Implementation:Tensorflow Tfjs Python Inference

Knowledge Sources	Tensorflow_Tfjs TensorFlow.js API
Domains	Deep_Learning, Inference, Python_Tooling
Last Updated	2026-02-10 06:00 GMT

Overview

This module provides a Python helper function for running TensorFlow.js model inference by invoking a Node.js binary via subprocess. It constructs the CLI command with the required model path, input directory, and output directory arguments, then executes it and captures stdout/stderr. If the subprocess exits with a non-zero return code, a ValueError is raised containing the stderr output.

Code Reference

Source Location

tfjs-inference/python/inference.py (GitHub)

Function Signature

def predict(binary_path,
            model_path,
            inputs_dir,
            outputs_dir,
            backend=None,
            tf_output_name_file=None):

Imports

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import subprocess

I/O Contract

Parameter	Type	Description
binary_path	str	Path to the Node.js inference binary (absolute preferred)
model_path	str	Directory containing the TensorFlow.js model JSON file
inputs_dir	str	Directory with input data, shape, and dtype files
outputs_dir	str	Directory where output data, shape, and dtype files are written
backend	str or None	Optional backend selection (cpu or wasm); defaults to cpu
tf_output_name_file	str or None	Optional file specifying TF output names; uses model defaults if absent

Output	Description
(none)	On success, inference results are written to outputs_dir
ValueError	Raised if subprocess exits with non-zero status, containing stderr

Implementation Details

The function builds a command list from the provided arguments and spawns the process using subprocess.Popen. Communication is performed via popen.communicate() to capture both stdout and stderr. Optional flags for backend selection and output name configuration are appended only when provided.

tfjs_inference_command = [
    binary_path, model_path_option, inputs_dir_option, outputs_dir_option
]

if tf_output_name_file:
    tfjs_inference_command.append('--tf_output_name_file=' + tf_output_name_file)

if backend:
    tfjs_inference_command.append('--backend=' + backend)

popen = subprocess.Popen(
    tfjs_inference_command,
    stdin=subprocess.PIPE,
    stdout=subprocess.PIPE,
    stderr=subprocess.PIPE)
stdout, stderr = popen.communicate()

Usage Example

from inference import predict

predict(
    binary_path='/usr/local/bin/tfjs-inference',
    model_path='/models/my_model/',
    inputs_dir='/data/inputs/',
    outputs_dir='/data/outputs/',
    backend='cpu'
)
# Results are written to /data/outputs/ as data, shape, and dtype files

Related Pages

• Tensorflow_Tfjs_MathBackendCPU - The CPU backend that may serve as the execution backend for inference