Implementation:Tensorflow Tfjs Tf LoadLayersModel

Summary

tf.loadLayersModel is the primary TensorFlow.js API for loading a pre-trained Keras-style layers model from a URL or custom I/O handler. It fetches the model topology (architecture) and binary weight files, deserializes them, and returns a fully reconstructed LayersModel ready for inference or further training.

API

tf.loadLayersModel(pathOrIOHandler: string|io.IOHandler, options?: io.LoadOptions): Promise<LayersModel>

Source

• tfjs-layers/src/models.ts:L248-270

Type

API Doc

Signature

export async function loadLayersModel(
  pathOrIOHandler: string|io.IOHandler,
  options?: io.LoadOptions
): Promise<LayersModel>

// LoadOptions
// {
//   strict?: boolean (default true),
//   onProgress?: (fraction: number) => void
// }

Parameters

Parameter	Type	Required	Description
pathOrIOHandler	io.IOHandler	Yes	URL to the model.json file, or a custom IOHandler for loading from non-HTTP sources (e.g., local filesystem, IndexedDB)
options	io.LoadOptions	No	Optional configuration for the loading process

LoadOptions

Option	Type	Default	Description
strict	boolean	true	If true, all model weights must match loaded weights exactly; mismatches throw an error
onProgress	(fraction: number) => void	undefined	Callback invoked during weight download with a fraction (0 to 1) indicating progress

I/O

• Inputs: URL to model.json or an IOHandler instance, plus optional load options
• Outputs: Promise<LayersModel> with both model topology and weights fully loaded

Loading Process

The function performs the following steps internally:

1. Parse the model.json to extract the model topology and weight manifest.
2. Resolve binary weight shard URLs relative to the model.json location.
3. Download all binary weight shards (potentially in parallel).
4. Deserialize the binary data into typed arrays matching each weight's dtype and shape.
5. Reconstruct the model architecture from the topology.
6. Assign loaded weight values to the corresponding model variables.

Example

// Load pretrained GPT-2 weights
const model = await tf.loadLayersModel(
  'https://storage.googleapis.com/gpt2/model.json',
  {onProgress: (frac) => console.log(`${(frac*100).toFixed(0)}% loaded`)}
);

Implements

Principle:Tensorflow_Tfjs_Pretrained_Weight_Loading

Environment:Tensorflow_Tfjs_Browser_Runtime Heuristic:Tensorflow_Tfjs_Backend_Selection_Strategy

Domains

NLP Model_Loading

Sources

TensorFlow.js

Related Pages

Environments

• Environment:Tensorflow_Tfjs_Browser_Runtime -- Browser runtime (WebGL / WebGPU / WASM / CPU backends)

Heuristics

• Heuristic:Tensorflow_Tfjs_Backend_Selection_Strategy -- Choose the optimal compute backend for your environment