Implementation:Tensorflow Serving Build Signature Def

Knowledge Sources	TensorFlow Serving TF SavedModel Utils
Domains	Model_Serialization, API_Design
Last Updated	2026-02-13 17:00 GMT

Overview

Wrapper for TensorFlow's build_tensor_info() and build_signature_def() utilities to construct SignatureDef protobufs for SavedModel export.

Description

These TensorFlow utility functions construct the protobuf messages that define a model's serving interface. build_tensor_info() converts a TensorFlow tensor into a TensorInfo proto (capturing name, dtype, and shape), while build_signature_def() assembles multiple TensorInfos into a complete SignatureDef with named inputs, outputs, and a method name.

In the TensorFlow Serving MNIST example, two signatures are constructed:

A classification signature (default) using serialized TF examples as input
A prediction signature (named "predict_images") using raw image tensors

Usage

Use these functions after training a model and before calling SavedModelBuilder. They bridge the gap between TensorFlow tensors in a session and the serialized serving interface.

Code Reference

Source Location

Repository: tensorflow/serving
File: tensorflow_serving/example/mnist_saved_model.py
Lines: L107-139

Signature

# build_tensor_info: Convert tensor to TensorInfo proto
tf.compat.v1.saved_model.utils.build_tensor_info(
    tensor  # tf.Tensor - the tensor to describe
) -> TensorInfo

# build_signature_def: Assemble inputs/outputs into SignatureDef
tf.compat.v1.saved_model.signature_def_utils.build_signature_def(
    inputs: dict,       # map of alias names to TensorInfo
    outputs: dict,      # map of alias names to TensorInfo
    method_name: str    # CLASSIFY_METHOD_NAME, PREDICT_METHOD_NAME, or REGRESS_METHOD_NAME
) -> SignatureDef

Import

import tensorflow as tf
# Access via:
# tf.compat.v1.saved_model.utils.build_tensor_info
# tf.compat.v1.saved_model.signature_def_utils.build_signature_def
# tf.compat.v1.saved_model.signature_constants.CLASSIFY_METHOD_NAME
# tf.compat.v1.saved_model.signature_constants.PREDICT_METHOD_NAME

I/O Contract

Inputs

Name	Type	Required	Description
tensor	tf.Tensor	Yes	For build_tensor_info: the tensor to describe
inputs	dict[str, TensorInfo]	Yes	For build_signature_def: named input tensor info map
outputs	dict[str, TensorInfo]	Yes	For build_signature_def: named output tensor info map
method_name	str	Yes	For build_signature_def: inference method constant

Outputs

Name	Type	Description
TensorInfo	protobuf	Proto describing tensor name, dtype, and shape
SignatureDef	protobuf	Complete serving signature with inputs, outputs, method

Usage Examples

Building Classification and Prediction Signatures

import tensorflow as tf

# Assume trained tensors exist: serialized_tf_example, prediction_classes, values, x, y

# 1. Build TensorInfo for each tensor
classification_inputs = tf.compat.v1.saved_model.utils.build_tensor_info(serialized_tf_example)
classification_outputs_classes = tf.compat.v1.saved_model.utils.build_tensor_info(prediction_classes)
classification_outputs_scores = tf.compat.v1.saved_model.utils.build_tensor_info(values)

# 2. Build classification signature (default serving signature)
classification_signature = tf.compat.v1.saved_model.signature_def_utils.build_signature_def(
    inputs={
        tf.compat.v1.saved_model.signature_constants.CLASSIFY_INPUTS: classification_inputs
    },
    outputs={
        tf.compat.v1.saved_model.signature_constants.CLASSIFY_OUTPUT_CLASSES: classification_outputs_classes,
        tf.compat.v1.saved_model.signature_constants.CLASSIFY_OUTPUT_SCORES: classification_outputs_scores
    },
    method_name=tf.compat.v1.saved_model.signature_constants.CLASSIFY_METHOD_NAME
)

# 3. Build prediction signature
tensor_info_x = tf.compat.v1.saved_model.utils.build_tensor_info(x)
tensor_info_y = tf.compat.v1.saved_model.utils.build_tensor_info(y)

prediction_signature = tf.compat.v1.saved_model.signature_def_utils.build_signature_def(
    inputs={'images': tensor_info_x},
    outputs={'scores': tensor_info_y},
    method_name=tf.compat.v1.saved_model.signature_constants.PREDICT_METHOD_NAME
)

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment