def save_serving_model(save_dir, exe, feed_vars, test_fetches, infer_prog):
feed_var_names = [var.name for var in feed_vars.values()]
fetch_list = sorted(test_fetches.items(), key=lambda i: i[0])
target_vars = [var[1] for var in fetch_list]
feed_var_names = prune_feed_vars(feed_var_names, target_vars, infer_prog)
output_dir = save_dir
serving_client = os.path.join(output_dir, 'serving_client')
serving_server = os.path.join(output_dir, 'serving_server')
logger.info(
"Export serving model to {}, client side: {}, server side: {}. input: {}, output: "
"{}...".format(output_dir, serving_client, serving_server,
feed_var_names, [str(var.name) for var in target_vars]))
feed_dict = {x: infer_prog.global_block().var(x) for x in feed_var_names}
fetch_dict = {x.name: x for x in target_vars}
import paddle_serving_client.io as serving_io
serving_client = os.path.join(save_dir, 'serving_client')
serving_server = os.path.join(save_dir, 'serving_server')
serving_io.save_model(
client_config_folder=serving_client,
server_model_folder=serving_server,
feed_var_dict=feed_dict,
fetch_var_dict=fetch_dict,
main_program=infer_prog)
shutil.copy(serving_client+'/serving_client_conf.prototxt', save_dir+'/serving_server_conf.prototxt')
shutil.rmtree('-p')
def main():
args = parse_args()
model = architectures.__dict__[args.model]()
place = fluid.CPUPlace()
exe = fluid.Executor(place)
startup_prog = fluid.Program()
infer_prog = fluid.Program()
with fluid.program_guard(infer_prog, startup_prog):
with fluid.unique_name.guard():
image = create_input(args.img_size)
out = create_model(args, model, image, class_dim=args.class_dim)
infer_prog = infer_prog.clone(for_test=True)
fluid.load(program=infer_prog,
model_path=args.pretrained_model,
executor=exe)
model_path = os.path.join(args.output_path, "ppcls_model")
conf_path = os.path.join(args.output_path, "ppcls_client_conf")
serving_io.save_model(model_path, conf_path, {"image": image},
{"prediction": out}, infer_prog)
def save_serving_model(self, model_path, client_conf_path):
feed_vars = {}
target_vars = {}
for target in self._target_names:
tmp_target = self._main_program.block(0)._find_var_recursive(
target)
target_vars[target] = tmp_target
for feed in self._feed_names:
tmp_feed = self._main_program.block(0)._find_var_recursive(feed)
feed_vars[feed] = tmp_feed
serving_io.save_model(model_path, client_conf_path, feed_vars,
target_vars, self._main_program)
def export_serving_model(args):
"""
Export PaddlePaddle inference model for prediction depolyment and serving.
"""
print("Exporting serving model...")
startup_prog = fluid.Program()
infer_prog = fluid.Program()
image, logit_out = build_model(infer_prog,
startup_prog,
phase=ModelPhase.PREDICT)
# Use CPU for exporting inference model instead of GPU
place = fluid.CPUPlace()
exe = fluid.Executor(place)
exe.run(startup_prog)
infer_prog = infer_prog.clone(for_test=True)
if os.path.exists(cfg.TEST.TEST_MODEL):
print('load test model:', cfg.TEST.TEST_MODEL)
try:
fluid.load(infer_prog, os.path.join(cfg.TEST.TEST_MODEL, 'model'),
exe)
except:
fluid.io.load_params(exe,
cfg.TEST.TEST_MODEL,
main_program=infer_prog)
else:
print("TEST.TEST_MODEL diretory is empty!")
exit(-1)
from paddle_serving_client.io import save_model
save_model(
cfg.FREEZE.SAVE_DIR + "/serving_server",
cfg.FREEZE.SAVE_DIR + "/serving_client",
{image.name: image},
{logit_out.name: logit_out},
infer_prog,
)
print("Serving model exported!")
print("Exporting serving model config...")
deploy_cfg_path = export_inference_config()
print("Serving model saved : [%s]" % (deploy_cfg_path))
def save_serving_model(FLAGS, exe, feed_vars, test_fetches, infer_prog):
cfg_name = os.path.basename(FLAGS.config).split('.')[0]
save_dir = os.path.join(FLAGS.output_dir, cfg_name)
feed_var_names = [var.name for var in feed_vars.values()]
fetch_list = sorted(test_fetches.items(), key=lambda i: i[0])
target_vars = [var[1] for var in fetch_list]
feed_var_names = prune_feed_vars(feed_var_names, target_vars, infer_prog)
serving_client = os.path.join(FLAGS.output_dir, 'serving_client')
serving_server = os.path.join(FLAGS.output_dir, 'serving_server')
logger.info(
"Export serving model to {}, client side: {}, server side: {}. input: {}, output: "
"{}...".format(FLAGS.output_dir, serving_client, serving_server,
feed_var_names, [str(var.name) for var in target_vars]))
feed_dict = {x: infer_prog.global_block().var(x) for x in feed_var_names}
fetch_dict = {x.name: x for x in target_vars}
import paddle_serving_client.io as serving_io
serving_client = os.path.join(save_dir, 'serving_client')
serving_server = os.path.join(save_dir, 'serving_server')
serving_io.save_model(serving_client, serving_server, feed_dict,
fetch_dict, infer_prog)
label = fluid.layers.data(name="label", shape=[1], dtype="int64")
dataset = fluid.DatasetFactory().create_dataset()
filelist = ["train_data/%s" % x for x in os.listdir("train_data")]
dataset.set_use_var([data, label])
pipe_command = "python imdb_reader.py"
dataset.set_pipe_command(pipe_command)
dataset.set_batch_size(128)
dataset.set_filelist(filelist)
dataset.set_thread(10)
avg_cost, acc, prediction = lstm_net(data, label, dict_dim)
optimizer = fluid.optimizer.SGD(learning_rate=0.01)
optimizer.minimize(avg_cost)
exe = fluid.Executor(fluid.CPUPlace())
exe.run(fluid.default_startup_program())
epochs = 6
import paddle_serving_client.io as serving_io
for i in range(epochs):
exe.train_from_dataset(program=fluid.default_main_program(),
dataset=dataset,
debug=False)
logger.info("TRAIN --> pass: {}".format(i))
if i == 5:
serving_io.save_model("{}_model".format(model_name),
"{}_client_conf".format(model_name),
{"words": data}, {"prediction": prediction},
fluid.default_main_program())
def train():
args = parse_args()
sparse_only = args.sparse_only
if not os.path.isdir(args.model_output_dir):
os.mkdir(args.model_output_dir)
dense_input = fluid.layers.data(name="dense_input",
shape=[dense_feature_dim],
dtype='float32')
sparse_input_ids = [
fluid.layers.data(name="C" + str(i),
shape=[1],
lod_level=1,
dtype="int64") for i in range(1, 27)
]
label = fluid.layers.data(name='label', shape=[1], dtype='int64')
#nn_input = None if sparse_only else dense_input
nn_input = dense_input
predict_y, loss, auc_var, batch_auc_var, infer_vars = dnn_model(
nn_input, sparse_input_ids, label, args.embedding_size,
args.sparse_feature_dim)
optimizer = fluid.optimizer.SGD(learning_rate=1e-4)
optimizer.minimize(loss)
exe = fluid.Executor(fluid.CPUPlace())
exe.run(fluid.default_startup_program())
dataset = fluid.DatasetFactory().create_dataset("InMemoryDataset")
dataset.set_use_var([dense_input] + sparse_input_ids + [label])
python_executable = "python3.6"
pipe_command = "{} criteo_reader.py {}".format(python_executable,
args.sparse_feature_dim)
dataset.set_pipe_command(pipe_command)
dataset.set_batch_size(128)
thread_num = 10
dataset.set_thread(thread_num)
whole_filelist = [
"raw_data/part-%d" % x for x in range(len(os.listdir("raw_data")))
]
print(whole_filelist)
dataset.set_filelist(whole_filelist[:100])
dataset.load_into_memory()
fluid.layers.Print(auc_var)
epochs = 1
for i in range(epochs):
exe.train_from_dataset(program=fluid.default_main_program(),
dataset=dataset,
debug=True)
print("epoch {} finished".format(i))
import paddle_serving_client.io as server_io
feed_var_dict = {}
feed_var_dict['dense_input'] = dense_input
for i, sparse in enumerate(sparse_input_ids):
feed_var_dict["embedding_{}.tmp_0".format(i)] = sparse
fetch_var_dict = {"prob": predict_y}
feed_kv_dict = {}
feed_kv_dict['dense_input'] = dense_input
for i, emb in enumerate(infer_vars):
feed_kv_dict["embedding_{}.tmp_0".format(i)] = emb
fetch_var_dict = {"prob": predict_y}
server_io.save_model("ctr_serving_model", "ctr_client_conf", feed_var_dict,
fetch_var_dict, fluid.default_main_program())
server_io.save_model("ctr_serving_model_kv", "ctr_client_conf_kv",
feed_kv_dict, fetch_var_dict,
fluid.default_main_program())
model_name = "bert_chinese_L-12_H-768_A-12"
module = hub.Module(name=model_name)
inputs, outputs, program = module.context(trainable=True,
max_seq_len=int(sys.argv[1]))
place = fluid.core_avx.CPUPlace()
exe = fluid.Executor(place)
input_ids = inputs["input_ids"]
position_ids = inputs["position_ids"]
segment_ids = inputs["segment_ids"]
input_mask = inputs["input_mask"]
pooled_output = outputs["pooled_output"]
sequence_output = outputs["sequence_output"]
feed_var_names = [
input_ids.name, position_ids.name, segment_ids.name, input_mask.name
]
target_vars = [pooled_output, sequence_output]
serving_io.save_model(
"bert_seq{}_model".format(sys.argv[1]),
"bert_seq{}_client".format(sys.argv[1]), {
"input_ids": input_ids,
"position_ids": position_ids,
"segment_ids": segment_ids,
"input_mask": input_mask,
}, {
"pooled_output": pooled_output,
"sequence_output": sequence_output
}, program)
batch_size=16)
test_reader = paddle.batch(paddle.reader.shuffle(
paddle.dataset.uci_housing.test(), buf_size=500),
batch_size=16)
x = fluid.data(name='x', shape=[None, 13], dtype='float32')
y = fluid.data(name='y', shape=[None, 1], dtype='float32')
y_predict = fluid.layers.fc(input=x, size=1, act=None)
cost = fluid.layers.square_error_cost(input=y_predict, label=y)
avg_loss = fluid.layers.mean(cost)
sgd_optimizer = fluid.optimizer.SGD(learning_rate=0.01)
sgd_optimizer.minimize(avg_loss)
place = fluid.CPUPlace()
feeder = fluid.DataFeeder(place=place, feed_list=[x, y])
exe = fluid.Executor(place)
exe.run(fluid.default_startup_program())
import paddle_serving_client.io as serving_io
for pass_id in range(30):
for data_train in train_reader():
avg_loss_value, = exe.run(fluid.default_main_program(),
feed=feeder.feed(data_train),
fetch_list=[avg_loss])
serving_io.save_model("uci_housing_model", "uci_housing_client", {"x": x},
{"price": y_predict}, fluid.default_main_program())