#!/usr/bin/env python
# coding=utf-8
# Copyright 2017-2019 The THUMT Authors
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import argparse
import os
import six
import numpy as np
import tensorflow as tf
import thumt.data.dataset as dataset
import thumt.data.record as record
import thumt.data.vocab as vocabulary
import thumt.models as models
import thumt.utils.distribute as distribute
import thumt.utils.hooks as hooks
import thumt.utils.inference as inference
import thumt.utils.optimizers as optimizers
import thumt.utils.parallel as parallel
def parse_args(args=None):
parser = argparse.ArgumentParser(
description="Training neural machine translation models",
usage="trainer.py [<args>] [-h | --help]"
)
# input files
parser.add_argument("--input", type=str, nargs=2,
help="Path of source and target corpus")
parser.add_argument("--record", type=str,
help="Path to tf.Record data")
parser.add_argument("--output", type=str, default="train",
help="Path to saved models")
parser.add_argument("--vocabulary", type=str, nargs=2,
help="Path of source and target vocabulary")
parser.add_argument("--validation", type=str,
help="Path of validation file")
parser.add_argument("--references", type=str, nargs="+",
help="Path of reference files")
parser.add_argument("--checkpoint", type=str,
help="Path to pre-trained checkpoint")
parser.add_argument("--half", action="store_true",
help="Enable FP16 training")
parser.add_argument("--distribute", action="store_true",
help="Enable distributed training")
# model and configuration
parser.add_argument("--model", type=str, required=True,
help="Name of the model")
parser.add_argument("--parameters", type=str, default="",
help="Additional hyper parameters")
return parser.parse_args(args)
def default_parameters():
params = tf.contrib.training.HParams(
input=["", ""],
output="",
record="",
model="transformer",
vocab=["", ""],
# Default training hyper parameters
num_threads=6,
batch_size=4096,
max_length=256,
length_multiplier=1,
mantissa_bits=2,
warmup_steps=4000,
train_steps=100000,
buffer_size=10000,
constant_batch_size=False,
device_list=[0],
update_cycle=1,
initializer="uniform_unit_scaling",
initializer_gain=1.0,
loss_scale=128,
scale_l1=0.0,
scale_l2=0.0,
optimizer="Adam",
adam_beta1=0.9,
adam_beta2=0.999,
adam_epsilon=1e-8,
clip_grad_norm=5.0,
learning_rate=1.0,
learning_rate_decay="linear_warmup_rsqrt_decay",
learning_rate_boundaries=[0],
learning_rate_values=[0.0],
keep_checkpoint_max=20,
keep_top_checkpoint_max=5,
# Validation
eval_steps=2000,
eval_secs=0,
eval_batch_size=32,
top_beams=1,
beam_size=4,
decode_alpha=0.6,
decode_length=50,
validation="",
references=[""],
save_checkpoint_secs=0,
save_checkpoint_steps=1000,
# Setting this to True can save disk spaces, but cannot restore
# training using the saved checkpoint
only_save_trainable=False
)
return params
def import_params(model_dir, model_name, params):
model_dir = os.path.abspath(model_dir)
p_name = os.path.join(model_dir, "params.json")
m_name = os.path.join(model_dir, model_name + ".json")
if not tf.gfile.Exists(p_name) or not tf.gfile.Exists(m_name):
return params
with tf.gfile.Open(p_name) as fd:
tf.logging.info("Restoring hyper parameters from %s" % p_name)
json_str = fd.readline()
params.parse_json(json_str)
with tf.gfile.Open(m_name) as fd:
tf.logging.info("Restoring model parameters from %s" % m_name)
json_str = fd.readline()
params.parse_json(json_str)
return params
def export_params(output_dir, name, params):
if not tf.gfile.Exists(output_dir):
tf.gfile.MkDir(output_dir)
# Save params as params.json
filename = os.path.join(output_dir, name)
with tf.gfile.Open(filename, "w") as fd:
fd.write(params.to_json())
def collect_params(all_params, params):
collected = tf.contrib.training.HParams()
for k in six.iterkeys(params.values()):
collected.add_hparam(k, getattr(all_params, k))
return collected
def merge_parameters(params1, params2):
params = tf.contrib.training.HParams()
for (k, v) in six.iteritems(params1.values()):
params.add_hparam(k, v)
params_dict = params.values()
for (k, v) in six.iteritems(params2.values()):
if k in params_dict:
# Override
setattr(params, k, v)
else:
params.add_hparam(k, v)
return params
def override_parameters(params, args):
params.model = args.model
params.input = args.input or params.input
params.output = args.output or params.output
params.record = args.record or params.record
params.vocab = args.vocabulary or params.vocab
params.validation = args.validation or params.validation
params.references = args.references or params.references
params.parse(args.parameters)
params.vocabulary = {
"source": vocabulary.load_vocabulary(params.vocab[0]),
"target": vocabulary.load_vocabulary(params.vocab[1])
}
params.vocabulary["source"] = vocabulary.process_vocabulary(
params.vocabulary["source"], params
)
params.vocabulary["target"] = vocabulary.process_vocabulary(
params.vocabulary["target"], params
)
control_symbols = [params.pad, params.bos, params.eos, params.unk]
params.mapping = {
"source": vocabulary.get_control_mapping(
params.vocabulary["source"],
control_symbols
),
"target": vocabulary.get_control_mapping(
params.vocabulary["target"],
control_symbols
)
}
return params
def get_initializer(params):
if params.initializer == "uniform":
max_val = params.initializer_gain
return tf.random_uniform_initializer(-max_val, max_val)
elif params.initializer == "normal":
return tf.random_normal_initializer(0.0, params.initializer_gain)
elif params.initializer == "normal_unit_scaling":
return tf.variance_scaling_initializer(params.initializer_gain,
mode="fan_avg",
distribution="normal")
elif params.initializer == "uniform_unit_scaling":
return tf.variance_scaling_initializer(params.initializer_gain,
mode="fan_avg",
distribution="uniform")
else:
raise ValueError("Unrecognized initializer: %s" % params.initializer)
def get_learning_rate_decay(learning_rate, global_step, params):
if params.learning_rate_decay in ["linear_warmup_rsqrt_decay", "noam"]:
step = tf.to_float(global_step)
warmup_steps = tf.to_float(params.warmup_steps)
multiplier = params.hidden_size ** -0.5
decay = multiplier * tf.minimum((step + 1) * (warmup_steps ** -1.5),
(step + 1) ** -0.5)
return learning_rate * decay
elif params.learning_rate_decay == "piecewise_constant":
return tf.train.piecewise_constant(tf.to_int32(global_step),
params.learning_rate_boundaries,
params.learning_rate_values)
elif params.learning_rate_decay == "none":
return learning_rate
else:
raise ValueError("Unknown learning_rate_decay")
def session_config(params):
optimizer_options = tf.OptimizerOptions(opt_level=tf.OptimizerOptions.L1,
do_function_inlining=True)
graph_options = tf.GraphOptions(optimizer_options=optimizer_options)
config = tf.ConfigProto(allow_soft_placement=True,
graph_options=graph_options)
if distribute.is_distributed_training_mode():
config.gpu_options.visible_device_list = str(distribute.local_rank())
elif params.device_list:
device_str = ",".join([str(i) for i in params.device_list])
config.gpu_options.visible_device_list = device_str
return config
def decode_target_ids(inputs, params):
decoded = []
vocab = params.vocabulary["target"]
for item in inputs:
syms = []
for idx in item:
if isinstance(idx, six.integer_types):
sym = vocab[idx]
else:
sym = idx
if not isinstance(sym, six.string_types):
sym = sym.decode("utf-8")
if sym == params.eos:
break
if sym == params.pad:
break
syms.append(sym)
decoded.append(syms)
return decoded
def restore_variables(checkpoint):
if not checkpoint:
return tf.no_op("restore_op")
# Load checkpoints
tf.logging.info("Loading %s" % checkpoint)
var_list = tf.train.list_variables(checkpoint)
reader = tf.train.load_checkpoint(checkpoint)
values = {}
for (name, shape) in var_list:
tensor = reader.get_tensor(name)
name = name.split(":")[0]
values[name] = tensor
var_list = tf.trainable_variables()
ops = []
for var in var_list:
name = var.name.split(":")[0]
if name in values:
tf.logging.info("Restore %s" % var.name)
ops.append(tf.assign(var, values[name]))
return tf.group(*ops, name="restore_op")
def print_variables():
all_weights = {v.name: v for v in tf.trainable_variables()}
total_size = 0
for v_name in sorted(list(all_weights)):
v = all_weights[v_name]
tf.logging.info("%s\tshape %s", v.name[:-2].ljust(80),
str(v.shape).ljust(20))
v_size = np.prod(np.array(v.shape.as_list())).tolist()
total_size += v_size
tf.logging.info("Total trainable variables size: %d", total_size)
def main(args):
if args.distribute:
distribute.enable_distributed_training()
tf.logging.set_verbosity(tf.logging.INFO)
model_cls = models.get_model(args.model)
params = default_parameters()
# Import and override parameters
# Priorities (low -> high):
# default -> saved -> command
params = merge_parameters(params, model_cls.get_parameters())
params = import_params(args.output, args.model, params)
override_parameters(params, args)
# Export all parameters and model specific parameters
if distribute.rank() == 0:
export_params(params.output, "params.json", params)
export_params(
params.output,
"%s.json" % args.model,
collect_params(params, model_cls.get_parameters())
)
# Build Graph
with tf.Graph().as_default():
if not params.record:
# Build input queue
features = dataset.get_training_input(params.input, params)
else:
features = record.get_input_features(
os.path.join(params.record, "*train*"), "train", params
)
# Build model
initializer = get_initializer(params)
regularizer = tf.contrib.layers.l1_l2_regularizer(
scale_l1=params.scale_l1, scale_l2=params.scale_l2)
model = model_cls(params)
# Create global step
global_step = tf.train.get_or_create_global_step()
dtype = tf.float16 if args.half else None
# Multi-GPU setting
sharded_losses = parallel.parallel_model(
model.get_training_func(initializer, regularizer, dtype),
features,
params.device_list
)
loss = tf.add_n(sharded_losses) / len(sharded_losses)
loss = loss + tf.losses.get_regularization_loss()
if distribute.rank() == 0:
print_variables()
learning_rate = get_learning_rate_decay(params.learning_rate,
global_step, params)
learning_rate = tf.convert_to_tensor(learning_rate, dtype=tf.float32)
tf.summary.scalar("loss", loss)
tf.summary.scalar("learning_rate", learning_rate)
# Create optimizer
if params.optimizer == "Adam":
opt = tf.train.AdamOptimizer(learning_rate,
beta1=params.adam_beta1,
beta2=params.adam_beta2,
epsilon=params.adam_epsilon)
elif params.optimizer == "LazyAdam":
opt = tf.contrib.opt.LazyAdamOptimizer(learning_rate,
beta1=params.adam_beta1,
beta2=params.adam_beta2,
epsilon=params.adam_epsilon)
else:
raise RuntimeError("Optimizer %s not supported" % params.optimizer)
opt = optimizers.MultiStepOptimizer(opt, params.update_cycle)
if args.half:
opt = optimizers.LossScalingOptimizer(opt, params.loss_scale)
# Optimization
grads_and_vars = opt.compute_gradients(
loss, colocate_gradients_with_ops=True)
if params.clip_grad_norm:
grads, var_list = list(zip(*grads_and_vars))
grads, _ = tf.clip_by_global_norm(grads, params.clip_grad_norm)
grads_and_vars = zip(grads, var_list)
train_op = opt.apply_gradients(grads_and_vars,
global_step=global_step)
# Validation
if params.validation and params.references[0]:
files = [params.validation] + list(params.references)
eval_inputs = dataset.sort_and_zip_files(files)
eval_input_fn = dataset.get_evaluation_input
else:
eval_input_fn = None
# Hooks
train_hooks = [
tf.train.StopAtStepHook(last_step=params.train_steps),
tf.train.NanTensorHook(loss),
tf.train.LoggingTensorHook(
{
"step": global_step,
"loss": loss,
"source": tf.shape(features["source"]),
"target": tf.shape(features["target"])
},
every_n_iter=1
)
]
broadcast_hook = distribute.get_broadcast_hook()
if broadcast_hook:
train_hooks.append(broadcast_hook)
if distribute.rank() == 0:
# Add hooks
save_vars = tf.trainable_variables() + [global_step]
saver = tf.train.Saver(
var_list=save_vars if params.only_save_trainable else None,
max_to_keep=params.keep_checkpoint_max,
sharded=False
)
tf.add_to_collection(tf.GraphKeys.SAVERS, saver)
train_hooks.append(
hooks.MultiStepHook(
tf.train.CheckpointSaverHook(
checkpoint_dir=params.output,
save_secs=params.save_checkpoint_secs or None,
save_steps=params.save_checkpoint_steps or None,
saver=saver),
step=params.update_cycle)
)
if eval_input_fn is not None:
train_hooks.append(
hooks.MultiStepHook(
hooks.EvaluationHook(
lambda f: inference.create_inference_graph(
[model], f, params
),
lambda: eval_input_fn(eval_inputs, params),
lambda x: decode_target_ids(x, params),
params.output,
session_config(params),
device_list=params.device_list,
max_to_keep=params.keep_top_checkpoint_max,
eval_secs=params.eval_secs,
eval_steps=params.eval_steps
),
step=params.update_cycle
)
)
checkpoint_dir = params.output
else:
checkpoint_dir = None
restore_op = restore_variables(args.checkpoint)
def restore_fn(step_context):
step_context.session.run(restore_op)
# Create session, do not use default CheckpointSaverHook
with tf.train.MonitoredTrainingSession(
checkpoint_dir=checkpoint_dir, hooks=train_hooks,
save_checkpoint_secs=None,
config=session_config(params)) as sess:
# Restore pre-trained variables
sess.run_step_fn(restore_fn)
while not sess.should_stop():
sess.run(train_op)
if __name__ == "__main__":
main(parse_args())
