@tensorflow/tfjs
Version:
An open-source machine learning framework.
144 lines (114 loc) • 4.57 kB
text/typescript
/**
* @license
* Copyright 2020 Google Inc. All Rights Reserved.
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
* =============================================================================
*/
/**
* Entry point for cli tool to build custom tfjs modules
*/
import * as fs from 'fs';
import * as path from 'path';
import * as yargs from 'yargs';
import {OP_SCOPE_SUFFIX} from '@tensorflow/tfjs-core';
import {bail} from './util';
import {CustomTFJSBundleConfig, SupportedBackends, ModuleProvider} from './types';
import {getModuleProvider} from './esm_module_provider';
// Will be configured when loading the config file.
let moduleProvider: ModuleProvider;
const BASE_PATH = process.env.BASE_PATH || process.cwd();
const DEFAULT_CUSTOM_BUNDLE_ARGS: Partial<CustomTFJSBundleConfig> = {
entries: [],
models: [],
kernels: [],
forwardModeOnly: true,
backends: ['cpu', 'webgl'],
moduleOptions: {},
};
const argParser = yargs.options({
config: {
description: 'Path to custom module config file.',
type: 'string',
demandOption: true
}
});
const args = argParser.argv;
function validateArgs(): CustomTFJSBundleConfig {
let configFilePath = args.config;
if (configFilePath == null) {
bail(`Error: no config file passed`);
}
configFilePath = path.resolve(BASE_PATH, configFilePath);
if (!fs.existsSync(configFilePath)) {
bail(`Error: config file does not exist at ${configFilePath}`);
}
let config;
try {
config = JSON.parse(fs.readFileSync(configFilePath, 'utf-8'));
} catch (error) {
bail(`Error could not read/parse JSON config file. \n ${error.message}`);
}
if (config.outputPath == null) {
bail('Error: config must specify "outputPath" property');
}
console.log(`Using custom module configuration from ${configFilePath}.`);
const finalConfig = Object.assign({}, DEFAULT_CUSTOM_BUNDLE_ARGS, config);
if (finalConfig.entries.length !== 0) {
bail('Error: config.entries not yet supported');
}
// if (finalConfig.models.length !== 0) {
// TODO validate that all these paths exist.
// bail('Error: config.models not yet supported');
// }
for (const requestedBackend of finalConfig.backends) {
if (requestedBackend !== SupportedBackends.cpu &&
requestedBackend !== SupportedBackends.webgl &&
requestedBackend !== SupportedBackends.wasm) {
bail(`Error: Unsupported backend specified '${requestedBackend}'`);
}
}
// Normalize the paths to absolute paths.
function normalizePath(p: string) {
return path.resolve(BASE_PATH, p);
}
finalConfig.models = finalConfig.models.map(normalizePath);
finalConfig.entries = finalConfig.entries.map(normalizePath);
finalConfig.normalizedOutputPath = normalizePath(finalConfig.outputPath);
moduleProvider = getModuleProvider(finalConfig.moduleOptions);
console.log('Final Configuration', finalConfig);
return finalConfig;
}
function getKernelNamesForConfig(config: CustomTFJSBundleConfig) {
// Later on this will do a union of kernels from entries, models and
// kernels, (and kernels used by the converter itself) Currently we only
// support directly listing kernels. remember that this also needs to handle
// kernels used by gradients if forwardModeOnly is false.
// Ops in core that are implemented as custom ops may appear in tf.profile
// they will have __op as a suffix. These do not have corresponding backend
// kernels so we need to filter them out.
function isNotCustomOp(kernelName: string) {
// opSuffix value is defined in tfjs-core/src/operation.ts
// duplicating it here to avoid an export.
return !kernelName.endsWith(OP_SCOPE_SUFFIX);
}
return config.kernels.filter(isNotCustomOp);
}
const customConfig = validateArgs();
const kernelsToInclude = getKernelNamesForConfig(customConfig);
customConfig.kernels = kernelsToInclude;
if (moduleProvider != null) {
moduleProvider.produceCustomTFJSModule(customConfig);
} else {
throw new Error('No module provider has been initialized.');
}