ps_dnn_trainer.py

# Copyright (c) 2022 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 ast
import copy
import os
import struct
import sys

import numpy as np
import yaml
from ps_dnn_model import StaticModel

import paddle
from paddle.distributed import fleet
from paddle.distributed.fleet.base import role_maker
from paddle.distributed.ps.utils.ps_program_builder import (
    debug_program,
    logger,
    new_pass,
    ps_log_root_dir,
)

__dir__ = os.path.dirname(os.path.abspath(__file__))
sys.path.append(os.path.abspath(os.path.join(__dir__, '..')))


def is_distributed_env():
    node_role = os.getenv("TRAINING_ROLE")
    print(f"-- Role: {node_role} --")
    if node_role is None:
        return False
    else:
        return True


class YamlHelper:
    def load_yaml(self, yaml_file, other_part=None):
        part_list = ["runner", "hyper_parameters"]
        if other_part:
            part_list += other_part
        running_config = self.get_all_inters_from_yaml(yaml_file, part_list)
        running_config = self.workspace_adapter(running_config)
        return running_config

    def print_yaml(self, config):
        print(self.pretty_print_envs(config))

    def parse_yaml(self, config):
        vs = [int(i) for i in yaml.__version__.split(".")]
        if vs[0] < 5:
            use_full_loader = False
        elif vs[0] > 5:
            use_full_loader = True
        else:
            if vs[1] >= 1:
                use_full_loader = True
            else:
                use_full_loader = False

        if os.path.isfile(config):
            with open(config, 'r', encoding="utf-8") as rb:
                if use_full_loader:
                    _config = yaml.load(rb.read(), Loader=yaml.FullLoader)
                else:
                    _config = yaml.load(rb.read())
                return _config
        else:
            raise ValueError(f"config {config} can not be supported")

    def get_all_inters_from_yaml(self, file, filters):
        _envs = self.parse_yaml(file)
        all_flattens = {}

        def fatten_env_namespace(namespace_nests, local_envs):
            for k, v in local_envs.items():
                if isinstance(v, dict):
                    nests = copy.deepcopy(namespace_nests)
                    nests.append(k)
                    fatten_env_namespace(nests, v)
                else:
                    global_k = ".".join(namespace_nests + [k])
                    all_flattens[global_k] = v

        fatten_env_namespace([], _envs)
        ret = {}
        for k, v in all_flattens.items():
            for f in filters:
                if k.startswith(f):
                    ret[k] = v
        return ret

    def workspace_adapter(self, config):
        workspace = config.get("workspace")
        for k, v in config.items():
            if isinstance(v, str) and "{workspace}" in v:
                config[k] = v.replace("{workspace}", workspace)
        return config

    def pretty_print_envs(self, envs, header=None):
        spacing = 2
        max_k = 40
        max_v = 45

        for k, v in envs.items():
            max_k = max(max_k, len(k))

        h_format = "    " + "|{{:>{}s}}{}{{:^{}s}}|\n".format(
            max_k, " " * spacing, max_v
        )
        l_format = "    " + f"|{{:>{max_k}s}}{{}}{{:^{max_v}s}}|\n"
        length = max_k + max_v + spacing

        border = "    +" + "".join(["="] * length) + "+"
        line = "    +" + "".join(["-"] * length) + "+"

        draws = ""
        draws += border + "\n"

        if header:
            draws += h_format.format(header[0], header[1])
        else:
            draws += h_format.format("Ps Benchmark Envs", "Value")

        draws += line + "\n"

        for k, v in sorted(envs.items()):
            if isinstance(v, str) and len(v) >= max_v:
                str_v = "... " + v[-41:]
            else:
                str_v = v

            draws += l_format.format(k, " " * spacing, str(str_v))

        draws += border

        _str = f"\n{draws}\n"
        return _str


def get_user_defined_strategy(config):
    if not is_distributed_env():
        logger.warn(
            "Not Find Distributed env, Change To local train mode. If you want train with fleet, please use [fleetrun] command."
        )
        # return None
    sync_mode = config.get("runner.sync_mode")
    assert sync_mode in ["async", "sync", "geo", "heter", "gpubox"]
    if sync_mode == "sync":
        strategy = paddle.distributed.fleet.DistributedStrategy()
        strategy.a_sync = False
    elif sync_mode == "async":
        strategy = paddle.distributed.fleet.DistributedStrategy()
        strategy.a_sync = True
        strategy.is_fl_ps_mode = (
            True if config.get("runner.is_fl_ps_mode") == 1 else False
        )
        if strategy.is_fl_ps_mode:
            strategy.pipeline = False
            micro_num = 1
            strategy.pipeline_configs = {
                "accumulate_steps": micro_num
            }  # num_microbatches
    elif sync_mode == "geo":
        strategy = paddle.distributed.fleet.DistributedStrategy()
        strategy.a_sync = True
        strategy.a_sync_configs = {"k_steps": config.get("runner.geo_step")}
    elif sync_mode == "heter":
        strategy = paddle.distributed.fleet.DistributedStrategy()
        strategy.a_sync = True
        strategy.a_sync_configs = {"heter_worker_device_guard": "gpu"}
        strategy.pipeline = True
        strategy.pipeline_configs = {
            "accumulate_steps": config.get('runner.micro_num')
        }
    elif sync_mode == "gpubox":
        print(f"sync_mode = {sync_mode}")
        strategy = paddle.distributed.fleet.DistributedStrategy()
        strategy.a_sync = True
        strategy.a_sync_configs = {"use_ps_gpu": 1}

    strategy.trainer_desc_configs = {
        "dump_fields_path": config.get("runner.dump_fields_path", ""),
        "dump_fields": config.get("runner.dump_fields", []),
        "dump_param": config.get("runner.dump_param", []),
        "stat_var_names": config.get("stat_var_names", []),
        "local_sparse": config.get("runner.local_sparse", []),
        "remote_sparse": config.get("runner.remote_sparse", []),
    }
    print("strategy:", strategy.trainer_desc_configs)

    if config.get("runner.fs_client.uri") is not None:
        strategy.fs_client_param = {
            "uri": config.get("runner.fs_client.uri", ""),
            "user": config.get("runner.fs_client.user", ""),
            "passwd": config.get("runner.fs_client.passwd", ""),
            "hadoop_bin": config.get("runner.fs_client.hadoop_bin", "hadoop"),
        }
    print("strategy:", strategy.fs_client_param)

    strategy.adam_d2sum = config.get("hyper_parameters.adam_d2sum", True)
    table_config = {}
    for x in config:
        if x.startswith("table_parameters"):
            table_name = x.split('.')[1]
            if table_name not in table_config:
                table_config[table_name] = {}
            table_config[table_name][x] = config[x]
    print("table_config:", table_config)
    strategy.sparse_table_configs = table_config
    print("strategy table config:", strategy.sparse_table_configs)
    a_sync_configs = strategy.a_sync_configs
    a_sync_configs["launch_barrier"] = False
    # a_sync_configs["launch_barrier"] = True
    strategy.a_sync_configs = a_sync_configs
    print("launch_barrier: ", strategy.a_sync_configs["launch_barrier"])

    return strategy


def get_distributed_strategy(user_defined_strategy):  # pslib
    from paddle.incubate.distributed.fleet.parameter_server.distribute_transpiler.distributed_strategy import (
        StrategyFactory,
    )

    k_steps = user_defined_strategy.a_sync_configs["k_steps"]
    strategy = None

    if not user_defined_strategy.a_sync and k_steps == 0:
        strategy = StrategyFactory.create_sync_strategy()

    if user_defined_strategy.a_sync and k_steps == 0:
        strategy = StrategyFactory.create_async_strategy()

    if user_defined_strategy.a_sync and k_steps > 0:
        strategy = StrategyFactory.create_geo_strategy(k_steps)

    if not strategy:
        raise ValueError("k_steps must be invalid value, please check")

    return strategy


def get_model(config):
    abs_dir = config['config_abs_dir']
    sys.path.append(abs_dir)
    static_model = StaticModel(config)
    return static_model


def parse_args():
    parser = argparse.ArgumentParser("PsTest train script")
    parser.add_argument(
        '-m', '--config_yaml', type=str, required=True, help='config file path'
    )
    parser.add_argument(
        '-bf16',
        '--pure_bf16',
        type=ast.literal_eval,
        default=False,
        help="whether use bf16",
    )

    parser.add_argument(
        '--run_minimize', type=int, default=0, help="test single pass"
    )
    parser.add_argument(
        '--run_single_pass', type=int, default=0, help="test single pass"
    )
    parser.add_argument(
        '--run_the_one_ps', type=int, default=0, help="test the_one_ps"
    )
    parser.add_argument(
        '--debug_new_minimize', type=int, default=0, help="test single pass"
    )
    parser.add_argument(
        '--debug_new_pass', type=int, default=0, help="test single pass"
    )
    parser.add_argument(
        '--applied_pass_name', type=str, default="", help="test single pass"
    )
    parser.add_argument(
        '--debug_the_one_ps', type=int, default=0, help="test the_one_ps"
    )

    args = parser.parse_args()
    args.abs_dir = os.path.dirname(os.path.abspath(args.config_yaml))
    yaml_helper = YamlHelper()
    config = yaml_helper.load_yaml(args.config_yaml)
    config["yaml_path"] = args.config_yaml
    config["config_abs_dir"] = args.abs_dir
    config["pure_bf16"] = args.pure_bf16
    config['run_minimize'] = args.run_minimize
    config['run_single_pass'] = args.run_single_pass
    config['run_the_one_ps'] = args.run_the_one_ps
    config['debug_new_minimize'] = args.debug_new_minimize
    config['debug_new_pass'] = args.debug_new_pass
    config['applied_pass_name'] = args.applied_pass_name
    config['debug_the_one_ps'] = args.debug_the_one_ps
    yaml_helper.print_yaml(config)
    return config


def bf16_to_fp32(val):
    return np.float32(struct.unpack('<f', struct.pack('<I', val << 16))[0])


class DnnTrainer:
    def __init__(self, config):
        self.metrics = {}
        self.config = config
        self.input_data = None
        self.reader = None
        self.exe = None
        self.train_result_dict = {}
        self.train_result_dict["speed"] = []
        self.model = None
        self.pure_bf16 = self.config['pure_bf16']
        self.role_maker = role_maker.PaddleCloudRoleMaker()

    def init_fleet_with_gloo(self, use_gloo=False):
        if use_gloo:
            os.environ["PADDLE_WITH_GLOO"] = "1"
            fleet.init(self.role_maker)
        else:
            fleet.init()

        if fleet.is_server():
            print(f"server: {fleet.server_index()} started")
        else:
            print(f"worker: {fleet.worker_index()} started")

    def run_minimize(self):
        self.init_fleet_with_gloo()
        self.model = get_model(self.config)
        print("cpu_num: {}".format(os.getenv("CPU_NUM")))
        self.input_data = self.model.create_feeds()
        self.metrics = self.model.net(self.input_data)
        loss = self.model._cost
        user_defined_strategy = get_user_defined_strategy(self.config)
        learning_rate = self.config.get(
            "hyper_parameters.optimizer.learning_rate"
        )
        sync_mode = self.config.get("runner.sync_mode")
        inner_optimizer = paddle.optimizer.Adam(learning_rate, lazy_mode=True)

        self.role_maker._generate_role()  # 必要
        if self.config['debug_new_minimize'] == 1:
            print("entering run_minimize -- new")
            from paddle.distributed.fleet.meta_optimizers.ps_optimizer import (
                ParameterServerOptimizer,
            )

            ps_optimizer = ParameterServerOptimizer(inner_optimizer)
            ps_optimizer._set_basic_info(
                loss, self.role_maker, inner_optimizer, user_defined_strategy
            )
            ps_optimizer.minimize_impl(loss)
        else:
            print("entering run_minimize -- old")
            fleet_obj = fleet.distributed_optimizer(
                inner_optimizer, user_defined_strategy
            )  # Fleet object
            fleet_obj.minimize(loss)

        if fleet.is_server():
            _main_file = (
                ps_log_root_dir
                + sync_mode
                + '_run_minimize'
                + '_debug:_'
                + str(self.config['debug_new_minimize'])
                + '_server_main.prototxt'
            )
            debug_program(_main_file, loss.block.program)
        elif fleet.is_worker():
            _main_file = (
                ps_log_root_dir
                + sync_mode
                + '_run_minimize'
                + '_debug:_'
                + str(self.config['debug_new_minimize'])
                + '_worker_main.prototxt'
            )
            debug_program(_main_file, loss.block.program)
        elif self.role_maker._is_heter_worker():
            _main_file = (
                ps_log_root_dir
                + sync_mode
                + '_run_minimize'
                + '_debug:_'
                + str(self.config['debug_new_minimize'])
                + '_heter_worker_main.prototxt'
            )
            debug_program(_main_file, loss.block.program)

    def run_single_pass(self):
        self.init_fleet_with_gloo()
        self.model = get_model(config)
        input_data = self.model.create_feeds()
        metrics = self.model.net(input_data)
        loss = self.model._cost
        user_defined_strategy = get_user_defined_strategy(config)
        learning_rate = config.get("hyper_parameters.optimizer.learning_rate")
        sync_mode = self.config.get("runner.sync_mode")
        inner_optimizer = paddle.optimizer.Adam(learning_rate, lazy_mode=True)
        startup_program = paddle.static.default_startup_program()
        inner_optimizer.minimize(loss, startup_program)
        if self.config['debug_new_pass'] == 1:
            print(
                "entering run {} - new".format(str(config["applied_pass_name"]))
            )
            from paddle.distributed.fleet.meta_optimizers.ps_optimizer import (
                ParameterServerOptimizer,
            )

            ps_optimizer = ParameterServerOptimizer(inner_optimizer)
            ps_optimizer._set_basic_info(
                loss, self.role_maker, inner_optimizer, user_defined_strategy
            )
            ps_optimizer._set_origin_programs([loss])
            ps_optimizer._init_ps_pass_context(loss, startup_program)
            _main = ps_optimizer.pass_ctx._attrs['cloned_main']

            append_send_ops_pass = new_pass(
                config["applied_pass_name"], ps_optimizer.pass_ctx._attrs
            )
            append_send_ops_pass.apply([_main], [None], ps_optimizer.pass_ctx)
        else:
            print(
                "entering run {} - old".format(str(config["applied_pass_name"]))
            )
            from paddle.incubate.distributed.fleet.parameter_server.ir import (
                public,
            )

            dist_strategy = get_distributed_strategy(user_defined_strategy)
            compiled_config = public.CompileTimeStrategy(
                loss.block.program,
                startup_program,
                dist_strategy,
                self.role_maker,
            )

            _main = compiled_config.origin_main_program.clone()
            _startup = compiled_config.origin_startup_program.clone()
            from paddle.incubate.distributed.fleet.parameter_server.ir import (
                trainer_pass as worker,
            )

            _main = worker.append_send_ops_pass(_main, compiled_config)

        if fleet.is_server():
            _main_file = (
                ps_log_root_dir
                + sync_mode
                + "_"
                + str(config["applied_pass_name"])
                + '_debug:_'
                + str(self.config['debug_new_pass'])
                + '_server_main.prototxt'
            )
            debug_program(_main_file, _main)
        elif fleet.is_worker():
            _main_file = (
                ps_log_root_dir
                + sync_mode
                + "_"
                + str(config["applied_pass_name"])
                + '_debug:_'
                + str(self.config['debug_new_pass'])
                + '_worker_main.prototxt'
            )
            debug_program(_main_file, _main)

    def run_the_one_ps(self):
        self.init_fleet_with_gloo()
        self.model = get_model(self.config)
        self.input_data = self.model.create_feeds()
        self.metrics = self.model.net(self.input_data)
        loss = self.model._cost
        user_defined_strategy = get_user_defined_strategy(self.config)
        learning_rate = self.config.get(
            "hyper_parameters.optimizer.learning_rate"
        )
        sync_mode = self.config.get("runner.sync_mode")
        inner_optimizer = paddle.optimizer.Adam(learning_rate, lazy_mode=True)

        self.role_maker._generate_role()  # 必要
        if self.config['debug_the_one_ps'] == 1:
            print("entering run_the_one_ps -- new")

            from paddle.distributed.fleet.meta_optimizers.ps_optimizer import (
                ParameterServerOptimizer,
            )

            ps_optimizer = ParameterServerOptimizer(inner_optimizer)
            ps_optimizer._set_basic_info(
                loss, self.role_maker, inner_optimizer, user_defined_strategy
            )
            ps_optimizer.minimize_impl(loss)

            from paddle.distributed.ps.the_one_ps import TheOnePSRuntime

            _runtime_handle = TheOnePSRuntime()  # ps 目录下重构版的 TheOnePSRuntime
            _runtime_handle._set_basic_info(ps_optimizer.pass_ctx._attrs)
            if fleet.is_worker():
                worker_desc = (
                    _runtime_handle.ps_desc_builder.build_worker_desc()
                )
                with open(
                    ps_log_root_dir + sync_mode + '_' + 'new_worker_ps_desc',
                    'w',
                ) as f:
                    f.write(worker_desc)
            if fleet.is_server():
                server_desc = (
                    _runtime_handle.ps_desc_builder.build_server_desc()
                )
                with open(
                    ps_log_root_dir + sync_mode + '_' + 'new_server_ps_desc',
                    'w',
                ) as f:
                    f.write(server_desc)

        else:
            pass
        '''
            print("entering run_the_one_ps -- old")
            fleet_obj = fleet.distributed_optimizer(
                inner_optimizer, user_defined_strategy)
            fleet_obj.minimize(loss)
            if fleet.is_worker():
                worker_desc = fleet_obj._runtime_handle._get_fleet_proto(is_server=False, is_sync=False)
                server_desc = fleet_obj._runtime_handle._get_fleet_proto(is_server=True, is_sync=False)
                with open(ps_log_root_dir + sync_mode + '_' + 'worker_ps_desc', 'w') as f:
                    f.write(str(worker_desc) + str(server_desc))
            if fleet.is_server():
                server_desc = fleet_obj._runtime_handle._get_fleet_proto(is_server=True, is_sync=False)
                with open(ps_log_root_dir + sync_mode + '_' + 'server_ps_desc', 'w') as f:
                    f.write(str(server_desc) + str(fleet_obj._runtime_handle._get_fs_client_desc().to_string()))
        '''
        if fleet.is_server():
            _main_file = (
                ps_log_root_dir
                + sync_mode
                + '_run_the_one_ps'
                + '_debug:_'
                + str(self.config['debug_the_one_ps'])
                + '_server_main.prototxt'
            )
            debug_program(_main_file, loss.block.program)
        elif fleet.is_worker():
            _main_file = (
                ps_log_root_dir
                + sync_mode
                + '_run_the_one_ps'
                + '_debug:_'
                + str(self.config['debug_the_one_ps'])
                + '_worker_main.prototxt'
            )
            debug_program(_main_file, loss.block.program)
        elif self.role_maker._is_heter_worker():
            _main_file = (
                ps_log_root_dir
                + sync_mode
                + '_run_the_one_ps'
                + '_debug:_'
                + str(self.config['debug_the_one_ps'])
                + '_heter_worker_main.prototxt'
            )
            debug_program(_main_file, loss.block.program)


if __name__ == "__main__":
    paddle.enable_static()
    config = parse_args()
    print(">>>>>>>>>> python process started")
    os.environ["CPU_NUM"] = str(config.get("runner.thread_num"))
    benchmark_main = DnnTrainer(config)
    if config['run_single_pass'] == 1:
        benchmark_main.run_single_pass()
    elif config['run_minimize'] == 1:
        benchmark_main.run_minimize()
    elif config['run_the_one_ps'] == 1:
        benchmark_main.run_the_one_ps()