Class GraphModel<ModelURL>

Releases the memory used by the weight tensors and resourceManager.

Doc

Dispose intermediate tensors for model debugging mode (flag KEEP_INTERMEDIATE_TENSORS is true).

Doc

Executes inference for the model for given input tensors.

Returns

A single tensor if provided with a single output or no outputs are provided and there is only one default output, otherwise return a tensor array. The order of the tensor array is the same as the outputs if provided, otherwise the order of outputNodes attribute of the model.

Doc

Executes inference for the model for given input tensors in async fashion, use this method when your model contains control flow ops.

Returns

A Promise of single tensor if provided with a single output or no outputs are provided and there is only one default output, otherwise return a tensor map.

Doc

Get intermediate tensors for model debugging mode (flag KEEP_INTERMEDIATE_TENSORS is true).

Doc

Loads the model and weight files, construct the in memory weight map and compile the inference graph.

Synchronously construct the in memory weight map and compile the inference graph. Also initialize hashtable if any.

Doc

Execute the inference for the input tensors.

See

inputNodes

You can also feed any intermediate nodes using the NamedTensorMap as the input type. For example, given the graph InputNode => Intermediate => OutputNode, you can execute the subgraph Intermediate => OutputNode by calling model.execute('IntermediateNode' : tf.tensor(...));

This is useful for models that uses tf.dynamic_rnn, where the intermediate state needs to be fed manually.

For batch inference execution, the tensors for each input need to be concatenated together. For example with mobilenet, the required input shape is [1, 244, 244, 3], which represents the [batch, height, width, channel]. If we are provide a batched data of 100 images, the input tensor should be in the shape of [100, 244, 244, 3].

Returns

Inference result tensors. The output would be single tf.Tensor if model has single output node, otherwise Tensor[] or NamedTensorMap[] will be returned for model with multiple outputs.

Doc

Save the configuration and/or weights of the GraphModel.

An IOHandler is an object that has a save method of the proper signature defined. The save method manages the storing or transmission of serialized data ("artifacts") that represent the model's topology and weights onto or via a specific medium, such as file downloads, local storage, IndexedDB in the web browser and HTTP requests to a server. TensorFlow.js provides IOHandler implementations for a number of frequently used saving mediums, such as tf.io.browserDownloads and tf.io.browserLocalStorage. See tf.io for more details.

This method also allows you to refer to certain types of IOHandlers as URL-like string shortcuts, such as 'localstorage://' and 'indexeddb://'.

Example 1: Save model's topology and weights to browser local storage; then load it back.

const modelUrl =
   'https://storage.googleapis.com/tfjs-models/savedmodel/mobilenet_v2_1.0_224/model.json';
const model = await tf.loadGraphModel(modelUrl);
const zeros = tf.zeros([1, 224, 224, 3]);
model.predict(zeros).print();

const saveResults = await model.save('localstorage://my-model-1');

const loadedModel = await tf.loadGraphModel('localstorage://my-model-1');
console.log('Prediction from loaded model:');
model.predict(zeros).print();

Returns

A Promise of SaveResult, which summarizes the result of the saving, such as byte sizes of the saved artifacts for the model's topology and weight values.

Doc