>
    from tensorflowjs import converters
  File "/usr/local/lib/python3.10/site-packages/tensorflowjs/converters/__init__.py", line 21, in <module>
    from tensorflowjs.converters.converter import convert
  File "/usr/local/lib/python3.10/site-packages/tensorflowjs/converters/converter.py", line 38, in <module>
    from tensorflowjs.converters import tf_saved_model_conversion_v2
  File "/usr/local/lib/python3.10/site-packages/tensorflowjs/converters/tf_saved_model_conversion_v2.py", line 28, in <module>
    import tensorflow_decision_forests
  File "/usr/local/lib/python3.10/site-packages/tensorflow_decision_forests/__init__.py", line 64, in <module>
    from tensorflow_decision_forests import keras
  File "/usr/local/lib/python3.10/site-packages/tensorflow_decision_forests/keras/__init__.py", line 53, in <module>
    from tensorflow_decision_forests.keras import core
  File "/usr/local/lib/python3.10/site-packages/tensorflow_decision_forests/keras/core.py", line 62, in <module>
    from tensorflow_decision_forests.keras import core_inference
  File "/usr/local/lib/python3.10/site-packages/tensorflow_decision_forests/keras/core_inference.py", line 36, in <module>
    from tensorflow_decision_forests.tensorflow.ops.inference import api as tf_op
  File "/usr/local/lib/python3.10/site-packages/tensorflow_decision_forests/tensorflow/ops/inference/api.py", line 179, in <module>
    from tensorflow_decision_forests.tensorflow.ops.inference import op
  File "/usr/local/lib/python3.10/site-packages/tensorflow_decision_forests/tensorflow/ops/inference/op.py", line 15, in <module>
    from tensorflow_decision_forests.tensorflow.ops.inference.op_dynamic import *
  File "/usr/local/lib/python3.10/site-packages/tensorflow_decision_forests/tensorflow/ops/inference/op_dynamic.py", line 24, in <module>
    raise e
  File "/usr/local/lib/python3.10/site-packages/tensorflow_decision_forests/tensorflow/ops/inference/op_dynamic.py", line 21, in <module>
    ops = tf.load_op_library(resource_loader.get_path_to_datafile("inference.so"))
  File "/usr/local/lib/python3.10/site-packages/tensorflow/python/framework/load_library.py", line 54, in load_op_library
    lib_handle = py_tf.TF_LoadLibrary(library_filename)
tensorflow.python.framework.errors_impl.NotFoundError: /usr/local/lib/python3.10/site-packages/tensorflow_decision_forests/tensorflow/ops/inference/inference.so: cannot open shared object file: No such file or directory

and just python r.py on the python file that has

import tensorflow as tf
import tensorflowjs as tfjs

# tf.keras.applications.MobileNetV3Large(
#     input_shape=None,
#     alpha=1.0,
#     minimalistic=False,
#     include_top=True,
#     weights='imagenet',
#     input_tensor=None,
#     classes=1000,
#     pooling=None,
#     dropout_rate=0.2,
#     classifier_activation='softmax',
#     include_preprocessing=True
# )

model = tf.keras.applications.MobileNetV3Large(
    input_shape=(224, 224, 3), weights='imagenet', classifier_activation='softmax'
)

tfjs.converters.save_keras_model(model, "sample_data/tfjs_model_keras_MobileNetV3Large")