Merge pull request #367 from frankwhzhang/dy2static

add dygraph to static

Author: seemingwang

models/rank/dnn/config.yaml

@@ -22,7 +22,7 @@ runner:
   use_gpu: False
   use_auc: True
   train_batch_size: 2
-  epochs: 3
+  epochs: 1
   print_interval: 2
   model_save_path: "output_model_dnn"
   infer_batch_size: 2

models/rank/dnn/config_to_static.yaml

@@ -0,0 +1,55 @@
+# 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.
+
+# workspace
+#workspace: "models/rank/dnn"
+
+
+runner:
+  train_data_dir: "data/sample_data/train"
+  train_reader_path: "criteo_reader" # importlib format
+  model_init_path: "output_model_dnn/0" # model_init
+  use_gpu: False
+  use_auc: True
+  train_batch_size: 2
+  epochs: 1
+  print_interval: 2
+  model_save_path: "output_model_dnn2" # save path
+  infer_batch_size: 2
+  infer_reader_path: "criteo_reader" # importlib format
+  test_data_dir: "data/sample_data/train"
+  infer_load_path: "output_model_dnn"
+  infer_start_epoch: 0
+  infer_end_epoch: 3
+
+  # distribute_config
+  sync_mode: "async"
+  split_file_list: False
+  thread_num: 1
+
+
+# hyper parameters of user-defined network
+hyper_parameters:
+  # optimizer config
+  optimizer:
+    class: Adam
+    learning_rate: 0.001
+    strategy: async
+  # user-defined <key, value> pairs
+  sparse_inputs_slots: 27
+  sparse_feature_number: 1000001
+  sparse_feature_dim: 9
+  dense_input_dim: 13
+  fc_sizes: [512, 256, 128, 32]
+  distributed_embedding: 0

tools/to_static.py

@@ -0,0 +1,98 @@
+# 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.
+
+#
+# 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 paddle
+import os
+import paddle.nn as nn
+import time
+import logging
+import sys
+import importlib
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+#sys.path.append(__dir__)
+sys.path.append(os.path.abspath(os.path.join(__dir__, '..')))
+
+from utils.utils_single import load_yaml, load_dy_model_class, get_abs_model, create_data_loader
+from utils.save_load import load_model, save_model, save_jit_model
+from paddle.io import DistributedBatchSampler, DataLoader
+import argparse
+
+logging.basicConfig(
+    format='%(asctime)s - %(levelname)s - %(message)s', level=logging.INFO)
+logger = logging.getLogger(__name__)
+
+
+def parse_args():
+    parser = argparse.ArgumentParser(description='paddle-rec run')
+    parser.add_argument("-m", "--config_yaml", type=str)
+    args = parser.parse_args()
+    args.abs_dir = os.path.dirname(os.path.abspath(args.config_yaml))
+    args.config_yaml = get_abs_model(args.config_yaml)
+    return args
+
+
+def main(args):
+    paddle.seed(12345)
+    # load config
+    config = load_yaml(args.config_yaml)
+    dy_model_class = load_dy_model_class(args.abs_dir)
+    config["config_abs_dir"] = args.abs_dir
+    # tools.vars
+    use_gpu = config.get("runner.use_gpu", True)
+    train_data_dir = config.get("runner.train_data_dir", None)
+    epochs = config.get("runner.epochs", None)
+    print_interval = config.get("runner.print_interval", None)
+    model_save_path = config.get("runner.model_save_path", "model_output")
+    model_init_path = config.get("runner.model_init_path", None)
+
+    logger.info("**************common.configs**********")
+    logger.info(
+        "use_gpu: {}, train_data_dir: {}, epochs: {}, print_interval: {}, model_save_path: {}".
+        format(use_gpu, train_data_dir, epochs, print_interval,
+               model_save_path))
+    logger.info("**************common.configs**********")
+
+    place = paddle.set_device('gpu' if use_gpu else 'cpu')
+
+    dy_model = dy_model_class.create_model(config)
+
+    load_model(model_init_path, dy_model)
+    # example dnn model forward
+    dy_model = paddle.jit.to_static(
+        dy_model,
+        input_spec=[[
+            paddle.static.InputSpec(
+                shape=[None, 1], dtype='int64') for jj in range(26)
+        ], paddle.static.InputSpec(
+            shape=[None, 13], dtype='float32')])
+    save_jit_model(dy_model, model_save_path, prefix='tostatic')
+
+
+if __name__ == '__main__':
+    args = parse_args()
+    main(args)

tools/utils/save_load.py

@@ -30,6 +30,14 @@ def save_model(net, optimizer, model_path, epoch_id, prefix='rec'):
     logger.info("Already save model in {}".format(model_path))
 
 
+def save_jit_model(net, model_path, prefix='tostatic'):
+    _mkdir_if_not_exist(model_path)
+    model_prefix = os.path.join(model_path, prefix)
+    #paddle.save(net.state_dict(), model_prefix + ".pdparams")
+    paddle.jit.save(net, model_prefix)
+    logger.info("Already save jit model in {}".format(model_path))
+
+
 def load_model(model_path, net, prefix='rec'):
     logger.info("start load model from {}".format(model_path))
     model_prefix = os.path.join(model_path, prefix)