run.py 10.7 KB
Newer Older
T
tangwei 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14
# Copyright (c) 2020 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.

T
tangwei 已提交
15
import os
T
tangwei 已提交
16
import subprocess
T
tangwei 已提交
17

T
tangwei 已提交
18 19
import argparse
import tempfile
T
tangwei 已提交
20
import yaml
X
fix  
xjqbest 已提交
21
import copy
22 23 24
from paddlerec.core.factory import TrainerFactory
from paddlerec.core.utils import envs
from paddlerec.core.utils import util
T
tangwei 已提交
25

T
tangwei 已提交
26 27
engines = {}
device = ["CPU", "GPU"]
T
tangwei 已提交
28
clusters = ["SINGLE", "LOCAL_CLUSTER", "CLUSTER"]
T
tangwei 已提交
29
engine_choices = [
X
fix  
xjqbest 已提交
30 31
    "SINGLE_TRAIN", "LOCAL_CLUSTER", "CLUSTER", "TDM_SINGLE",
    "TDM_LOCAL_CLUSTER", "TDM_CLUSTER", "SINGLE_INFER"
T
tangwei 已提交
32
]
T
tangwei 已提交
33
custom_model = ['TDM']
C
fix  
chengmo 已提交
34
model_name = ""
T
tangwei 已提交
35 36


T
tangwei 已提交
37
def engine_registry():
T
tangwei 已提交
38 39 40
    engines["TRANSPILER"] = {}
    engines["PSLIB"] = {}

X
fix  
xjqbest 已提交
41 42
    engines["TRANSPILER"]["SINGLE_TRAIN"] = single_train_engine
    engines["TRANSPILER"]["SINGLE_INFER"] = single_infer_engine
T
tangwei 已提交
43 44 45 46 47
    engines["TRANSPILER"]["LOCAL_CLUSTER"] = local_cluster_engine
    engines["TRANSPILER"]["CLUSTER"] = cluster_engine
    engines["PSLIB"]["SINGLE"] = local_mpi_engine
    engines["PSLIB"]["LOCAL_CLUSTER"] = local_mpi_engine
    engines["PSLIB"]["CLUSTER"] = cluster_mpi_engine
T
tangwei 已提交
48

T
tangwei 已提交
49

X
fix  
xjqbest 已提交
50
def get_inters_from_yaml(file, filters):
T
tangwei 已提交
51 52 53 54 55 56
    with open(file, 'r') as rb:
        _envs = yaml.load(rb.read(), Loader=yaml.FullLoader)

    flattens = envs.flatten_environs(_envs)
    inters = {}
    for k, v in flattens.items():
X
fix  
xjqbest 已提交
57 58 59
        for f in filters:
            if k.startswith(f):
                inters[k] = v
T
tangwei 已提交
60
    return inters
T
tangwei 已提交
61 62


X
fix  
xjqbest 已提交
63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95
def get_all_inters_from_yaml(file, filters):
    with open(file, 'r') as rb:
        _envs = yaml.load(rb.read(), Loader=yaml.FullLoader)
    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)
            elif (k == "dataset" or k == "phase" or
                  k == "runner") and isinstance(v, list):
                for i in v:
                    if i.get("name") is None:
                        raise ValueError("name must be in dataset list ", v)
                    nests = copy.deepcopy(namespace_nests)
                    nests.append(k)
                    nests.append(i["name"])
                    fatten_env_namespace(nests, i)
            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


C
chengmo 已提交
96
def get_engine(args):
T
tangwei 已提交
97
    transpiler = get_transpiler()
X
fix  
xjqbest 已提交
98 99 100
    with open(args.model, 'r') as rb:
        envs = yaml.load(rb.read(), Loader=yaml.FullLoader)
    run_extras = get_all_inters_from_yaml(args.model, ["train.", "runner."])
X
fix  
xjqbest 已提交
101 102 103

    engine = run_extras.get("train.engine", None)
    if engine is None:
X
fix  
xjqbest 已提交
104
        engine = run_extras.get("runner." + envs["mode"] + ".class", None)
X
fix  
xjqbest 已提交
105
    if engine is None:
X
fix  
xjqbest 已提交
106
        engine = "single_train"
T
tangwei 已提交
107 108
    engine = engine.upper()
    if engine not in engine_choices:
T
tangwei 已提交
109 110
        raise ValueError("train.engin can not be chosen in {}".format(
            engine_choices))
T
tangwei 已提交
111

T
tangwei 已提交
112
    print("engines: \n{}".format(engines))
T
tangwei 已提交
113

T
tangwei 已提交
114
    run_engine = engines[transpiler].get(engine, None)
T
tangwei 已提交
115

T
tangwei 已提交
116 117 118 119
    return run_engine


def get_transpiler():
T
tangwei 已提交
120
    FNULL = open(os.devnull, 'w')
T
tangwei 已提交
121 122 123 124
    cmd = [
        "python", "-c",
        "import paddle.fluid as fluid; fleet_ptr = fluid.core.Fleet(); [fleet_ptr.copy_table_by_feasign(10, 10, [2020, 1010])];"
    ]
T
tangwei 已提交
125 126 127
    proc = subprocess.Popen(cmd, stdout=FNULL, stderr=FNULL, cwd=os.getcwd())
    ret = proc.wait()
    if ret == -11:
T
tangwei 已提交
128
        return "PSLIB"
T
tangwei 已提交
129
    else:
T
tangwei 已提交
130
        return "TRANSPILER"
T
tangwei 已提交
131 132


T
tangwei 已提交
133 134 135
def set_runtime_envs(cluster_envs, engine_yaml):
    if cluster_envs is None:
        cluster_envs = {}
T
tangwei 已提交
136

T
tangwei 已提交
137
    engine_extras = get_inters_from_yaml(engine_yaml, "train.trainer.")
138 139
    if "train.trainer.threads" in engine_extras and "CPU_NUM" in cluster_envs:
        cluster_envs["CPU_NUM"] = engine_extras["train.trainer.threads"]
T
tangwei 已提交
140

T
tangwei 已提交
141
    envs.set_runtime_environs(cluster_envs)
142
    envs.set_runtime_environs(engine_extras)
T
fix bug  
tangwei 已提交
143 144 145

    need_print = {}
    for k, v in os.environ.items():
T
tangwei 已提交
146
        if k.startswith("train.trainer."):
T
fix bug  
tangwei 已提交
147 148 149
            need_print[k] = v

    print(envs.pretty_print_envs(need_print, ("Runtime Envs", "Value")))
T
tangwei 已提交
150 151


C
chengmo 已提交
152 153 154 155
def get_trainer_prefix(args):
    if model_name in custom_model:
        return model_name.upper()
    return ""
T
tangwei 已提交
156

C
chengmo 已提交
157

X
fix  
xjqbest 已提交
158
def single_train_engine(args):
C
chengmo 已提交
159
    trainer = get_trainer_prefix(args) + "SingleTrainer"
C
chengmo 已提交
160
    single_envs = {}
C
chengmo 已提交
161
    single_envs["train.trainer.trainer"] = trainer
C
chengmo 已提交
162
    single_envs["train.trainer.threads"] = "2"
X
fix  
xjqbest 已提交
163
    single_envs["train.trainer.engine"] = "single_train"
C
chengmo 已提交
164
    single_envs["train.trainer.platform"] = envs.get_platform()
C
chengmo 已提交
165
    print("use {} engine to run model: {}".format(trainer, args.model))
X
fix  
xjqbest 已提交
166 167 168
    set_runtime_envs(single_envs, args.model)
    trainer = TrainerFactory.create(args.model)
    return trainer
X
fix  
xjqbest 已提交
169

X
fix  
xjqbest 已提交
170 171 172 173 174 175 176 177 178

def single_infer_engine(args):
    trainer = get_trainer_prefix(args) + "SingleInfer"
    single_envs = {}
    single_envs["train.trainer.trainer"] = trainer
    single_envs["train.trainer.threads"] = "2"
    single_envs["train.trainer.engine"] = "single_infer"
    single_envs["train.trainer.platform"] = envs.get_platform()
    print("use {} engine to run model: {}".format(trainer, args.model))
X
fix  
xjqbest 已提交
179 180 181
    set_runtime_envs(single_envs, args.model)
    trainer = TrainerFactory.create(args.model)
    return trainer
C
chengmo 已提交
182

X
fix  
xjqbest 已提交
183

T
tangwei 已提交
184
def cluster_engine(args):
T
tangwei 已提交
185 186
    def update_workspace(cluster_envs):
        workspace = cluster_envs.get("engine_workspace", None)
T
tangwei 已提交
187

T
tangwei 已提交
188 189
        if not workspace:
            return
T
tangwei 已提交
190
        path = envs.path_adapter(workspace)
T
tangwei 已提交
191 192 193
        for name, value in cluster_envs.items():
            if isinstance(value, str):
                value = value.replace("{workspace}", path)
T
tangwei 已提交
194
                value = envs.windows_path_converter(value)
T
tangwei 已提交
195 196 197
                cluster_envs[name] = value

    def master():
T
tangwei 已提交
198
        role = "MASTER"
199
        from paddlerec.core.engine.cluster.cluster import ClusterEngine
T
tangwei 已提交
200 201 202 203
        with open(args.backend, 'r') as rb:
            _envs = yaml.load(rb.read(), Loader=yaml.FullLoader)

        flattens = envs.flatten_environs(_envs, "_")
T
tangwei 已提交
204
        flattens["engine_role"] = role
T
tangwei 已提交
205
        flattens["engine_run_config"] = args.model
T
tangwei 已提交
206 207 208 209
        flattens["engine_temp_path"] = tempfile.mkdtemp()
        update_workspace(flattens)

        envs.set_runtime_environs(flattens)
T
tangwei 已提交
210 211
        print(envs.pretty_print_envs(flattens, ("Submit Runtime Envs", "Value"
                                                )))
T
tangwei 已提交
212 213 214 215 216

        launch = ClusterEngine(None, args.model)
        return launch

    def worker():
T
tangwei 已提交
217
        role = "WORKER"
T
tangwei 已提交
218 219 220 221
        trainer = get_trainer_prefix(args) + "ClusterTrainer"
        cluster_envs = {}
        cluster_envs["train.trainer.trainer"] = trainer
        cluster_envs["train.trainer.engine"] = "cluster"
T
tangwei 已提交
222 223
        cluster_envs["train.trainer.threads"] = envs.get_runtime_environ(
            "CPU_NUM")
T
tangwei 已提交
224
        cluster_envs["train.trainer.platform"] = envs.get_platform()
C
chengmo 已提交
225 226
        print("launch {} engine with cluster to with model: {}".format(
            trainer, args.model))
T
tangwei 已提交
227
        set_runtime_envs(cluster_envs, args.model)
T
tangwei 已提交
228

T
bug fix  
tangwei12 已提交
229 230
        trainer = TrainerFactory.create(args.model)
        return trainer
T
tangwei 已提交
231

T
tangwei 已提交
232 233 234
    role = os.getenv("PADDLE_PADDLEREC_ROLE", "MASTER")

    if role == "WORKER":
T
tangwei 已提交
235 236 237
        return worker()
    else:
        return master()
C
chengmo 已提交
238 239


T
tangwei 已提交
240
def cluster_mpi_engine(args):
T
tangwei 已提交
241 242
    print("launch cluster engine with cluster to run model: {}".format(
        args.model))
T
tangwei 已提交
243

T
fix bug  
tangwei 已提交
244
    cluster_envs = {}
T
tangwei 已提交
245
    cluster_envs["train.trainer.trainer"] = "CtrCodingTrainer"
T
tangwei 已提交
246
    cluster_envs["train.trainer.platform"] = envs.get_platform()
T
tangwei 已提交
247

T
tangwei 已提交
248
    set_runtime_envs(cluster_envs, args.model)
T
tangwei 已提交
249

T
tangwei 已提交
250 251 252 253 254
    trainer = TrainerFactory.create(args.model)
    return trainer


def local_cluster_engine(args):
255
    from paddlerec.core.engine.local_cluster import LocalClusterEngine
C
chengmo 已提交
256

C
chengmo 已提交
257
    trainer = get_trainer_prefix(args) + "ClusterTrainer"
C
chengmo 已提交
258 259 260
    cluster_envs = {}
    cluster_envs["server_num"] = 1
    cluster_envs["worker_num"] = 1
C
chengmo 已提交
261
    cluster_envs["start_port"] = envs.find_free_port()
C
chengmo 已提交
262
    cluster_envs["log_dir"] = "logs"
C
chengmo 已提交
263
    cluster_envs["train.trainer.trainer"] = trainer
C
chengmo 已提交
264 265 266 267 268 269
    cluster_envs["train.trainer.strategy"] = "async"
    cluster_envs["train.trainer.threads"] = "2"
    cluster_envs["train.trainer.engine"] = "local_cluster"
    cluster_envs["train.trainer.platform"] = envs.get_platform()

    cluster_envs["CPU_NUM"] = "2"
T
tangwei 已提交
270 271
    print("launch {} engine with cluster to run model: {}".format(trainer,
                                                                  args.model))
C
chengmo 已提交
272 273 274 275 276 277

    set_runtime_envs(cluster_envs, args.model)
    launch = LocalClusterEngine(cluster_envs, args.model)
    return launch


T
tangwei 已提交
278
def local_mpi_engine(args):
T
tangwei 已提交
279 280
    print("launch cluster engine with cluster to run model: {}".format(
        args.model))
281
    from paddlerec.core.engine.local_mpi import LocalMPIEngine
T
tangwei 已提交
282

T
tangwei 已提交
283 284
    print("use 1X1 MPI ClusterTraining at localhost to run model: {}".format(
        args.model))
T
tangwei 已提交
285

T
tangwei 已提交
286 287 288
    mpi = util.run_which("mpirun")
    if not mpi:
        raise RuntimeError("can not find mpirun, please check environment")
T
fix bug  
tangwei 已提交
289 290
    cluster_envs = {}
    cluster_envs["mpirun"] = mpi
T
tangwei 已提交
291
    cluster_envs["train.trainer.trainer"] = "CtrCodingTrainer"
T
fix bug  
tangwei 已提交
292
    cluster_envs["log_dir"] = "logs"
T
tangwei 已提交
293
    cluster_envs["train.trainer.engine"] = "local_cluster"
T
tangwei 已提交
294 295

    cluster_envs["train.trainer.platform"] = envs.get_platform()
T
tangwei 已提交
296

T
tangwei 已提交
297
    set_runtime_envs(cluster_envs, args.model)
T
tangwei 已提交
298 299 300 301
    launch = LocalMPIEngine(cluster_envs, args.model)
    return launch


T
tangwei 已提交
302
def get_abs_model(model):
303
    if model.startswith("paddlerec."):
T
tangwei 已提交
304
        dir = envs.path_adapter(model)
T
tangwei 已提交
305
        path = os.path.join(dir, "config.yaml")
T
tangwei 已提交
306 307 308 309 310 311 312
    else:
        if not os.path.isfile(model):
            raise IOError("model config: {} invalid".format(model))
        path = model
    return path


T
tangwei 已提交
313
if __name__ == "__main__":
314
    parser = argparse.ArgumentParser(description='paddle-rec run')
T
tangwei 已提交
315
    parser.add_argument("-m", "--model", type=str)
T
tangwei 已提交
316
    parser.add_argument("-b", "--backend", type=str, default=None)
T
tangwei 已提交
317

T
tangwei 已提交
318 319 320
    abs_dir = os.path.dirname(os.path.abspath(__file__))
    envs.set_runtime_environs({"PACKAGE_BASE": abs_dir})

T
tangwei 已提交
321
    args = parser.parse_args()
T
tangwei 已提交
322

C
fix  
chengmo 已提交
323
    model_name = args.model.split('.')[-1]
T
tangwei 已提交
324
    args.model = get_abs_model(args.model)
T
tangwei 已提交
325
    engine_registry()
T
tangwei 已提交
326

C
chengmo 已提交
327
    which_engine = get_engine(args)
T
tangwei 已提交
328 329
    engine = which_engine(args)
    engine.run()