Implementation:Tensorflow Tfjs Layer Utils

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

Overview

This module provides utility functions for printing human-readable summaries of TensorFlow.js LayersModel instances. It formats layer names, types, input shapes, output shapes, parameter counts, and connectivity information in a tabular text layout. The main entry point is printSummary, which adapts its column layout based on whether the model has a sequential-like topology or a more complex graph structure.

Code Reference

Source Location

tfjs-layers/src/utils/layer_utils.ts (GitHub)

Key Imports

import {Container} from '../engine/container';
import {Layer, Node} from '../engine/topology';
import {countParamsInWeights} from './variable_utils';

Main Function

printSummary

export function printSummary(
    model: Container,
    lineLength?: number,
    positions?: number[],
    printFn: (message?: any, ...optionalParams: any[]) => void = console.log
): void

Prints a formatted summary of the model to the console (or a custom print function). Columns include:

• Layer (type) - Layer name and class name
• Input Shape - Shapes of inbound tensors
• Output shape - The output tensor shape
• Param # - Number of trainable parameters
• Receives inputs (non-sequential models only) - Connectivity info

Layout Defaults

Model Type	Line Length	Column Positions
Sequential-like	90	[0.32, 0.61, 0.89, 1]
Non-sequential (graph)	115	[0.24, 0.48, 0.70, 0.80, 1]

Internal Functions

• countTrainableParams(model): Counts total trainable parameters, using collectedTrainableWeights if available.
• isModelSequentialLike(model): Determines if the model has a linear topology (single path, no shared layers).
• printRow(fields, positions, printFn): Formats a row of fields at specified column positions.
• printLayerSummary(layer, positions, printFn): Prints one layer without connectivity info (sequential models).
• printLayerSummaryWithConnections(layer, positions, relevantNodes, printFn): Prints one layer with inbound connection details (graph models).

I/O Contract

Function	Input	Output
printSummary	Container (model), optional line length, positions, print function	void (text output via printFn)

The summary footer includes total params, trainable params, and non-trainable params counts.

Usage Example

import * as tf from '@tensorflow/tfjs';

const model = tf.sequential();
model.add(tf.layers.dense({units: 64, inputShape: [128]}));
model.add(tf.layers.dense({units: 10}));

model.summary();  // internally calls printSummary

// Custom print function to capture output
const lines: string[] = [];
model.summary(undefined, undefined, (line: string) => lines.push(line));

Related Pages

• Tensorflow_Tfjs_Training_Utils - Training weight utilities
• Tensorflow_Tfjs_Embedding_Layer - Example layer whose params appear in summaries
• Tensorflow_Tfjs_DepthwiseConv2D_Layer - Convolutional layer with parameter counts