Tensorflow

Supported event types: logs

The Tensorflow filter plugin allows running machine learning inference tasks on the records of data coming from input plugins or stream processors. This filter uses Tensorflow Lite as the inference engine, and requires Tensorflow Lite shared library to be present during build and at runtime.

Tensorflow Lite is a lightweight open source deep learning framework used for mobile and IoT applications. Tensorflow Lite only handles inference, not training. It loads pre-trained models (.tflite files) that are converted into Tensorflow Lite format (FlatBuffer). You can read more on converting Tensorflow models.

The Tensorflow plugin for Fluent Bit has the following limitations:

• Currently, supports single-input models

• Uses Tensorflow 2.3 header files

Configuration parameters

The plugin supports the following configuration parameters:

Key	Description	Default
include_input_fields	Include all input fields in filter's output.	true
input_field	Specify the name of the field in the record to apply inference on.	none
model_file	Path to the model file (.tflite) to be loaded by Tensorflow Lite.	none
normalization_value	Divide input values to normalization_value.	none

Creating a Tensorflow Lite shared library

To create a Tensorflow Lite shared library:

1. Clone the Tensorflow repository.

2. Install the Bazel package manager.

3. Run the following command to create the shared library:

   bazel build -c opt //tensorflow/lite/c:tensorflowlite_c  # see https://github.com/tensorflow/tensorflow/tree/master/tensorflow/lite/c

   The script creates the shared library bazel-bin/tensorflow/lite/c/libtensorflowlite_c.so.

4. Copy the library to a location such as /usr/lib that can be used by Fluent Bit.

Building Fluent Bit with Tensorflow filter plugin

The Tensorflow filter plugin is disabled by default. You must build Fluent Bit with the Tensorflow plugin enabled. In addition, it requires access to Tensorflow Lite header files to compile. Therefore, you must pass the address of the Tensorflow source code on your machine to the build script:

cmake -DFLB_FILTER_TENSORFLOW=On -DTensorflow_DIR=<AddressOfTensorflowSourceCode> ...

Command line

If the Tensorflow plugin initializes correctly, it reports successful creation of the interpreter, and prints a summary of model's input and output types and dimensions.

The command:

fluent-bit -i mqtt -p 'tag=mqtt.data' -F tensorflow -m '*' -p 'input_field=image' -p 'model_file=/home/user/model.tflite' -p 'include_input_fields=false' -p 'normalization_value=255' -o stdout

produces an output like:

[2020/08/04 20:00:00] [ info] Tensorflow Lite interpreter created!
[2020/08/04 20:00:00] [ info] [tensorflow] ===== input #1 =====
[2020/08/04 20:00:00] [ info] [tensorflow] type: FLOAT32  dimensions: {1, 224, 224, 3}
[2020/08/04 20:00:00] [ info] [tensorflow] ===== output #1 ====
[2020/08/04 20:00:00] [ info] [tensorflow] type: FLOAT32  dimensions: {1, 2}

Configuration file

fluent-bit.yaml

service:
  flush: 1
  daemon: off
  log_level: info

pipeline:
  inputs:
    - name: mqtt
      tag: mqtt.data

  filters:
    - name: tensorflow
      match: mqtt.data
      input_field: image
      model_file: /home/m/model.tflite
      include_input_fields: false
      normalization_value: 255

  outputs:
    - name: stdout
      match: '*'

fluent-bit.conf