fluid_benchmark.py 13.6 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
# Copyright (c) 2018 PaddlePaddle Authors. 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.

import argparse
import cProfile
import time
import os

import numpy as np

import paddle.fluid as fluid
import paddle.fluid.core as core
import paddle.fluid.profiler as profiler
import paddle.fluid.transpiler.distribute_transpiler as distribute_transpiler

27
from args import *
28 29


X
Xin Pan 已提交
30 31
def append_nccl2_prepare(trainer_id):
    if trainer_id >= 0:
32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57
        # append gen_nccl_id at the end of startup program
        trainer_id = int(os.getenv("PADDLE_TRAINER_ID"))
        port = os.getenv("PADDLE_PSERVER_PORT")
        worker_ips = os.getenv("PADDLE_TRAINER_IPS")
        worker_endpoints = []
        for ip in worker_ips.split(","):
            worker_endpoints.append(':'.join([ip, port]))
        num_trainers = len(worker_endpoints)
        current_endpoint = os.getenv("PADDLE_CURRENT_IP") + ":" + port
        worker_endpoints.remove(current_endpoint)

        nccl_id_var = fluid.default_startup_program().global_block().create_var(
            name="NCCLID",
            persistable=True,
            type=fluid.core.VarDesc.VarType.RAW)
        fluid.default_startup_program().global_block().append_op(
            type="gen_nccl_id",
            inputs={},
            outputs={"NCCLID": nccl_id_var},
            attrs={
                "endpoint": current_endpoint,
                "endpoint_list": worker_endpoints,
                "trainer_id": trainer_id
            })
        return nccl_id_var, num_trainers, trainer_id
    else:
X
Xin Pan 已提交
58 59
        raise Exception("must set positive PADDLE_TRAINER_ID env variables for "
                        "nccl-based dist train.")
60 61


62
def dist_transpile(trainer_id, args):
X
Xin Pan 已提交
63
    if trainer_id < 0:
64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83
        return None, None

    # the port of all pservers, needed by both trainer and pserver
    port = os.getenv("PADDLE_PSERVER_PORT", "6174")
    # comma separated ips of all pservers, needed by trainer and
    # pserver
    pserver_ips = os.getenv("PADDLE_PSERVER_IPS", "")
    eplist = []
    for ip in pserver_ips.split(","):
        eplist.append(':'.join([ip, port]))
    pserver_endpoints = ",".join(eplist)
    # total number of workers/trainers in the job, needed by
    # trainer and pserver
    trainers = int(os.getenv("PADDLE_TRAINERS"))
    # the IP of the local machine, needed by pserver only
    current_endpoint = os.getenv("PADDLE_CURRENT_IP", "") + ":" + port
    # the role, should be either PSERVER or TRAINER
    training_role = os.getenv("PADDLE_TRAINING_ROLE")

    t = distribute_transpiler.DistributeTranspiler()
84 85 86 87
    t.transpile(
        trainer_id,
        pservers=pserver_endpoints,
        trainers=trainers,
88
        sync_mode=not args.async_mode)
89 90 91 92 93 94 95 96 97 98
    if training_role == "PSERVER":
        pserver_program = t.get_pserver_program(current_endpoint)
        pserver_startup_program = t.get_startup_program(current_endpoint,
                                                        pserver_program)
        return pserver_program, pserver_startup_program
    elif training_role == "TRAINER":
        train_program = t.get_trainer_program()
        return train_program, fluid.default_startup_program()
    else:
        raise ValueError(
G
gongweibao 已提交
99
            'PADDLE_TRAINING_ROLE environment variable must be either TRAINER or PSERVER'
100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124
        )


def test(exe, inference_program, test_reader, feeder, batch_acc):
    accuracy_evaluator = fluid.metrics.Accuracy()
    for batch_id, data in enumerate(test_reader()):
        acc = exe.run(inference_program,
                      feed=feeder.feed(data),
                      fetch_list=[batch_acc])
        accuracy_evaluator.update(value=np.array(acc), weight=len(data))

    return accuracy_evaluator.eval()


# TODO(wuyi): replace train, train_parallel, test functions with new trainer
# API once it is ready.
def train(avg_loss, infer_prog, optimizer, train_reader, test_reader, batch_acc,
          args, train_prog, startup_prog):
    if os.getenv("PADDLE_TRAINING_ROLE") == "PSERVER":
        place = core.CPUPlace()
        exe = fluid.Executor(place)
        exe.run(startup_prog)
        exe.run(train_prog)
        return

125 126 127 128
    if args.use_fake_data:
        raise Exception(
            "fake data is not supported in single GPU test for now.")

129 130 131
    place = core.CPUPlace() if args.device == 'CPU' else core.CUDAPlace(0)
    exe = fluid.Executor(place)
    exe.run(startup_prog)
Y
yi.wu 已提交
132

133
    # Use inference_transpiler to speedup
Y
yi.wu 已提交
134 135 136 137 138 139
    if not args.use_reader_op:
        feed_var_list = [
            var for var in train_prog.global_block().vars.itervalues()
            if var.is_data
        ]
        feeder = fluid.DataFeeder(feed_var_list, place)
140 141 142 143

    iters, num_samples, start_time = 0, 0, time.time()
    for pass_id in range(args.pass_num):
        train_losses = []
Y
yi.wu 已提交
144 145
        if not args.use_reader_op:
            reader_generator = train_reader()
146 147 148 149 150
        batch_id = 0
        data = None
        while True:
            if not args.use_reader_op:
                data = next(reader_generator, None)
Y
yi.wu 已提交
151 152 153
                if data == None:
                    break
            if iters == args.iterations:
154
                reader_generator.close()
155
                break
156 157 158
            if iters == args.skip_batch_num:
                start_time = time.time()
                num_samples = 0
159

Y
yi.wu 已提交
160
            if args.use_reader_op:
Y
yi.wu 已提交
161 162 163 164
                try:
                    loss = exe.run(train_prog, fetch_list=[avg_loss])
                except fluid.core.EnforceNotMet as ex:
                    break
Y
yi.wu 已提交
165 166 167 168
            else:
                loss = exe.run(train_prog,
                               feed=feeder.feed(data),
                               fetch_list=[avg_loss])
169
            iters += 1
170
            batch_id += 1
Y
yi.wu 已提交
171 172 173
            # FIXME(wuyi): For use_reader_op, if the current
            # pass is not the last, the last batch of this pass
            # is also equal to args.batch_size.
Y
update  
yi.wu 已提交
174
            if args.use_reader_op:
Y
yi.wu 已提交
175
                num_samples += args.batch_size * args.gpus
Y
update  
yi.wu 已提交
176 177
            else:
                num_samples += len(data)
178 179 180
            train_losses.append(loss)
            print("Pass: %d, Iter: %d, Loss: %f\n" %
                  (pass_id, iters, np.mean(train_losses)))
L
Luo Tao 已提交
181
        print_train_time(start_time, time.time(), num_samples)
L
Luo Tao 已提交
182
        print("Pass: %d, Loss: %f" % (pass_id, np.mean(train_losses))),
183
        # evaluation
G
guochaorong 已提交
184
        if not args.no_test and batch_acc and not args.use_reader_op:
185 186 187 188
            if args.use_inference_transpiler:
                t = fluid.InferenceTranspiler()
                t.transpile(infer_prog, place)

189 190 191 192 193 194 195 196 197 198 199 200 201
            pass_test_acc = test(exe, infer_prog, test_reader, feeder,
                                 batch_acc)
            print(", Test Accuracy: %f" % pass_test_acc)
        print("\n")
        # TODO(wuyi): add warmup passes to get better perf data.
        exit(0)


# TODO(wuyi): replace train, train_parallel, test functions with new trainer
# API once it is ready.
def train_parallel(avg_loss, infer_prog, optimizer, train_reader, test_reader,
                   batch_acc, args, train_prog, startup_prog, nccl_id_var,
                   num_trainers, trainer_id):
Y
yi.wu 已提交
202 203 204 205 206 207 208 209
    place = core.CPUPlace() if args.device == 'CPU' else core.CUDAPlace(0)
    if not args.use_reader_op:
        feed_var_list = [
            var for var in train_prog.global_block().vars.itervalues()
            if var.is_data
        ]
        feeder = fluid.DataFeeder(feed_var_list, place)

210 211 212
    # generate fake:
    if args.use_fake_data:
        for var in feed_var_list:
W
Wu Yi 已提交
213
            v = startup_prog.global_block()._clone_variable(var)
214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229
            var.persistable = True
            v.persistable = True

            real_shape = list(var.shape)
            real_shape[0] = args.batch_size / args.gpus
            startup_prog.global_block().append_op(
                outputs={"Out": v},
                type="fill_constant",
                attrs={"shape": real_shape,
                       "value": 1.0,
                       "dtype": var.dtype})

    if nccl_id_var and trainer_id == 0:
        #FIXME(wuyi): wait other trainer to start listening
        time.sleep(30)

230 231 232 233 234 235 236 237 238 239 240
    startup_exe = fluid.Executor(place)
    startup_exe.run(startup_prog)
    strategy = fluid.ExecutionStrategy()
    strategy.num_threads = 1
    strategy.allow_op_delay = False
    exe = fluid.ParallelExecutor(
        True,
        avg_loss.name,
        exec_strategy=strategy,
        num_trainers=num_trainers,
        trainer_id=trainer_id)
241

242 243 244 245
    for pass_id in range(args.pass_num):
        num_samples = 0
        iters = 0
        start_time = time.time()
Y
yi.wu 已提交
246 247
        if not args.use_reader_op:
            reader_generator = train_reader()
248 249 250 251 252
        batch_id = 0
        data = None
        while True:
            if not args.use_reader_op:
                data = next(reader_generator, None)
Y
yi.wu 已提交
253 254 255
                if data == None:
                    break
            if iters == args.iterations:
256
                reader_generator.close()
257
                break
X
Xin Pan 已提交
258 259 260 261 262
            if args.profile and pass_id == 0 and batch_id == 5:
                profiler.start_profiler("All")
            elif args.profile and pass_id == 0 and batch_id == 10:
                profiler.stop_profiler("total", "/tmp/profile_%d" % trainer_id)

263 264 265
            if iters == args.skip_batch_num:
                start_time = time.time()
                num_samples = 0
Y
yi.wu 已提交
266
            if args.use_fake_data or args.use_reader_op:
Y
yi.wu 已提交
267 268 269 270
                try:
                    loss, = exe.run([avg_loss.name])
                except fluid.core.EnforceNotMet as ex:
                    break
271 272
            else:
                loss, = exe.run([avg_loss.name], feed=feeder.feed(data))
Y
update  
yi.wu 已提交
273
            if args.use_reader_op:
Y
yi.wu 已提交
274
                num_samples += args.batch_size * args.gpus
Y
update  
yi.wu 已提交
275 276
            else:
                num_samples += len(data)
277 278 279 280
            iters += 1
            if batch_id % 1 == 0:
                print("Pass %d, batch %d, loss %s" %
                      (pass_id, batch_id, np.array(loss)))
281
            batch_id += 1
Y
yi.wu 已提交
282

L
Luo Tao 已提交
283
        print_train_time(start_time, time.time(), num_samples)
G
guochaorong 已提交
284 285 286
        if not args.no_test and batch_acc and not args.use_reader_op:
            # we have not implement record io for test
            # skip test when use args.use_reader_op
287 288 289 290 291 292 293 294
            test_acc = test(startup_exe, infer_prog, test_reader, feeder,
                            batch_acc)
            print("Pass: %d, Test Accuracy: %f\n" % (pass_id, test_acc))


def print_arguments(args):
    vars(args)['use_nvprof'] = (vars(args)['use_nvprof'] and
                                vars(args)['device'] == 'GPU')
L
Luo Tao 已提交
295
    print('----------- Configuration Arguments -----------')
296 297 298 299 300
    for arg, value in sorted(vars(args).iteritems()):
        print('%s: %s' % (arg, value))
    print('------------------------------------------------')


L
Luo Tao 已提交
301 302 303 304 305 306 307
def print_train_time(start_time, end_time, num_samples):
    train_elapsed = end_time - start_time
    examples_per_sec = num_samples / train_elapsed
    print('\nTotal examples: %d, total time: %.5f, %.5f examples/sed\n' %
          (num_samples, train_elapsed, examples_per_sec))


308 309 310 311 312 313 314 315
def print_paddle_envs():
    print('----------- Configuration envs -----------')
    for k in os.environ:
        if "PADDLE_" in k:
            print "ENV %s:%s" % (k, os.environ[k])
    print('------------------------------------------------')


316 317 318
def main():
    args = parse_args()
    print_arguments(args)
319
    print_paddle_envs()
320 321
    if args.no_random:
        fluid.default_startup_program().random_seed = 1
X
Xin Pan 已提交
322 323 324 325

    # the unique trainer id, starting from 0, needed by trainer
    # only
    nccl_id_var, num_trainers, trainer_id = (
Y
yi.wu 已提交
326
        None, 1, int(os.getenv("PADDLE_TRAINER_ID", "0")))
327 328 329 330 331 332 333 334 335 336 337 338 339

    if args.use_cprof:
        pr = cProfile.Profile()
        pr.enable()
    model_def = __import__("models.%s" % args.model, fromlist=["models"])
    train_args = list(model_def.get_model(args))
    train_args.append(args)
    # Run optimizer.minimize(avg_loss)
    train_args[2].minimize(train_args[0])
    if args.memory_optimize:
        fluid.memory_optimize(fluid.default_main_program())

    if args.update_method == "pserver":
340
        train_prog, startup_prog = dist_transpile(trainer_id, args)
341 342 343 344 345 346 347 348 349 350 351 352 353 354 355
        if not train_prog:
            raise Exception(
                "Must configure correct environments to run dist train.")
        train_args.extend([train_prog, startup_prog])
        if args.gpus > 1 and os.getenv("PADDLE_TRAINING_ROLE") == "TRAINER":
            train_args.extend([nccl_id_var, num_trainers, trainer_id])
            train_parallel(*train_args)
        train(*train_args)
        exit(0)

    # for other update methods, use default programs
    train_args.append(fluid.default_main_program())
    train_args.append(fluid.default_startup_program())

    if args.update_method == "nccl2":
X
Xin Pan 已提交
356
        nccl_id_var, num_trainers, trainer_id = append_nccl2_prepare(trainer_id)
357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372
    if args.gpus == 1:
        # NOTE: parallel executor use profiler interanlly
        if args.use_nvprof and args.device == 'GPU':
            with profiler.cuda_profiler("cuda_profiler.txt", 'csv') as nvprof:
                train(*train_args)
        else:
            train(*train_args)
    else:
        if args.device == "CPU":
            raise Exception("Only support GPU perf with parallel exe")
        train_args.extend([nccl_id_var, num_trainers, trainer_id])
        train_parallel(*train_args)


if __name__ == "__main__":
    main()