main.py

#
# -*- coding: utf-8 -*-
#
# Copyright (c) 2021 Intel Corporation
#
# 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.
#
#

import time
import sys
import numpy as np
import unicodedata
import six
import re
import tensorflow as tf
from absl import app
from argparse import ArgumentParser
import pandas as pd
from utils import tokenizer
from utils.tokenizer import Subtokenizer
from utils import metrics

flags = tf.compat.v1.flags
FLAGS = flags.FLAGS

flags.DEFINE_integer("batch_size", 64,
                     "run batch size")

flags.DEFINE_string("input_graph", None,
                    "The path of input model file.")

flags.DEFINE_string("inputs_file", None,
                    "File saved to an output file.")

flags.DEFINE_string("reference_file", None,
                    "File containing reference translation.")

flags.DEFINE_string("vocab_file", None,
                    "Path to subtoken vocabulary file.")

flags.DEFINE_string("config", None,
                    "Config json file")

flags.DEFINE_string("output_model", None,
                    "The output model of the quantized model.")

flags.DEFINE_string("mode", "tune",
                     "One of three options: 'benchmark'/'accuracy'/'tune'.")

flags.DEFINE_integer("iters", -1,
                     "The iteration used for benchmark.")

class UnicodeRegex(object):
    def __init__(self):
        punctuation = self.property_chars("P")
        self.nondigit_punct_re = re.compile(r"([^\d])([" + punctuation + r"])")
        self.punct_nondigit_re = re.compile(r"([" + punctuation + r"])([^\d])")
        self.symbol_re = re.compile("([" + self.property_chars("S") + "])")

    def property_chars(self, prefix):
        return "".join(six.unichr(x) for x in range(sys.maxunicode)
                    if unicodedata.category(six.unichr(x)).startswith(prefix))

uregex = UnicodeRegex()

def bleu_tokenize(string):
    string = uregex.nondigit_punct_re.sub(r"\1 \2 ", string)
    string = uregex.punct_nondigit_re.sub(r" \1 \2", string)
    string = uregex.symbol_re.sub(r" \1 ", string)
    return string.split()

class bleu(object):
    def __init__(self):
        self.translations = []
        self.labels = []

    def reset(self):
        self.translations = []
        self.labels = []

    def update(self, pred, label):
        if len(label) != len(pred):
            raise ValueError("Reference and translation files have different number "
                             "of lines. If training only a few steps (100-200), the "
                             "translation may be empty.")
        label = [x.lower() for x in label]
        pred = [x.lower() for x in pred]
        label = [bleu_tokenize(x) for x in label]
        pred = [bleu_tokenize(x) for x in pred]
        self.labels.extend(label)
        self.translations.extend(pred)

    def result(self):
        return metrics.compute_bleu(self.labels, self.translations) * 100

def collate_fn(batch):
    """Puts each data field into a pd frame with outer dimension batch size"""
    elem = batch[0]
    if isinstance(elem, tuple):
        batch = zip(*batch)
        return [collate_fn(samples) for samples in batch]
    elif isinstance(elem, np.ndarray):
        return [list(elem) for elem in batch]
    elif isinstance(elem, str):
        return batch
    else:
        return pd.DataFrame(batch).fillna(0).values.astype(np.int32)

def load_graph(file_name):
    tf.compat.v1.logging.info('Loading graph from: ' + file_name)
    with tf.io.gfile.GFile(file_name, "rb") as f:
        graph_def = tf.compat.v1.GraphDef()
        graph_def.ParseFromString(f.read())
    with tf.Graph().as_default() as graph:
        tf.import_graph_def(graph_def, name='')
    return graph

def eval_func(infer_graph, iteration=-1):
    if isinstance(infer_graph, tf.compat.v1.GraphDef):
        graph = tf.Graph() 
        with graph.as_default():
            tf.import_graph_def(infer_graph, name='') 
        infer_graph = graph

    subtokenizer = Subtokenizer(FLAGS.vocab_file)
    input_tensor = infer_graph.get_tensor_by_name('input_tensor:0')
    output_tensor = infer_graph.get_tensor_by_name(\
        'model/Transformer/strided_slice_19:0')
    ds = Dataset(FLAGS.inputs_file, FLAGS.reference_file, FLAGS.vocab_file)
    from neural_compressor.data import DATALOADERS
    dataloader = DATALOADERS['tensorflow'](ds, batch_size=FLAGS.batch_size, 
                                           collate_fn=collate_fn)
    config = tf.compat.v1.ConfigProto()
    config.use_per_session_threads = 1
    config.inter_op_parallelism_threads = 1
    sess = tf.compat.v1.Session(graph=infer_graph, config=config)
    time_list = []
    bleu_eval = bleu()
    predictions = []
    labels = []
    warmup = 10
    if iteration != -1:
        assert iteration >= warmup, 'iteration must be larger than warmup'
    for idx, (input_data, label) in enumerate(dataloader):
        if idx < iteration or iteration == -1:
            time_start = time.time()
            out = sess.run([output_tensor], {input_tensor: input_data})
            duration = time.time() - time_start
            time_list.append(duration)
            predictions.append(out)
            labels.extend(label)
        else:
            break
    latency = np.array(time_list[warmup: ]).mean() / FLAGS.batch_size
    print('Batch size = {}'.format(FLAGS.batch_size))
    print('Latency: {:.3f} ms'.format(latency * 1000))
    print('Throughput: {:.3f} items/sec'.format(1./ latency))
    
    # only calculate accuracy when running out all predictions
    if iteration == -1:
        decode = []
        for i,tr in enumerate(predictions):
            for j,itr in enumerate(tr):
                for k, otr in enumerate(itr):
                    try:
                        index = list(otr).index(tokenizer.EOS_ID)
                        decode.append(subtokenizer.decode(otr[:index]))
                    except:
                        decode.append(subtokenizer.decode(otr))
        bleu_eval.update(decode, labels)
        print('Accuracy is {:.3f}'.format(bleu_eval.result()))
        return bleu_eval.result()
 
class Dataset(object):
    def __init__(self, inputs_file, reference_file, vocab_file):
        with tf.io.gfile.GFile(inputs_file) as f:
            records = f.read().split("\n")
            inputs = [record.strip() for record in records]
            if not inputs[-1]:
                inputs.pop()
    
        self.ref_lines = tokenizer.native_to_unicode(
            tf.io.gfile.GFile(reference_file).read()).strip().splitlines()
        
        subtokenizer = Subtokenizer(vocab_file)
        self.batch = []
        token_lens=[]
        for i, line in enumerate(inputs):
            enc = subtokenizer.encode(line, add_eos=True)
            token_lens.append((i, len(enc)))
    
        sorted_by_token_input_lens = sorted(token_lens, key=lambda x: x[1], reverse=True)
    
        sorted_inputs = [None] * len(sorted_by_token_input_lens)
        sorted_keys = [0] * len(sorted_by_token_input_lens)
    
        lines = []
        for i, (index, _) in enumerate(sorted_by_token_input_lens):
            sorted_inputs[i] = inputs[index]
            sorted_keys[index] = i
            enc=subtokenizer.encode(sorted_inputs[i], add_eos=True)
            lines.append([enc])
        for i in sorted_keys:
            self.batch.append(lines[i])

    def __getitem__(self, index):
        data = self.batch[index]
        label = self.ref_lines[index]
        return data[0], label

    def __len__(self):
        return len(self.batch)

def main(_):
    graph = load_graph(FLAGS.input_graph)
    if FLAGS.mode == 'tune':
        from neural_compressor.experimental import Quantization, common
        quantizer = Quantization(FLAGS.config)
        ds = Dataset(FLAGS.inputs_file, FLAGS.reference_file, FLAGS.vocab_file)
        quantizer.calib_dataloader = common.DataLoader(ds, collate_fn=collate_fn, \
                                                 batch_size=FLAGS.batch_size)
        quantizer.model = common.Model(graph)
        quantizer.eval_func = eval_func
        q_model = quantizer.fit()
        try:
            q_model.save(FLAGS.output_model)
        except Exception as e:
            print("Failed to save model due to {}".format(str(e)))
    elif FLAGS.mode == 'benchmark':
        eval_func(graph, FLAGS.iters)
    elif FLAGS.mode == 'accuracy':
        eval_func(graph, -1)
        
if __name__ == "__main__":
    tf.compat.v1.app.run()