netron/source/keras.js

import * as json from './json.js';
import * as python from './python.js';

const keras = {};
const tfjs = {};

keras.ModelFactory = class {

    async match(context) {
        const identifier = context.identifier;
        const extension = identifier.lastIndexOf('.') > 0 ? identifier.split('.').pop().toLowerCase() : '';
        const group = await context.peek('hdf5');
        if (group && group.attributes && group.attributes.get('CLASS') !== 'hickle') {
            if (identifier === 'model.weights.h5') {
                return context.set('keras.model.weights.h5', group);
            }
            if (identifier === 'parameter.h5') {
                return context.set('hdf5.parameter.h5', group);
            }
            return context.set('keras.h5', group);
        }
        const json = await context.peek('json');
        if (json) {
            if (json.mxnet_version || (json.nodes && json.arg_nodes && json.heads)) {
                return null;
            }
            if (json.model_config || (json.class_name && json.config)) {
                return context.set('keras.config.json', json);
            }
            if (identifier === 'metadata.json' && json.keras_version) {
                return context.set('keras.metadata.json', json);
            }
        }
        const container = await tfjs.Container.open(context);
        if (container) {
            return context.set('tfjs', container);
        }
        const pickle = await context.peek('pkl');
        if (pickle && pickle.__class__ &&
            pickle.__class__.__module__ === 'keras.engine.sequential' &&
            pickle.__class__.__name__ === 'Sequential') {
            return context.set('tfjs.pickle', pickle);
        }
        // model.weights.npz
        const entries = await context.peek('npz');
        const regex = /^(__root__|layers\/.+|_layer_checkpoint_dependencies\/.+)\.npy$/;
        if (entries instanceof Map && entries.size > 0 && Array.from(entries).every(([name]) => regex.test(name))) {
            return context.set('keras.model.weights.npz', entries);
        }
        // keras_metadata.pb
        if (extension === 'pb' && context.stream && context.stream.length > 16) {
            const tags = await context.tags('pb');
            if (tags.size === 1 && tags.get(1) === 2) {
                const stream = context.stream;
                const buffer = stream.peek(Math.min(stream.length, 1024));
                const content = String.fromCharCode.apply(null, buffer);
                if (/root"/.test(content) && /\{\s*"class_name"\s*:/.test(content)) {
                    return context.set('keras.pb.SavedMetadata');
                }
            }
        }
        return null;
    }

    filter(context, match) {
        if (context.type === 'keras.metadata.json' && (match.type === 'keras.config.json' || match.type === 'keras.model.weights.h5' || match.type === 'keras.model.weights.npz')) {
            return false;
        }
        if (context.type === 'keras.config.json' && (match.type === 'keras.model.weights.h5' || match.type === 'keras.model.weights.npz')) {
            return false;
        }
        if (context.type === 'tfjs' && match.type === 'tf.tfjs.weights') {
            return false;
        }
        return true;
    }

    async open(context) {
        const request_json = async (context, name) => {
            try {
                context = await context.fetch(name);
            } catch {
                return null;
            }
            return await context.read('json');
        };
        const _create_config = (weights_store) => {
            const config = {};
            config.class_name = 'Model';
            config.config = {};
            config.config.layers = [];
            const snake_to_pascal_case = (name) => {
                return name.replace(/(^|_|\d)([a-z])/g, (match, p1, p2) => p1 === '_' ? p2.toUpperCase() : p1 + p2.toUpperCase());
            };
            for (const [name, value] of weights_store) {
                const layer = {};
                layer.name = name;
                layer.class_name = name.split('/').pop().replace(/_[0-9]+$/, '');
                layer.class_name = snake_to_pascal_case(layer.class_name);
                layer.config = {};
                layer.config.name = name;
                layer._trainable_variables = value;
                config.config.layers.push(layer);
            }
            return config;
        };
        const _load_state = (trackable, weights_store, assets_store, inner_path) => {
            inner_path = inner_path || '';
            if (trackable && trackable.config && Array.isArray(trackable.config.layers)) {
                /* eslint-disable no-use-before-define */
                _load_container_state(trackable, weights_store, assets_store, inner_path ? `${inner_path}/layers` : 'layers');
                /* eslint-enable no-use-before-define */
            } else {
                const weights = weights_store.get(inner_path);
                if (weights) {
                    trackable._trainable_variables = weights;
                }
            }
        };
        const _load_container_state = (container, weights_store, assets_store, inner_path) => {
            const used_names = new Map();
            for (const trackable of container.config.layers) {
                const pascal_to_snake_case = (name) => {
                    name = name.replace(/\W+/g, "");
                    name = name.replace(/(.)([A-Z][a-z]+)/g, (match, p1, p2) => `${p1}_${p2}`);
                    name = name.replace(/([a-z])([A-Z])/g, (match, p1, p2) => `${p1}_${p2}`);
                    return name.toLowerCase();
                };
                let name = pascal_to_snake_case(trackable.class_name);
                if (used_names.has(name)) {
                    const next = used_names.get(name) + 1;
                    used_names.set(name, next);
                    name = `${name}_${next}`;
                } else {
                    used_names.set(name, 0);
                }
                _load_state(trackable, weights_store, assets_store, `${inner_path}/${name}`);
            }
        };
        const read_weights_hdf5 = (group) => {
            const weights_store = new Map();
            const stack = [[group, '']];
            while (stack.length > 0) {
                const [group, path] = stack.pop();
                if (group.groups instanceof Map) {
                    const checkpoint = group.groups.get('layers') || group.groups.get('_layer_checkpoint_dependencies');
                    if (checkpoint) {
                        for (const [key, layer] of checkpoint.groups) {
                            const name = `${path ? `${path}/` : ''}layers/${key}`;
                            stack.push([layer, name]);
                            const values = [];
                            for (const vars of layer.groups) {
                                for (const [name, group] of vars[1].groups) {
                                    const variable = group.value;
                                    if (variable) {
                                        const layout = variable.littleEndian ? '<' : '>';
                                        const tensor = new keras.Tensor(name, variable.shape, variable.type, null, null, layout, variable.data);
                                        values.push(tensor);
                                    }
                                }
                            }
                            if (values.length > 0) {
                                weights_store.set(name, values);
                            }
                        }
                    }
                }
            }
            return weights_store;
        };
        const read_weights_numpy = (entries) => {
            const weights_store = new Map();
            for (const [path, array] of entries) {
                const file = path.split('/').map((name) => name === '_layer_checkpoint_dependencies' ? 'layers' : name).join('/');
                if (file.endsWith('.npy') && file.startsWith('layers/')) {
                    if (array.dtype.name === 'object' && array.shape.length === 0 && Array.isArray(array.data) && array.data.length === 1) {
                        const values = Object.values(array.data[0]).map((array) => {
                            const stride = array.strides.map((stride) => stride / array.itemsize);
                            const dataType = array.dtype.__name__;
                            const values = dataType === 'string' || dataType === 'object' ? array.flatten().tolist() : array.tobytes();
                            const encoding = dataType === 'string' || dataType === 'object' ? '|' : array.dtype.byteorder;
                            return new keras.Tensor('', array.shape, dataType, stride, null, encoding, values);
                        });
                        if (values.length > 0) {
                            const name = file.replace(/\.npy$/, '');
                            weights_store.set(name, values);
                        }
                    }
                }
            }
            return weights_store;
        };
        const request_weights = async (context) => {
            const formats = [
                ['model.weights.h5', 'hdf5', read_weights_hdf5],
                ['model.weights.npz', 'npz', read_weights_numpy],
            ];
            for (const [name, type, callback] of formats) {
                let content = null;
                try {
                    /* eslint-disable no-await-in-loop */
                    content = await context.fetch(name);
                    /* eslint-enable no-await-in-loop */
                } catch {
                    // continue regardless of error
                }
                if (content) {
                    /* eslint-disable no-await-in-loop */
                    const obj = await content.peek(type);
                    /* eslint-enable no-await-in-loop */
                    if (obj) {
                        return callback(obj);
                    }
                }
            }
            return new Map();
        };
        const open_model = async (format, producer, backend, config, weights) => {
            const metadata = await context.metadata('keras-metadata.json');
            return new keras.Model(metadata, format, producer, backend, config, weights);
        };
        switch (context.type) {
            case 'keras.config.json': {
                const obj = context.value;
                const config = obj.model_config ? obj.model_config : obj;
                const backend = obj.backend || '';
                let version = obj.keras_version ? obj.keras_version : null;
                if (!version) {
                    const metadata = await request_json(context, 'metadata.json');
                    if (metadata && metadata.keras_version) {
                        version = metadata.keras_version;
                    }
                }
                const format = `Keras${version ? ` v${version}` : ''}`;
                const weights_store = await request_weights(context);
                _load_state(config, weights_store);
                return open_model(format, '', backend, config, null);
            }
            case 'keras.model.weights.h5': {
                const group = context.value;
                const weights_store = read_weights_hdf5(group);
                const metadata = await request_json(context, 'metadata.json');
                let config = await request_json(context, 'config.json');
                const name = config ? 'Keras' : 'Keras Weights';
                const format = name + (metadata && metadata.keras_version ? ` v${metadata.keras_version}` : '');
                if (config) {
                    _load_state(config, weights_store);
                } else {
                    config = _create_config(weights_store);
                }
                return await open_model(format, '', '', config, null);
            }
            case 'keras.model.weights.npz': {
                const entries = context.value;
                const weights_store = read_weights_numpy(entries);
                const metadata = await request_json(context, 'metadata.json');
                let config = await request_json(context, 'config.json');
                const name = config ? 'Keras' : 'Keras Weights';
                const format = name + (metadata && metadata.keras_version ? ` v${metadata.keras_version}` : '');
                if (config) {
                    _load_state(config, weights_store);
                } else {
                    config = _create_config(weights_store);
                }
                return await open_model(format, '', '', config, null);
            }
            case 'keras.metadata.json': {
                const metadata = context.value;
                let config = await request_json(context, 'config.json');
                const name = config ? 'Keras' : 'Keras Weights';
                const format = name + (metadata.keras_version ? ` v${metadata.keras_version}` : '');
                const weights_store = await request_weights(context);
                if (!config && (!weights_store || weights_store.size === 0)) {
                    throw new keras.Error("'config.json' or 'model.weights.*' not present.");
                }
                if (config) {
                    _load_state(config, weights_store);
                } else {
                    config = _create_config(weights_store);
                }
                return await open_model(format, '', '', config, null);
            }
            case 'hdf5.parameter.h5':
            case 'keras.h5': {
                const find_root_group = (root_group) => {
                    const kerasmodel = root_group.group('model/kerasmodel');
                    if (kerasmodel && kerasmodel.attributes.has('model_config')) {
                        return kerasmodel;
                    }
                    return root_group;
                };
                const read_model_config = (group) => {
                    if (group.attributes.has('model_config')) {
                        const buffer = group.attributes.get('model_config');
                        const reader = json.TextReader.open(buffer);
                        if (reader) {
                            return reader.read();
                        }
                    }
                    return null;
                };
                const load_attributes_from_hdf5_group = (group, name) => {
                    if (group.attributes.has(name)) {
                        return group.attributes.get(name);
                    }
                    if (group.attributes.has(`${name}0`)) {
                        let index = 0;
                        let value = [];
                        while (group.attributes.has(name + index.toString())) {
                            const chunk = group.attributes.get(name + index.toString());
                            value = value.concat(chunk);
                            index++;
                        }
                        return value;
                    }
                    return null;
                };
                const weights = new keras.Weights();
                const group = context.value;
                const root_group = find_root_group(group);
                const model_config = read_model_config(root_group);
                if (model_config) {
                    const backend = root_group.attributes.get('backend') || '';
                    const version = root_group.attributes.get('keras_version') || '';
                    const format = `Keras${version ? ` v${version}` : ''}`;
                    const model_weights_group = root_group.group('model_weights');
                    if (model_weights_group) {
                        const layer_names = load_attributes_from_hdf5_group(model_weights_group, 'layer_names');
                        for (const layer_name of layer_names) {
                            const layer_weights = model_weights_group.group(layer_name);
                            if (layer_weights) {
                                const weight_names = load_attributes_from_hdf5_group(layer_weights, 'weight_names');
                                if (Array.isArray(weight_names) && weight_names.length > 0) {
                                    for (const weight_name of weight_names) {
                                        const weight = layer_weights.group(weight_name);
                                        if (weight && weight.value) {
                                            const variable = weight.value;
                                            const tensor = new keras.Tensor(weight_name, variable.shape, variable.type, null, null, variable.littleEndian ? '<' : '>', variable.data);
                                            weights.add(layer_name, tensor);
                                        }
                                    }
                                }
                            }
                        }
                    }
                    if (!model_config) {
                        throw new keras.Error("'model_config' is not present.");
                    }
                    if (!model_config.class_name) {
                        throw new keras.Error("'class_name' is not present.");
                    }
                    return open_model(format, '', backend, model_config, weights);
                }
                const layer_names = load_attributes_from_hdf5_group(root_group, 'layer_names');
                if (layer_names && Array.isArray(layer_names)) {
                    const version = root_group.attributes.get('keras_version') || '';
                    const format = `Keras Weights${version ? ` v${version}` : ''}`;
                    const backend = root_group.attributes.get('backend') || '';
                    for (const layer_name of layer_names) {
                        const layer_weights = root_group.group(layer_name);
                        if (layer_weights) {
                            const weight_names = load_attributes_from_hdf5_group(layer_weights, 'weight_names');
                            if (Array.isArray(weight_names) && weight_names.length > 0) {
                                for (const weight_name of weight_names) {
                                    const weight = layer_weights.group(weight_name);
                                    if (weight && weight.value) {
                                        const variable = weight.value;
                                        const components = weight_name.split('/');
                                        components.pop();
                                        const name = (components.length === 0 || components[0] !== layer_name) ? [layer_name].concat(components).join('/') : components.join('/');
                                        const encoding = variable.littleEndian ? '<' : '>';
                                        const tensor = new keras.Tensor(weight_name, variable.shape, variable.type, null, null, encoding, variable.data);
                                        weights.add(name, tensor);
                                    }
                                }
                            }
                        }
                    }
                    return open_model(format, '', backend, null, weights);
                }
                const rootKeys = new Set(root_group.attributes.keys());
                rootKeys.delete('nb_layers');
                if (rootKeys.size > 0 || root_group.value !== null) {
                    throw new keras.Error('File format is not HDF5 Weights.');
                }
                const format = 'HDF5 Weights';
                let weights_group = root_group;
                if (root_group.attributes.size === 0 && root_group.value === null && root_group.groups.size === 1) {
                    const group = root_group.groups.values().next().value;
                    if (group.attributes.size === 0 && group.value === null) {
                        weights_group = group;
                    }
                }
                const tensorKeys = new Set(['name', 'shape', 'quantization']);
                const groups = Array.from(weights_group.groups.values());
                if (groups.every((group) => group.attributes.size === 0 && group.groups.length === 0 && group.value !== null)) {
                    for (const group of groups) {
                        const variable = group.value;
                        const layout = variable.littleEndian ? '<' : '>';
                        const tensor = new keras.Tensor(group.name, variable.shape, variable.type, null, null, layout, variable.type === 'string' ? variable.value : variable.data);
                        weights.add('', tensor);
                    }
                    return open_model(format, '', '', null, weights);
                }
                if (groups.every((group) => group.value === null && Array.from(group.attributes.keys()).filter((key) => !tensorKeys.has(key)).length === 0 && Array.from(group.groups.values()).every((variable) => Object.keys(variable.attributes).length === 0 && variable.value !== null))) {
                    for (const group of groups) {
                        const module = group.attributes.has('name') ? group.attributes.get('name') : group.name;
                        for (const variableGroup of group.groups.values()) {
                            if (variableGroup.attributes.size !== 0 || variableGroup.groups.size !== 0) {
                                throw new keras.Error('Variable format is not HDF5 Weights.');
                            }
                            const variable = variableGroup.value;
                            if (!variable) {
                                throw new keras.Error('Variable value is not HDF5 Weights.');
                            }
                            const name = module ? [module, variableGroup.name].join('/') : variableGroup.name;
                            const layout = variable.littleEndian ? '<' : '>';
                            const tensor = new keras.Tensor(name, variable.shape, variable.type, null, null, layout, variable.type === 'string' ? variable.value : variable.data);
                            weights.add(module, tensor);
                        }
                    }
                    return open_model(format, '', '', null, weights);
                }
                const walk = function(group) {
                    if (group.attributes.size === 0 && group.value === null && group.groups.size > 0) {
                        for (const subGroup of group.groups.values()) {
                            walk(subGroup);
                        }
                        return;
                    }
                    const subKeys = new Set(['index', 'need_grad']);
                    const attribtues = Array.from(group.attributes.keys());
                    const match = attribtues.filter((key) => !subKeys.has(key)).length === 0;
                    if (match && group.value !== null && group.groups.size === 0) {
                        const variable = group.value;
                        const variableName = group.path;
                        let moduleName = variableName;
                        const parts = variableName.split('/');
                        if (parts.length > 1) {
                            parts.pop();
                            moduleName = parts.join('/');
                        }
                        const layout = variable.littleEndian ? '<' : '>';
                        const tensor = new keras.Tensor(variableName, variable.shape, variable.type, null, null, layout, variable.type === 'string' ? variable.value : variable.data);
                        weights.add(moduleName, tensor);
                        return;
                    }
                    throw new keras.Error('Module group format is not HDF5 Weights.');
                };
                walk(weights_group);
                return open_model(format, '', '', null, weights);
            }
            case 'tfjs': {
                const target = context.value;
                await target.read();
                return open_model(target.format, target.producer, target.backend, target.config, target.weights);
            }
            case 'keras.pickle': {
                const obj = context.value;
                const execution = new python.Execution();
                const decoder = new TextDecoder('utf-8');
                const format = `Keras Pickle${obj.keras_version ? ` v${decoder.decode(obj.keras_version)}` : ''}`;
                const backend = obj.backend ? decoder.decode(obj.backend) : '';
                const reader = json.TextReader.open(obj.model_config);
                const model_config = reader.read();
                const weights = new keras.Weights();
                const model_weights_group = obj.model_weights;
                if (model_weights_group) {
                    const layer_names = model_weights_group.layer_names.map((buffer) => decoder.decode(buffer));
                    for (const layer_name of layer_names) {
                        const layer_weights = model_weights_group[layer_name];
                        if (layer_weights) {
                            const weight_names = layer_weights.weight_names.map((buffer) => decoder.decode(buffer));
                            if (Array.isArray(weight_names) && weight_names.length > 0) {
                                for (const weight_name of weight_names) {
                                    const buffer = layer_weights[weight_name];
                                    const pickle = execution.__import__('pickle');
                                    const unpickler = new pickle.Unpickler(buffer);
                                    const variable = unpickler.load();
                                    const tensor = new keras.Tensor(weight_name, variable.shape, variable.dtype.__name__, null, null, '<', variable.data);
                                    weights.add(layer_name, tensor);
                                }
                            }
                        }
                    }
                }
                return open_model(format, '', backend, model_config, weights);
            }
            case 'keras.pb.SavedMetadata': {
                keras.proto = await context.require('./keras-proto');
                const format = 'Keras Saved Metadata';
                const reader = await context.read('protobuf.binary');
                const saved_metadata = keras.proto.third_party.tensorflow.python.keras.protobuf.SavedMetadata.decode(reader);
                if (!saved_metadata || !Array.isArray(saved_metadata.nodes) ||
                    !saved_metadata.nodes.every((node) => node && typeof node.metadata === 'string' && node.metadata.length > 0)) {
                    throw new keras.Error('Invalid keras.protobuf.SavedMetadata.');
                }
                const objects = new Map();
                for (const node of saved_metadata.nodes) {
                    const reader = json.TextReader.open(node.metadata);
                    node.metadata = reader.read();
                    objects.set(node.node_path, node);
                }
                const model_config = objects.get('root').metadata;
                return open_model(format, '', '', model_config, null);
            }
            default: {
                throw new keras.Error(`Unsupported Keras format '${context.type}'.`);
            }
        }
    }
};

keras.Model = class {

    constructor(metadata, format, producer, backend, config, weights) {
        this.format = format;
        this.runtime = backend;
        this.producer = producer;
        metadata = new keras.GraphMetadata(metadata);
        this.modules = [new keras.Graph(metadata, config, weights)];
    }
};

keras.Graph = class {

    constructor(metadata, config, weights, group) {
        this.inputs = [];
        this.outputs = [];
        this.nodes = [];
        group = group || '';
        const values = new Map();
        values.map = (name, type, tensor) => {
            if (tensor) {
                return new keras.Value(name, type || null, tensor);
            }
            if (!values.has(name)) {
                values.set(name, new keras.Value(name, type || null, tensor || null));
            } else if (type || tensor) {
                throw new keras.Error(`Duplicate value '${name}'.`);
            }
            return values.get(name);
        };
        if (config) {
            const getInputType = (layer) => {
                if (layer && layer.config) {
                    let dataType = '?';
                    let shape = [];
                    const config = layer.config;
                    if (config.dtype) {
                        dataType = config.dtype;
                        delete config.dtype;
                    }
                    if (Array.isArray(config.batch_input_shape)) {
                        shape = config.batch_input_shape.map((s) => s === null ? '?' : s);
                        delete config.batch_input_shape;
                    } else if (config.batch_input_shape &&
                        config.batch_input_shape.class_name === '__tuple__' &&
                        Array.isArray(config.batch_input_shape.items)) {
                        shape = config.batch_input_shape.items.map((s) => s === null ? '?' : s);
                        delete config.batch_input_shape;
                    }
                    return new keras.TensorType(dataType, new keras.TensorShape(shape));
                }
                return null;
            };
            this.name = config.name || (config.config && config.config.name ? config.config.name : '');
            this.description = config.class_name;
            let baseType = config.class_name;
            switch (baseType) {
                case '__Function__':
                    this.type = 'function';
                    break;
                case 'Sequential':
                case 'Functional':
                case 'Model': {
                    break;
                }
                case 'Tokenizer': {
                    config = { config: { layers: [config] } };
                    baseType = 'Functional';
                    break;
                }
                default: {
                    const layers = Array.from(config.layers ? config.layers : config);
                    const sequential = layers.every((layer) => layer.inbound_nodes === undefined);
                    baseType = sequential ? 'Sequential' : 'Functional';
                    break;
                }
            }
            switch (baseType) {
                case 'Sequential': {
                    config = config.config;
                    const outputs = null;
                    let name = 'input';
                    let index = -1;
                    const layers = Array.from(config.layers ? config.layers : config);
                    while (layers.length > 0) {
                        const layer = layers.shift();
                        let current = index.toString();
                        index++;
                        if (index === 0) {
                            const type = getInputType(layer);
                            let remove = false;
                            if (layer.class_name === 'InputLayer' && layer.config && layer.config.name) {
                                name = layer.config.name;
                                remove = true;
                            }
                            const value = values.map(name, type);
                            const argument = new keras.Argument(name, [value]);
                            this.inputs.push(argument);
                            if (remove) {
                                continue;
                            }
                        }
                        const nodeInputs = [{ name }];
                        if (layer.config && layer.config.name) {
                            current = layer.config.name;
                        }
                        name = current;
                        let nodeOutputs = [name];
                        if (index === layers.length) {
                            if (outputs && outputs.length > 0) {
                                nodeOutputs = [outputs[0]];
                                name = null;
                            }
                        }
                        layer.inputs = nodeInputs;
                        layer.outputs = nodeOutputs;
                        const node = new keras.Node(metadata, layer, group, weights, values);
                        this.nodes.push(node);
                    }
                    if (name) {
                        const value = values.map(name);
                        const argument = new keras.Argument(name, [value]);
                        this.outputs.push(argument);
                    }
                    break;
                }
                case '__Function__':
                case 'Functional':
                case 'Model': {
                    config = config.config;
                    const nodes = new Map();
                    if (config.layers) {
                        const is_constant = (item) => {
                            return Array.isArray(item) && (item.length === 3 || item.length === 4) && item[0] === '_CONSTANT_VALUE' && item[1] === -1;
                        };
                        const is_connection = (item) => {
                            return Array.isArray(item) && (item.length === 3 || item.length === 4) && typeof item[0] === 'string' && typeof item[1] === 'number'  && typeof item[2] === 'number';
                        };
                        const read_value = (input_data) => {
                            if (!Array.isArray(input_data)) {
                                return input_data;
                            }
                            const transform = (value) => {
                                if (value.every((item) => is_constant(item))) {
                                    for (let i = 0; i < value.length; i++) {
                                        value[i] = value[i][2];
                                    }
                                } else if (value.every((item) => Array.isArray(item))) {
                                    const dims = value.map((item) => transform(item));
                                    const [dim] = dims;
                                    for (let i = 1; i < dims.length; i++) {
                                        if (dim.length === dims[i].length) {
                                            if (!dims[i].every((value, i) => value === dim[i])) {
                                                throw new python.Error('Invalid array shape.');
                                            }
                                        }
                                    }
                                    return [value.length].concat(dim);
                                }
                                return [value.length];
                            };
                            const shape = transform(input_data);
                            const flatten = (input) => input.reduce((a, b) => a.concat(Array.isArray(b) ? flatten(b) : b), []);
                            const value = flatten(input_data);
                            return { shape, value };
                        };
                        const functional = config.layers.every((layer) => Array.isArray(layer.inbound_nodes));
                        if (functional) {
                            const read_connection = (input_data) => {
                                const [node_name, node_index, tensor_index] = input_data;
                                const inbound_node_key = `${node_name}[${node_index}]`;
                                const inbound_node = nodes.get(inbound_node_key);
                                const tensor_key = `${node_name}[${node_index}][${tensor_index}]`;
                                if (inbound_node) {
                                    while (tensor_index >= inbound_node.outputs.length) {
                                        inbound_node.outputs.push(undefined);
                                    }
                                    inbound_node.outputs[tensor_index] = tensor_key;
                                }
                                return tensor_key;
                            };
                            const process_node = (node, inbound_node) => {
                                if (Array.isArray(inbound_node) && inbound_node.length === 4 && typeof inbound_node[0] === 'string') {
                                    const key = read_connection(inbound_node);
                                    node.inputs.push({ name: key });
                                    for (const [name, value] of Object.entries(inbound_node[3])) {
                                        if (is_connection(value)) {
                                            const key = read_connection(value);
                                            node.inputs.push({ name: key });
                                        } else if (Array.isArray(value)) {
                                            const array = read_value(value);
                                            node.args[name] = array;
                                        } else {
                                            node.args[name] = value;
                                        }
                                    }
                                } else if (Array.isArray(inbound_node)) {
                                    for (const input_data of inbound_node) {
                                        // [ 'conv2d', 0, 0 ] or [ 'conv2d', 0, 0, {} ]
                                        if (Array.isArray(input_data) && is_connection(input_data)) {
                                            const key = read_connection(input_data);
                                            node.inputs.push({ name: key });
                                        } else if (Array.isArray(input_data) && input_data.every((item) => is_connection(item))) {
                                            for (const input of input_data) {
                                                const key = read_connection(input);
                                                node.inputs.push({ name: key });
                                            }
                                        } else if (Array.isArray(input_data)) {
                                            const value = read_value(input_data);
                                            node.inputs.push(value);
                                        } else {
                                            throw new keras.Error(`Invalid inbound connection '${JSON.stringify(input_data)}'.`);
                                        }
                                    }
                                } else if (inbound_node && inbound_node.args) {
                                    for (const arg of inbound_node.args) {
                                        if (arg && arg.class_name === '__keras_tensor__' && arg.config && is_connection(arg.config.keras_history)) {
                                            const key = read_connection(arg.config.keras_history);
                                            node.inputs.push({ name: key });
                                        } else if (Array.isArray(arg) && arg.every((arg) => arg && arg.class_name === '__keras_tensor__' && arg.config && is_connection(arg.config.keras_history))) {
                                            for (const input of arg) {
                                                const key = read_connection(input.config.keras_history);
                                                node.inputs.push({ name: key });
                                            }
                                        }
                                    }
                                }
                            };
                            let legacy_format = true;
                            for (const layer of config.layers) {
                                if (Array.isArray(layer.inbound_nodes)) {
                                    for (const inbound_node of layer.inbound_nodes) {
                                        if (Array.isArray(inbound_node.args)) {
                                            legacy_format = false;
                                        }
                                    }
                                }
                            }
                            for (const layer of config.layers) {
                                const class_name = layer.class_name;
                                let first_index = 0;
                                if (legacy_format) {
                                    const keys = new Set(Object.keys(layer.config));
                                    const is_functional_config = keys.has('name') && keys.has('layers') && keys.has('input_layers') && keys.has('output_layers');
                                    if (class_name === 'Sequential' ||
                                        (is_functional_config && Array.isArray(layer.config.layers) && layer.config.layers.length > 0 && layer.config.layers[0].class_name === 'InputLayer')) {
                                        first_index++;
                                    }
                                }
                                if (Array.isArray(layer.inbound_nodes) && layer.inbound_nodes.length === 0) {
                                    layer.inputs = [];
                                    layer.outputs = [];
                                    layer.args = {};
                                    nodes.set(`${layer.name}[${first_index}]`, layer);
                                } else if (Array.isArray(layer.inbound_nodes) && layer.inbound_nodes.length === 1) {
                                    layer.inputs = [];
                                    layer.outputs = [];
                                    layer.args = {};
                                    [layer.inbound_node] = layer.inbound_nodes;
                                    nodes.set(`${layer.name}[${first_index}]`, layer);
                                } else {
                                    let config = {};
                                    switch (class_name) {
                                        case 'Functional':
                                        case 'Sequential':
                                        case 'Model': {
                                            config = layer;
                                            break;
                                        }
                                        default: {
                                            config.class_name = '__Function__';
                                            config.name = layer.name;
                                            config.config = {};
                                            config.config.layers = [{ ...layer }];
                                            delete config.config.layers[0].inbound_nodes;
                                            delete config.config.layers[0].input_layers;
                                            delete config.config.layers[0].output_layers;
                                            break;
                                        }
                                    }
                                    const type = new keras.Graph(metadata, config, weights, '');
                                    for (let i = 0; i < layer.inbound_nodes.length; i++) {
                                        const index = i + first_index;
                                        const key = `${layer.name}[${index}]`;
                                        const node = {};
                                        node.name = key;
                                        node.class_name = '__Function__';
                                        node.config = {};
                                        node.config.name = key;
                                        node.inputs = [];
                                        node.outputs = [];
                                        node.args = {};
                                        node.__type__ = type;
                                        node.inbound_node = layer.inbound_nodes[i];
                                        nodes.set(key, node);
                                    }
                                }
                            }
                            for (const entry of nodes) {
                                if (entry[1].inbound_node) {
                                    process_node(entry[1], entry[1].inbound_node);
                                }
                            }
                            if (Array.isArray(config.input_layers)) {
                                if (config.input_layers.length === 3 && typeof config.input_layers[0] === 'string' && Number.isInteger(config.input_layers[1]) && Number.isInteger(config.input_layers[2])) {
                                    config.input_layers = [config.input_layers];
                                }
                                for (let i = 0; i < config.input_layers.length; i++) {
                                    const input_data = config.input_layers[i];
                                    const name = read_connection(input_data);
                                    const [node_name, node_index] = input_data;
                                    const inbound_node_key = `${node_name}[${node_index}]`;
                                    const node = nodes.get(inbound_node_key);
                                    let type = null;
                                    if (node && node.class_name === 'InputLayer') {
                                        type = getInputType(node);
                                        nodes.delete(name);
                                        nodes.delete(inbound_node_key);
                                    }
                                    const value = values.map(name, type);
                                    const argument = new keras.Argument(node_name, [value]);
                                    this.inputs.push(argument);
                                }
                            }
                            if (Array.isArray(config.output_layers)) {
                                if (config.output_layers.length === 3 && typeof config.output_layers[0] === 'string' && Number.isInteger(config.output_layers[1]) && Number.isInteger(config.output_layers[2])) {
                                    config.output_layers = [config.output_layers];
                                }
                                for (let i = 0; i < config.output_layers.length; i++) {
                                    const output_data = config.output_layers[i];
                                    const [name] = output_data;
                                    const key = read_connection(output_data);
                                    const value = values.map(key);
                                    const argument = new keras.Argument(name, [value]);
                                    this.outputs.push(argument);
                                }
                            }
                        } else {
                            for (const layer of config.layers) {
                                layer.inputs = [];
                                layer.outputs = [];
                                layer.args = {};
                                nodes.set(`${layer.name}[0]`, layer);
                            }
                        }
                    }
                    for (const entry of nodes) {
                        const node = new keras.Node(metadata, entry[1], group, weights, values);
                        this.nodes.push(node);
                    }
                    break;
                }
                default: {
                    throw new keras.Error(`'${config.class_name}' is not supported.`);
                }
            }
        } else if (weights) {
            this.type = 'weights';
            for (const name of weights.keys()) {
                if (weights.get('', name).length <= 6) {
                    const layer = { class_name: 'Weights', config: { name } };
                    const node = new keras.Node(metadata, layer, '', weights, values);
                    this.nodes.push(node);
                }
            }
        }
    }
};

keras.Argument = class {

    constructor(name, value, type = null, visible = true) {
        this.name = name;
        this.value = value;
        this.type = type;
        this.visible = visible;
    }
};

keras.Value = class {

    constructor(name, type, initializer = null) {
        if (typeof name !== 'string') {
            throw new keras.Error(`Invalid value identifier '${JSON.stringify(name)}'.`);
        }
        this.name = name;
        this.type = !type && initializer ? initializer.type  : type;
        this.quantization = initializer && initializer.quantization ? initializer.quantization : null;
        this.initializer = initializer;
    }
};

keras.Node = class {

    constructor(metadata, layer, group, weights, values) {
        const config = layer.config || {};
        const args = layer.args || {};
        let inputs = layer.inputs || [];
        let outputs = layer.outputs || [];
        const name = config && config.name ? config.name : '';
        this.group = group || '';
        this.name = (this.group ? `${this.group}/` : '') + name;
        this.inputs = [];
        this.outputs = [];
        this.attributes = [];
        this.chain = [];
        let names = [name];
        let class_name = layer.class_name;
        let model = false;
        switch (class_name) {
            case '__Function__': {
                this.type = layer.__type__;
                model = true;
                break;
            }
            case 'Model':
            case 'Functional':
            case 'Sequential': {
                const name = layer.name || (layer.config ? layer.config.name : '');
                this.type = new keras.Graph(metadata, layer, weights, (group ? `${group}/` : '') + name);
                model = true;
                if (config) {
                    delete config.layers;
                    delete config.input_layers;
                    delete config.output_layers;
                }
                this.inputs = [new keras.Argument('inputs', inputs.map((input) => values.map(input.name)))];
                this.outputs = [new keras.Argument('outputs', outputs.map((name) => values.map(name)))];
                inputs = [];
                outputs = [];
                break;
            }
            case 'Wrapper':
            case 'Bidirectional':
            case 'TimeDistributed': {
                if (config && config.layer) {
                    const inner = config.layer;
                    delete config.layer;
                    this.inner = new keras.Node(metadata, inner, null, null, values);
                    if (class_name === 'Bidirectional' && inner.config.name) {
                        names = [`${name}/forward_${inner.config.name}`, `${name}/backward_${inner.config.name}`];
                        if (!group) {
                            group = name;
                        }
                    }
                }
                this.type = metadata.type(class_name) || { name: class_name };
                break;
            }
            case 'TFOpLambda': {
                if (config && config.function) {
                    class_name = config.function;
                    delete config.function;
                }
                this.type = metadata.type(class_name) || { name: class_name };
                break;
            }
            default: {
                this.type = metadata.type(class_name) || { name: class_name };
                break;
            }
        }

        if (layer._trainable_variables) {
            if (inputs.length === 0 && Array.isArray(this.type.inputs) && this.type.inputs.length > 0) {
                // weights-only, remove 'input' from type metadata
                this.type = { ...this.type };
                this.type.inputs = this.type.inputs.slice(1);
            }
            for (const variable of layer._trainable_variables) {
                inputs.push({ name: '', initializer: variable });
            }
        } else if (weights && !model) {
            for (const name of names) {
                let tensors = weights.get(group, name);
                if (tensors.length > 0) {
                    for (const initializer of tensors) {
                        inputs.push({ name: initializer.name, initializer });
                    }
                } else {
                    tensors = weights.get('', name);
                    for (const initializer of tensors) {
                        inputs.push({ name: initializer.name, initializer });
                    }
                }
            }
        }

        const attributes = [];

        const convertAttributeValue = (value) => {
            if (Array.isArray(value) || value !== Object(value)) {
                return value;
            }
            const obj = {};
            if (value.class_name) {
                obj.__type__ = value.class_name;
            }
            if (value.config) {
                const config = value.config;
                for (const [key, value] of Object.entries(config)) {
                    obj[key] = convertAttributeValue(value);
                }
            }
            return obj;
        };

        if (config && !Array.isArray(config)) {
            for (const [name, value] of Object.entries(config)) {
                if (class_name !== 'Activation' && name === 'activation' && value !== 'linear') {
                    if (typeof value === 'string') {
                        const config = { activation: value };
                        const node = new keras.Node(metadata, { class_name: 'Activation', config }, null, null, value);
                        this.chain.push(node);
                    } else if (value && typeof value.class_name === 'string' && value.config) {
                        const type = value.class_name;
                        if (!metadata.type(type)) {
                            metadata.add(type, { name: type, category: 'Activation' });
                        }
                        const node = new keras.Node(metadata, value, null, null, value);
                        this.chain.push(node);
                    }
                }
                if (name !== 'name' && name !== 'batch_input_shape') {
                    const schema = metadata.attribute(class_name, name);
                    attributes.push([schema, name, value]);
                }
            }
        }

        const innerType = this.inner ? this.inner.type : null;
        const innerMetadata = innerType ? metadata.type(innerType) : null;
        let inputIndex = 0;
        while (inputs.length > 0) {
            let list = false;
            let name = null;
            let visible = true;
            if (!innerMetadata || inputIndex === 0) {
                if (this.type && this.type.inputs && inputIndex < this.type.inputs.length) {
                    const input = this.type.inputs[inputIndex];
                    name = input.name;
                    if (class_name === 'BatchNormalization' && name === 'gamma' && config.scale === false) {
                        inputIndex++;
                        continue;
                    }
                    visible = input.visible !== false;
                    if (this.type.inputs[inputIndex].list) {
                        list = true;
                    }
                }
            } else {
                switch (class_name) {
                    case 'Bidirectional': {
                        let innerIndex = inputIndex;
                        if (innerMetadata && innerMetadata.inputs) {
                            if (innerIndex < innerMetadata.inputs.length) {
                                name = `forward_${innerMetadata.inputs[innerIndex].name}`;
                            } else {
                                innerIndex = innerIndex - innerMetadata.inputs.length + 1;
                                if (innerIndex < innerMetadata.inputs.length) {
                                    name = `backward_${innerMetadata.inputs[innerIndex].name}`;
                                }
                            }
                        }
                        visible = false;
                        break;
                    }
                    case 'TimeDistributed':
                        if (innerMetadata && innerMetadata.inputs && inputIndex < innerMetadata.inputs.length) {
                            name = innerMetadata.inputs[inputIndex].name;
                        }
                        break;
                    default:
                        break;
                }
            }
            const input = list ? inputs.splice(0, inputs.length) : [inputs.shift()];
            const inputArguments = input.map((input) => {
                if (input.name) {
                    return values.map(input.name, null, input.initializer);
                }
                if (input.initializer) {
                    return values.map(input.name, null, input.initializer);
                }
                if (input.value !== undefined) {
                    const tensor = new keras.Tensor('', input.shape, config.dtype || '?', null, null, '|', input.value);
                    return values.map('', null, tensor);
                }
                throw new keras.Error(`Invalid argument '${JSON.stringify(input.name)}'.`);
            });
            if (!name && inputArguments.length === 1 && inputArguments[0].initializer && inputArguments[0].initializer.name) {
                if (names.length === 1 && names[0] === '') {
                    name = inputArguments[0].initializer.name;
                } else {
                    const parts = inputArguments[0].initializer.name.split('/').pop().split(':').shift().split('_');
                    const inputName1 = parts.pop();
                    const inputName2 = parts.length > 0 ? [parts.pop(), inputName1].join('_') : '';
                    const inputNames = new Set(['recurrent_kernel', 'running_mean', 'running_std', 'moving_mean', 'moving_variance', 'depthwise_filter', 'pointwise_filter']);
                    name = inputNames.has(inputName2) ? inputName2 : inputName1;
                }
            }
            const argument = new keras.Argument(name || inputIndex.toString(), inputArguments, null, visible);
            this.inputs.push(argument);
            inputIndex++;
        }

        for (let i = 0; i < outputs.length; i++) {
            const output = outputs[i];
            const name = this.type && this.type.outputs && i < this.type.outputs.length && this.type.outputs[i] && this.type.outputs[i].name ? this.type.outputs[i].name : i.toString();
            const argument = new keras.Argument(name, output === undefined || output.length === 0 ? [] : [values.map(output)]);
            this.outputs.push(argument);
        }

        const inputTypes = new Map((this.type.inputs || []).map((input) => [input.name, input.type]));
        for (const [name, arg] of Object.entries(args)) {
            if (name !== 'name') {
                if ((arg && arg.name) || (inputTypes.has(name) && inputTypes.get(name) === 'Tensor' && arg)) {
                    if (arg.name) {
                        const value = values.map(arg.name);
                        const argument = new keras.Argument(name, [value]);
                        this.inputs.push(argument);
                    } else {
                        const tensor = new keras.Tensor('', arg.shape, config.dtype || '?', null, null, '|', arg.value);
                        const value = values.map('', null, tensor);
                        const argument = new keras.Argument(name, [value]);
                        this.inputs.push(argument);
                    }
                } else {
                    const schema = metadata.attribute(class_name, name);
                    this.attributes.push([schema, name, arg]);
                }
            }
        }

        this.attributes = attributes.map(([metadata, name, value]) => {
            let type = null;
            let visible = true;
            if (value && typeof value === 'object' && value.class_name && value.config) {
                value = convertAttributeValue(value);
            }
            switch (name) {
                case 'trainable':
                    type = 'boolean';
                    visible = false;
                    break;
                case 'dtype':
                    visible = false;
                    break;
                default: {
                    if (metadata) {
                        type = metadata.type ? metadata.type : type;
                        if (metadata.visible === false) {
                            visible = false;
                        } else if (metadata.default !== undefined) {
                            if (Array.isArray(value)) {
                                if (Array.isArray(metadata.default)) {
                                    visible = value.length !== metadata.default || !value.every((item, index) => item === metadata.default[index]);
                                } else {
                                    visible = !value.every((item) => item === metadata.default);
                                }
                            } else {
                                visible = value !== metadata.default;
                            }
                        }
                    }
                    break;
                }
            }
            return new keras.Argument(name, value, type, visible);
        });

        if (typeof this.type.name !== 'string' || !this.type.name.split) { // #416
            throw new keras.Error(`Unsupported node type '${JSON.stringify(this.type.name)}'.`);
        }
    }
};

keras.Tensor = class {

    constructor(name, shape, type, stride, quantization, encoding, data, location) {
        this.name = name;
        this.type = new keras.TensorType(type, new keras.TensorShape(shape));
        this.stride = stride;
        this.encoding = encoding;
        this._data = data;
        this.location = location;
        if (quantization && (quantization.scale !== 0 || quantization.min !== 0)) {
            this.quantization = {
                type: 'linear',
                scale: [quantization.scale],
                min: [quantization.min]
            };
        }
    }

    get values() {
        if (this.encoding === '|') {
            return this._data;
        }
        if (this._data === null) {
            return null;
        }
        return this._data instanceof Uint8Array ? this._data : this._data.peek();
    }
};

keras.TensorType = class {

    constructor(dataType, shape) {
        this.dataType = dataType;
        this.shape = shape;
    }

    toString() {
        return this.dataType + this.shape.toString();
    }
};

keras.TensorShape = class {

    constructor(dimensions) {
        this.dimensions = dimensions;
    }

    toString() {
        return this.dimensions && this.dimensions.length > 0 ? (`[${this.dimensions.map((dimension) => dimension.toString()).join(',')}]`) : '';
    }
};

keras.GraphMetadata = class {

    constructor(metadata) {
        this._metadata = metadata;
        this._types = new Map();
    }

    type(name) {
        if (this._types.has(name)) {
            return this._types.get(name);
        }
        return this._metadata.type(name);
    }

    attribute(type, name) {
        return this._metadata.attribute(type, name);
    }

    add(type, metadata) {
        this._types.set(type, metadata);
    }
};

keras.Weights = class {

    constructor() {
        this._map = new Map();
    }

    get empty() {
        return this._map.size === 0;
    }

    add(layer_name, tensor) {
        if (!this._map.has(layer_name)) {
            this._map.set(layer_name, []);
        }
        this._map.get(layer_name).push(tensor);
    }

    get(group, name) {
        if (group) {
            const list = this._map.get(group.split('/').shift());
            if (list) {
                const match1 = list.filter((tensor) => tensor.name.startsWith(`${name}/`));
                if (match1.length > 0) {
                    return match1;
                }
                const match2 = list.filter((tensor) => tensor.name.startsWith(`${group}/${name}/`));
                if (match2.length > 0) {
                    return match2;
                }
            }
        } else {
            const match1 = this._map.get(name);
            if (match1 && match1.length > 0) {
                return match1;
            }
            const match2 = this._map.get('');
            if (match2 && match2.length > 0) {
                const match3 = match2.filter((tensor) => tensor.name.startsWith(`${(group ? `${group}/` : '') + name}/`));
                if (match3.length > 0) {
                    return match3;
                }
            }
        }
        return [];
    }

    keys() {
        return this._map.keys();
    }
};

keras.Error = class extends Error {

    constructor(message) {
        super(message);
        this.name = 'Error loading Keras model.';
    }
};

tfjs.Container = class {

    static async open(context) {
        const json = await context.peek('json');
        if (json) {
            if (json.modelTopology && (json.format === 'layers-model' || json.modelTopology.class_name || json.modelTopology.model_config)) {
                return new tfjs.Container(context, '');
            }
            if (Array.isArray(json) && json.every((item) => item.weights && item.paths)) {
                return new tfjs.Container(context, 'weights.json');
            }
            if (json.tfjsVersion) {
                return new tfjs.Container(context, 'metadata');
            }
        }
        const identifier = context.identifier;
        if (/^.*group\d+-shard\d+of\d+(\.bin)?$/.test(identifier)) {
            return new tfjs.Container(context, 'weights.bin');
        }
        return null;
    }

    constructor(context, type) {
        this.context = context;
        this.type = type;
    }

    async read() {
        const context = this.context;
        switch (this.type) {
            case '': {
                const obj = await context.peek('json');
                return this._openModelJson(obj);
            }
            case 'weights.json': {
                this.format = 'TensorFlow.js Weights';
                this.config = null;
                const obj = await context.peek('json');
                const manifests = Array.from(obj);
                for (const manifest of manifests) {
                    for (const weight of manifest.weights) {
                        const name = weight.name;
                        const index = name.lastIndexOf('/');
                        weight.identifier = index === -1 ? name : name.substring(0, index);
                    }
                }
                return this._openManifests(manifests);
            }
            case 'weights.bin': {
                const content = await this.context.fetch('model.json');
                const obj = await content.read('json');
                return this._openModelJson(obj);
            }
            case 'metadata': {
                const content = await this.context.fetch('model.json');
                const obj = await content.read('json');
                return this._openModelJson(obj);
            }
            default: {
                throw new tfjs.Error(`Unsupported TensorFlow.js format '${this.type}'.`);
            }
        }
    }

    _openShards(manifests, shards) {
        this.weights = new keras.Weights();
        const dtype_size_map = new Map([
            ['float16', 2], ['float32', 4], ['float64', 8],
            ['int8', 1], ['int16', 2], ['int32', 4], ['int64', 8],
            ['uint8', 1], ['uint16', 2], ['uint32', 4], ['uint64', 8]
        ]);
        for (const manifest of manifests) {
            let buffer = null;
            let location = '';
            if (Array.isArray(manifest.paths) && manifest.paths.length > 0 && manifest.paths.every((path) => shards.has(path))) {
                const list = manifest.paths.map((path) => shards.get(path));
                location = manifest.paths.join(', ');
                const size = list.reduce((a, b) => a + b.length, 0);
                buffer = new Uint8Array(size);
                let offset = 0;
                for (const item of list) {
                    buffer.set(item, offset);
                    offset += item.length;
                }
            }
            let offset = 0;
            for (const weight of manifest.weights) {
                const dtype = weight.quantization && weight.quantization.dtype ? weight.quantization.dtype : weight.dtype;
                if (!dtype_size_map.has(dtype)) {
                    throw new keras.Error(`Unsupported weight data type size '${dtype}'.`);
                }
                const itemsize = dtype_size_map.get(dtype);
                const size = weight.shape.reduce((a, b) => a * b, 1);
                const length = itemsize * size;
                const data = buffer ? buffer.slice(offset, offset + length) : null;
                const tensor = new keras.Tensor(weight.name, weight.shape, dtype, null, weight.quantization, '<', data, location);
                this.weights.add(weight.identifier, tensor);
                offset += length;
            }
        }
    }

    async _openManifests(manifests) {
        const shards = new Map();
        for (const manifest of manifests) {
            for (const path of manifest.paths) {
                if (!shards.has(path)) {
                    const promise = this.context.fetch(path);
                    shards.set(path, promise);
                }
            }
        }
        const promises = shards.values();
        try {
            const contexts = await Promise.all(promises);
            for (const key of shards.keys()) {
                const context = contexts.shift();
                const buffer = context.stream.peek();
                shards.set(key, buffer);
            }
            this._openShards(manifests, shards);
        } catch {
            shards.clear();
            this._openShards(manifests, shards);
        }
    }

    _openModelJson(obj) {
        if (!obj || !obj.modelTopology || (obj.format !== 'layers-model' && !obj.modelTopology.model_config && !obj.modelTopology.config)) {
            throw new tfjs.Error('File format is not TensorFlow.js layers-model.');
        }
        const modelTopology = obj.modelTopology;
        if (obj.format) {
            this.format = `TensorFlow.js ${obj.format}`;
        } else if (modelTopology.keras_version) {
            const match = modelTopology.keras_version.match(/^(.+)\s+(\d.*)$/);
            const version = match ? `${match[1]} v${match[2]}` : `v${modelTopology.keras_version}`;
            this.format = `TensorFlow.js Keras ${version}`;
        } else {
            this.format = 'TensorFlow.js Keras';
        }
        this.producer = obj.convertedBy || obj.generatedBy || '';
        this.backend = modelTopology.backend || '';
        const manifests = obj.weightsManifest;
        for (const manifest of manifests) {
            for (const weight of manifest.weights) {
                weight.identifier = '';
            }
        }
        this.config = modelTopology.model_config ? modelTopology.model_config : modelTopology;
        return this._openManifests(manifests);
    }
};

tfjs.Error = class extends Error {

    constructor(message) {
        super(message);
        this.name = 'Error loading TensorFlow.js model.';
    }
};

export const ModelFactory = keras.ModelFactory;