Skip to content

P
Paddle

BaiXuePrincess / Paddle
与 Fork 源项目一致

Fork自 PaddlePaddle / Paddle

通知 1
Star 1
Fork 0
P
Paddle
提交 b853ac8c 编写于 作者: T typhoonzero
浏览文件
操作

Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into fix_test_sendrecv_portbind

上级 00f8e63b d2f9e193
展开全部 隐藏空白更改
内联 并排
Showing with 7570 addition and 314 deletion
benchmark/cluster/README.md
浏览文件 @ b853ac8c
......@@ -36,11 +36,41 @@
- Trainer Count: 100
- Metrics: mini-batch / sec
| Batch Size | 32 | 64 | 128 | 256 |
| -- | -- | -- | -- | -- |
| PaddlePaddle Fluid | - | - | - | - |
| PaddlePaddle v2 | - | - | - | - |
| TensorFlow | - | - | - | - |
<table>
<thead>
<tr>
<th>Batch Size </th>
<th> 32</th>
<th>64</th>
<th>128 </th>
<th>256</th>
</tr>
</thead>
<tbody>
<tr>
<td> PaddlePaddle Fluid</td>
<td>-</td>
<td>- </td>
<td>- </td>
<td>- </td>
</tr>
<tr>
<td>PaddlePaddle v2 </td>
<td>- </td>
<td>- </td>
<td>- </td>
<td>- </td>
</tr>
<tr>
<td>TensorFlow </td>
<td>- </td>
<td>- </td>
<td>- </td>
<td>- </td>
</tr>
</tbody>
</table>
### Measure the Performance for Different PServer Count
......@@ -48,11 +78,41 @@
- Batch Size: 64
- Metrics: mini-batch / sec
| PServer Count | 10 | 20 | 40 | 60 |
| -- | -- | -- | -- | -- |
| PaddlePaddle Fluid | - | - | - | - |
| PaddlePaddle v2 | - | - | - | - |
| TensorFlow | - | - | - | - |
<table>
<thead>
<tr>
<th>PServer Count </th>
<th>10</th>
<th>20</th>
<th>40 </th>
<th>60</th>
</tr>
</thead>
<tbody>
<tr>
<td> PaddlePaddle Fluid</td>
<td>-</td>
<td>- </td>
<td>- </td>
<td>- </td>
</tr>
<tr>
<td>PaddlePaddle v2 </td>
<td>- </td>
<td>- </td>
<td>- </td>
<td>- </td>
</tr>
<tr>
<td>TensorFlow </td>
<td>- </td>
<td>- </td>
<td>- </td>
<td>- </td>
</tr>
</tbody>
</table>
### Measure Parallel Efficiency By Increasing Trainer Count
......@@ -67,11 +127,69 @@ The parallel efficiency is:
$E = \div(S, N)$
| Trainer Counter | 1 | 10 | 20 | 30 | 40 | 50 | 60 | 70 | 80 | 90 | 100 |
| -- | -- | -- | -- | -- | -- | -- | -- | -- | -- | -- | -- |
| PaddlePaddle Fluid | - | - | - | - | - | - | - | - | - | - | - |
| PaddlePaddle v2 | - | - | - | - | - | - | - | - | - | - | - | - |
| TensorFlow | - | - | - | - | - | - | - | - | - | - | - | - | - |
<table>
<thead>
<tr>
<th>Trainer Counter </th>
<th>1</th>
<th>10</th>
<th>20 </th>
<th>30</th>
<th>40</th>
<th>50</th>
<th>60 </th>
<th>70</th>
<th>80</th>
<th>90</th>
<th>100 </th>
</tr>
</thead>
<tbody>
<tr>
<td> PaddlePaddle Fluid</td>
<td>-</td>
<td>- </td>
<td>- </td>
<td>- </td>
<td>-</td>
<td>- </td>
<td>- </td>
<td>- </td>
<td>-</td>
<td>- </td>
<td>- </td>
</tr>
<tr>
<td>PaddlePaddle v2 </td>
<td>- </td>
<td>- </td>
<td>- </td>
<td>- </td>
<td>-</td>
<td>- </td>
<td>- </td>
<td>- </td>
<td>-</td>
<td>- </td>
<td>- </td>
</tr>
<tr>
<td>TensorFlow </td>
<td>- </td>
<td>- </td>
<td>- </td>
<td>- </td>
<td>-</td>
<td>- </td>
<td>- </td>
<td>- </td>
<td>-</td>
<td>- </td>
<td>- </td>
</tr>
</tbody>
</table>
## Reproduce the benchmark
......
benchmark/cluster/vgg16/README.md
浏览文件 @ b853ac8c
......@@ -16,11 +16,41 @@ Setting environment variable: `MKL_NUM_THREADS=1`.
- Metrics: samples / sec
| Batch Size | 32 | 64 | 128 | 256 |
| -- | -- | -- | -- | -- |
| PaddlePaddle Fluid | 15.44 | 16.32 | 16.74 | 16.79 |
| PaddlePaddle v2 | 15.97 | 17.04 | 17.60 | 17.83 |
| TensorFlow | 9.09 | 9.10 | 9.24 | 8.66 |
<table>
<thead>
<tr>
<th>Batch Size </th>
<th> 32</th>
<th>64</th>
<th>128 </th>
<th>256</th>
</tr>
</thead>
<tbody>
<tr>
<td> PaddlePaddle Fluid</td>
<td> 15.44 </td>
<td> 16.32 </td>
<td> 16.74 </td>
<td> 16.79 </td>
</tr>
<tr>
<td>PaddlePaddle v2 </td>
<td> 15.97 </td>
<td> 17.04 </td>
<td> 17.60 </td>
<td> 17.83 </td>
</tr>
<tr>
<td>TensorFlow </td>
<td> 9.09 </td>
<td> 9.10 </td>
<td> 9.24 </td>
<td> 8.66 </td>
</tr>
</tbody>
</table>
### Different Batch Size
......@@ -28,12 +58,40 @@ Setting environment variable: `MKL_NUM_THREADS=1`.
- Trainer Count: 20
- Metrics: samples / sec
| Batch Size | 32 | 64 | 128 | 256 |
| -- | -- | -- | -- | -- |
| PaddlePaddle Fluid | 190.20 | 222.15 | 247.40 | 258.18 |
| PaddlePaddle v2 | 170.96 | 233.71 | 256.14 | 329.23 |
| TensorFlow | - | - | - | - |
<table>
<thead>
<tr>
<th>Batch Size </th>
<th> 32</th>
<th>64</th>
<th>128 </th>
<th>256</th>
</tr>
</thead>
<tbody>
<tr>
<td> PaddlePaddle Fluid</td>
<td> 190.20 </td>
<td> 222.15 </td>
<td> 247.40 </td>
<td> 258.18 </td>
</tr>
<tr>
<td>PaddlePaddle v2 </td>
<td> 170.96 </td>
<td> 233.71 </td>
<td> 256.14 </td>
<td> 329.23 </td>
</tr>
<tr>
<td>TensorFlow </td>
<td> - </td>
<td> - </td>
<td> - </td>
<td> - </td>
</tr>
</tbody>
</table>
### Accelerate Rate
......@@ -41,11 +99,41 @@ Setting environment variable: `MKL_NUM_THREADS=1`.
- Batch Size: 128
- Metrics: samples / sec
| Trainer Count | 20 | 40 | 80 | 100 |
| -- | -- | -- | -- | -- |
| PaddlePaddle Fluid | 263.29 (78.64%) | 518.80 (77.47%) | 836.26 (62.44%) | 1019.29 (60.89%) |
| PaddlePaddle v2 (need more tests) | 326.85 (92.85%) | 534.58 (75.93%) | 853.30 (60.60%) | 1041.99 (59.20%) |
| TensorFlow | - | - | - | - |
<table>
<thead>
<tr>
<th>Trainer Count </th>
<th>20</th>
<th>40</th>
<th>80</th>
<th>100</th>
</tr>
</thead>
<tbody>
<tr>
<td> PaddlePaddle Fluid</td>
<td> 263.29 (78.64%) </td>
<td> 518.80 (77.47%) </td>
<td> 836.26 (62.44%) </td>
<td> 1019.29 (60.89%) </td>
</tr>
<tr>
<td>PaddlePaddle v2 (need more tests) </td>
<td> 326.85 (92.85%) </td>
<td> 534.58 (75.93%) </td>
<td> 853.30 (60.60%) </td>
<td> 1041.99 (59.20%) </td>
</tr>
<tr>
<td>TensorFlow </td>
<td> - </td>
<td> - </td>
<td> - </td>
<td> - </td>
</tr>
</tbody>
</table>
### Different Pserver Count
......@@ -53,11 +141,41 @@ Setting environment variable: `MKL_NUM_THREADS=1`.
- Batch Size: 128
- Metrics: samples/ sec
| PServer Count | 3 | 6 |10 | 20 |
| -- | -- | -- | -- | -- |
| PaddlePaddle Fluid(should fix in next PR) | 589.1 | 592.6 | 656.4 | 655.8 |
| PaddlePaddle v2 | 593.4 | 791.3 | 729.7 | 821.7 |
| TensorFlow | - | - | - | - |
<table>
<thead>
<tr>
<th>PServer Count </th>
<th>3</th>
<th>6</th>
<th>10</th>
<th>20</th>
</tr>
</thead>
<tbody>
<tr>
<td> PaddlePaddle Fluid(should fix in next PR) </td>
<td> 589.1 </td>
<td> 592.6 </td>
<td> 656.4 </td>
<td> 655.8 </td>
</tr>
<tr>
<td>PaddlePaddle v2 (need more tests) </td>
<td> 593.4 </td>
<td> 791.3 </td>
<td> 729.7 </td>
<td> 821.7 </td>
</tr>
<tr>
<td>TensorFlow </td>
<td> - </td>
<td> - </td>
<td> - </td>
<td> - </td>
</tr>
</tbody>
</table>
*The performance gap between Fuild and v2 comes from the network interference.*
......
benchmark/fluid/machine_translation.py 0 → 100644
浏览文件 @ b853ac8c
# 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.
"""seq2seq model for fluid."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
import argparse
import time
import distutils.util
import paddle.v2 as paddle
import paddle.fluid as fluid
import paddle.fluid.core as core
import paddle.fluid.framework as framework
from paddle.fluid.executor import Executor
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument(
"--embedding_dim",
type=int,
default=512,
help="The dimension of embedding table. (default: %(default)d)")
parser.add_argument(
"--encoder_size",
type=int,
default=512,
help="The size of encoder bi-rnn unit. (default: %(default)d)")
parser.add_argument(
"--decoder_size",
type=int,
default=512,
help="The size of decoder rnn unit. (default: %(default)d)")
parser.add_argument(
"--batch_size",
type=int,
default=16,
help="The sequence number of a mini-batch data. (default: %(default)d)")
parser.add_argument(
"--dict_size",
type=int,
default=30000,
help="The dictionary capacity. Dictionaries of source sequence and "
"target dictionary have same capacity. (default: %(default)d)")
parser.add_argument(
"--pass_num",
type=int,
default=2,
help="The pass number to train. (default: %(default)d)")
parser.add_argument(
"--learning_rate",
type=float,
default=0.0002,
help="Learning rate used to train the model. (default: %(default)f)")
parser.add_argument(
"--infer_only", action='store_true', help="If set, run forward only.")
parser.add_argument(
"--beam_size",
type=int,
default=3,
help="The width for beam searching. (default: %(default)d)")
parser.add_argument(
"--use_gpu",
type=distutils.util.strtobool,
default=True,
help="Whether to use gpu. (default: %(default)d)")
parser.add_argument(
"--max_length",
type=int,
default=250,
help="The maximum length of sequence when doing generation. "
"(default: %(default)d)")
def lstm_step(x_t, hidden_t_prev, cell_t_prev, size):
def linear(inputs):
return fluid.layers.fc(input=inputs, size=size, bias_attr=True)
forget_gate = fluid.layers.sigmoid(x=linear([hidden_t_prev, x_t]))
input_gate = fluid.layers.sigmoid(x=linear([hidden_t_prev, x_t]))
output_gate = fluid.layers.sigmoid(x=linear([hidden_t_prev, x_t]))
cell_tilde = fluid.layers.tanh(x=linear([hidden_t_prev, x_t]))
cell_t = fluid.layers.sums(input=[
fluid.layers.elementwise_mul(
x=forget_gate, y=cell_t_prev), fluid.layers.elementwise_mul(
x=input_gate, y=cell_tilde)
])
hidden_t = fluid.layers.elementwise_mul(
x=output_gate, y=fluid.layers.tanh(x=cell_t))
return hidden_t, cell_t
def seq_to_seq_net(embedding_dim, encoder_size, decoder_size, source_dict_dim,
target_dict_dim, is_generating, beam_size, max_length):
"""Construct a seq2seq network."""
def bi_lstm_encoder(input_seq, gate_size):
# Linear transformation part for input gate, output gate, forget gate
# and cell activation vectors need be done outside of dynamic_lstm.
# So the output size is 4 times of gate_size.
input_forward_proj = fluid.layers.fc(input=input_seq,
size=gate_size * 4,
act=None,
bias_attr=False)
forward, _ = fluid.layers.dynamic_lstm(
input=input_forward_proj, size=gate_size * 4, use_peepholes=False)
input_reversed_proj = fluid.layers.fc(input=input_seq,
size=gate_size * 4,
act=None,
bias_attr=False)
reversed, _ = fluid.layers.dynamic_lstm(
input=input_reversed_proj,
size=gate_size * 4,
is_reverse=True,
use_peepholes=False)
return forward, reversed
src_word_idx = fluid.layers.data(
name='source_sequence', shape=[1], dtype='int64', lod_level=1)
src_embedding = fluid.layers.embedding(
input=src_word_idx,
size=[source_dict_dim, embedding_dim],
dtype='float32')
src_forward, src_reversed = bi_lstm_encoder(
input_seq=src_embedding, gate_size=encoder_size)
encoded_vector = fluid.layers.concat(
input=[src_forward, src_reversed], axis=1)
encoded_proj = fluid.layers.fc(input=encoded_vector,
size=decoder_size,
bias_attr=False)
backward_first = fluid.layers.sequence_pool(
input=src_reversed, pool_type='first')
decoder_boot = fluid.layers.fc(input=backward_first,
size=decoder_size,
bias_attr=False,
act='tanh')
def lstm_decoder_with_attention(target_embedding, encoder_vec, encoder_proj,
decoder_boot, decoder_size):
def simple_attention(encoder_vec, encoder_proj, decoder_state):
decoder_state_proj = fluid.layers.fc(input=decoder_state,
size=decoder_size,
bias_attr=False)
decoder_state_expand = fluid.layers.sequence_expand(
x=decoder_state_proj, y=encoder_proj)
concated = fluid.layers.concat(
input=[encoder_proj, decoder_state_expand], axis=1)
attention_weights = fluid.layers.fc(input=concated,
size=1,
act='tanh',
bias_attr=False)
attention_weights = fluid.layers.sequence_softmax(
input=attention_weights)
weigths_reshape = fluid.layers.reshape(
x=attention_weights, shape=[-1])
scaled = fluid.layers.elementwise_mul(
x=encoder_vec, y=weigths_reshape, axis=0)
context = fluid.layers.sequence_pool(input=scaled, pool_type='sum')
return context
rnn = fluid.layers.DynamicRNN()
cell_init = fluid.layers.fill_constant_batch_size_like(
input=decoder_boot,
value=0.0,
shape=[-1, decoder_size],
dtype='float32')
cell_init.stop_gradient = False
with rnn.block():
current_word = rnn.step_input(target_embedding)
encoder_vec = rnn.static_input(encoder_vec)
encoder_proj = rnn.static_input(encoder_proj)
hidden_mem = rnn.memory(init=decoder_boot, need_reorder=True)
cell_mem = rnn.memory(init=cell_init)
context = simple_attention(encoder_vec, encoder_proj, hidden_mem)
decoder_inputs = fluid.layers.concat(
input=[context, current_word], axis=1)
h, c = lstm_step(decoder_inputs, hidden_mem, cell_mem, decoder_size)
rnn.update_memory(hidden_mem, h)
rnn.update_memory(cell_mem, c)
out = fluid.layers.fc(input=h,
size=target_dict_dim,
bias_attr=True,
act='softmax')
rnn.output(out)
return rnn()
if not is_generating:
trg_word_idx = fluid.layers.data(
name='target_sequence', shape=[1], dtype='int64', lod_level=1)
trg_embedding = fluid.layers.embedding(
input=trg_word_idx,
size=[target_dict_dim, embedding_dim],
dtype='float32')
prediction = lstm_decoder_with_attention(trg_embedding, encoded_vector,
encoded_proj, decoder_boot,
decoder_size)
label = fluid.layers.data(
name='label_sequence', shape=[1], dtype='int64', lod_level=1)
cost = fluid.layers.cross_entropy(input=prediction, label=label)
avg_cost = fluid.layers.mean(x=cost)
feeding_list = ["source_sequence", "target_sequence", "label_sequence"]
return avg_cost, feeding_list
def to_lodtensor(data, place):
seq_lens = [len(seq) for seq in data]
cur_len = 0
lod = [cur_len]
for l in seq_lens:
cur_len += l
lod.append(cur_len)
flattened_data = np.concatenate(data, axis=0).astype("int64")
flattened_data = flattened_data.reshape([len(flattened_data), 1])
lod_t = core.LoDTensor()
lod_t.set(flattened_data, place)
lod_t.set_lod([lod])
return lod_t, lod[-1]
def lodtensor_to_ndarray(lod_tensor):
dims = lod_tensor.get_dims()
ndarray = np.zeros(shape=dims).astype('float32')
for i in xrange(np.product(dims)):
ndarray.ravel()[i] = lod_tensor.get_float_element(i)
return ndarray
def train():
avg_cost, feeding_list = seq_to_seq_net(
args.embedding_dim,
args.encoder_size,
args.decoder_size,
args.dict_size,
args.dict_size,
False,
beam_size=args.beam_size,
max_length=args.max_length)
# clone from default main program
inference_program = fluid.default_main_program().clone()
optimizer = fluid.optimizer.Adam(learning_rate=args.learning_rate)
optimizer.minimize(avg_cost)
fluid.memory_optimize(fluid.default_main_program())
train_batch_generator = paddle.batch(
paddle.reader.shuffle(
paddle.dataset.wmt14.train(args.dict_size), buf_size=1000),
batch_size=args.batch_size)
test_batch_generator = paddle.batch(
paddle.reader.shuffle(
paddle.dataset.wmt14.test(args.dict_size), buf_size=1000),
batch_size=args.batch_size)
place = core.CUDAPlace(0) if args.use_gpu else core.CPUPlace()
exe = Executor(place)
exe.run(framework.default_startup_program())
def do_validation():
total_loss = 0.0
count = 0
for batch_id, data in enumerate(test_batch_generator()):
src_seq = to_lodtensor(map(lambda x: x[0], data), place)[0]
trg_seq = to_lodtensor(map(lambda x: x[1], data), place)[0]
lbl_seq = to_lodtensor(map(lambda x: x[2], data), place)[0]
fetch_outs = exe.run(inference_program,
feed={
feeding_list[0]: src_seq,
feeding_list[1]: trg_seq,
feeding_list[2]: lbl_seq
},
fetch_list=[avg_cost],
return_numpy=False)
total_loss += lodtensor_to_ndarray(fetch_outs[0])[0]
count += 1
return total_loss / count
for pass_id in xrange(args.pass_num):
pass_start_time = time.time()
words_seen = 0
for batch_id, data in enumerate(train_batch_generator()):
src_seq, word_num = to_lodtensor(map(lambda x: x[0], data), place)
words_seen += word_num
trg_seq, word_num = to_lodtensor(map(lambda x: x[1], data), place)
words_seen += word_num
lbl_seq, _ = to_lodtensor(map(lambda x: x[2], data), place)
fetch_outs = exe.run(framework.default_main_program(),
feed={
feeding_list[0]: src_seq,
feeding_list[1]: trg_seq,
feeding_list[2]: lbl_seq
},
fetch_list=[avg_cost])
avg_cost_val = np.array(fetch_outs[0])
print('pass_id=%d, batch_id=%d, train_loss: %f' %
(pass_id, batch_id, avg_cost_val))
pass_end_time = time.time()
test_loss = do_validation()
time_consumed = pass_end_time - pass_start_time
words_per_sec = words_seen / time_consumed
print("pass_id=%d, test_loss: %f, words/s: %f, sec/pass: %f" %
(pass_id, test_loss, words_per_sec, time_consumed))
def infer():
pass
if __name__ == '__main__':
args = parser.parse_args()
if args.infer_only:
infer()
else:
train()
benchmark/fluid/mnist.py 0 → 100644
浏览文件 @ b853ac8c
# 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.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
import argparse
import time
import paddle.v2 as paddle
import paddle.fluid as fluid
import paddle.fluid.profiler as profiler
SEED = 1
DTYPE = "float32"
# random seed must set before configuring the network.
# fluid.default_startup_program().random_seed = SEED
def parse_args():
parser = argparse.ArgumentParser("mnist model benchmark.")
parser.add_argument(
'--batch_size', type=int, default=128, help='The minibatch size.')
parser.add_argument(
'--iterations', type=int, default=35, help='The number of minibatches.')
parser.add_argument(
'--pass_num', type=int, default=5, help='The number of passes.')
parser.add_argument(
'--device',
type=str,
default='GPU',
choices=['CPU', 'GPU'],
help='The device type.')
parser.add_argument(
'--infer_only', action='store_true', help='If set, run forward only.')
parser.add_argument(
'--use_cprof', action='store_true', help='If set, use cProfile.')
parser.add_argument(
'--use_nvprof',
action='store_true',
help='If set, use nvprof for CUDA.')
args = parser.parse_args()
return args
def print_arguments(args):
vars(args)['use_nvprof'] = (vars(args)['use_nvprof'] and
vars(args)['device'] == 'GPU')
print('----------- Configuration Arguments -----------')
for arg, value in sorted(vars(args).iteritems()):
print('%s: %s' % (arg, value))
print('------------------------------------------------')
def cnn_model(data):
conv_pool_1 = fluid.nets.simple_img_conv_pool(
input=data,
filter_size=5,
num_filters=20,
pool_size=2,
pool_stride=2,
act="relu")
conv_pool_2 = fluid.nets.simple_img_conv_pool(
input=conv_pool_1,
filter_size=5,
num_filters=50,
pool_size=2,
pool_stride=2,
act="relu")
# TODO(dzhwinter) : refine the initializer and random seed settting
SIZE = 10
input_shape = conv_pool_2.shape
param_shape = [reduce(lambda a, b: a * b, input_shape[1:], 1)] + [SIZE]
scale = (2.0 / (param_shape[0]**2 * SIZE))**0.5
predict = fluid.layers.fc(
input=conv_pool_2,
size=SIZE,
act="softmax",
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.NormalInitializer(
loc=0.0, scale=scale)))
return predict
def eval_test(exe, batch_acc, batch_size_tensor, inference_program):
test_reader = paddle.batch(
paddle.dataset.mnist.test(), batch_size=args.batch_size)
test_pass_acc = fluid.average.WeightedAverage()
for batch_id, data in enumerate(test_reader()):
img_data = np.array(map(lambda x: x[0].reshape([1, 28, 28]),
data)).astype(DTYPE)
y_data = np.array(map(lambda x: x[1], data)).astype("int64")
y_data = y_data.reshape([len(y_data), 1])
acc, weight = exe.run(inference_program,
feed={"pixel": img_data,
"label": y_data},
fetch_list=[batch_acc, batch_size_tensor])
test_pass_acc.add(value=acc, weight=weight)
pass_acc = test_pass_acc.eval()
return pass_acc
def run_benchmark(model, args):
if args.use_cprof:
pr = cProfile.Profile()
pr.enable()
start_time = time.time()
# Input data
images = fluid.layers.data(name='pixel', shape=[1, 28, 28], dtype=DTYPE)
label = fluid.layers.data(name='label', shape=[1], dtype='int64')
# Train program
predict = model(images)
cost = fluid.layers.cross_entropy(input=predict, label=label)
avg_cost = fluid.layers.mean(x=cost)
# Evaluator
batch_size_tensor = fluid.layers.create_tensor(dtype='int64')
batch_acc = fluid.layers.accuracy(
input=predict, label=label, total=batch_size_tensor)
# inference program
inference_program = fluid.default_main_program().clone()
with fluid.program_guard(inference_program):
inference_program = fluid.io.get_inference_program(
target_vars=[batch_acc, batch_size_tensor])
# Optimization
opt = fluid.optimizer.AdamOptimizer(
learning_rate=0.001, beta1=0.9, beta2=0.999)
opt.minimize(avg_cost)
fluid.memory_optimize(fluid.default_main_program())
# Initialize executor
place = fluid.CPUPlace() if args.device == 'CPU' else fluid.CUDAPlace(0)
exe = fluid.Executor(place)
# Parameter initialization
exe.run(fluid.default_startup_program())
# Reader
train_reader = paddle.batch(
paddle.dataset.mnist.train(), batch_size=args.batch_size)
accuracy = fluid.average.WeightedAverage()
for pass_id in range(args.pass_num):
accuracy.reset()
pass_start = time.time()
for batch_id, data in enumerate(train_reader()):
img_data = np.array(
map(lambda x: x[0].reshape([1, 28, 28]), data)).astype(DTYPE)
y_data = np.array(map(lambda x: x[1], data)).astype("int64")
y_data = y_data.reshape([len(y_data), 1])
start = time.time()
outs = exe.run(
fluid.default_main_program(),
feed={"pixel": img_data,
"label": y_data},
fetch_list=[avg_cost, batch_acc, batch_size_tensor]
) # The accuracy is the accumulation of batches, but not the current batch.
accuracy.add(value=outs[1], weight=outs[2])
end = time.time()
loss = np.array(outs[0])
acc = np.array(outs[1])
print("pass=%d, batch=%d, loss=%f, error=%f, elapse=%f" %
(pass_id, batch_id, loss, 1 - acc, (end - start) / 1000))
pass_end = time.time()
train_avg_acc = accuracy.eval()
test_avg_acc = eval_test(exe, batch_acc, batch_size_tensor,
inference_program)
print("pass=%d, train_avg_acc=%f, test_avg_acc=%f, elapse=%f" %
(pass_id, train_avg_acc, test_avg_acc,
(pass_end - pass_start) / 1000))
if __name__ == '__main__':
args = parse_args()
print_arguments(args)
if args.use_nvprof and args.device == 'GPU':
with profiler.cuda_profiler("cuda_profiler.txt", 'csv') as nvprof:
run_benchmark(cnn_model, args)
else:
run_benchmark(cnn_model, args)
benchmark/fluid/resnet.py 0 → 100644
浏览文件 @ b853ac8c
# 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.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import argparse
import functools
import numpy as np
import time
import cProfile, pstats, StringIO
import paddle.v2 as paddle
import paddle.fluid as fluid
import paddle.fluid.core as core
import paddle.fluid.profiler as profiler
def parse_args():
parser = argparse.ArgumentParser('Convolution model benchmark.')
parser.add_argument(
'--model',
type=str,
choices=['resnet_imagenet', 'resnet_cifar10'],
default='resnet_imagenet',
help='The model architecture.')
parser.add_argument(
'--batch_size', type=int, default=32, help='The minibatch size.')
parser.add_argument(
'--use_fake_data',
action='store_true',
help='use real data or fake data')
parser.add_argument(
'--skip_batch_num',
type=int,
default=5,
help='The first num of minibatch num to skip, for better performance test'
)
parser.add_argument(
'--iterations', type=int, default=80, help='The number of minibatches.')
parser.add_argument(
'--pass_num', type=int, default=100, help='The number of passes.')
parser.add_argument(
'--data_format',
type=str,
default='NCHW',
choices=['NCHW', 'NHWC'],
help='The data data_format, now only support NCHW.')
parser.add_argument(
'--device',
type=str,
default='GPU',
choices=['CPU', 'GPU'],
help='The device type.')
parser.add_argument(
'--data_set',
type=str,
default='flowers',
choices=['cifar10', 'flowers'],
help='Optional dataset for benchmark.')
parser.add_argument(
'--infer_only', action='store_true', help='If set, run forward only.')
parser.add_argument(
'--use_cprof', action='store_true', help='If set, use cProfile.')
parser.add_argument(
'--use_nvprof',
action='store_true',
help='If set, use nvprof for CUDA.')
parser.add_argument(
'--with_test',
action='store_true',
help='If set, test the testset during training.')
args = parser.parse_args()
return args
def print_arguments(args):
vars(args)['use_nvprof'] = (vars(args)['use_nvprof'] and
vars(args)['device'] == 'GPU')
print('----------- Configuration Arguments -----------')
for arg, value in sorted(vars(args).iteritems()):
print('%s: %s' % (arg, value))
print('------------------------------------------------')
def conv_bn_layer(input, ch_out, filter_size, stride, padding, act='relu'):
conv1 = fluid.layers.conv2d(
input=input,
filter_size=filter_size,
num_filters=ch_out,
stride=stride,
padding=padding,
act=None,
bias_attr=False)
return fluid.layers.batch_norm(input=conv1, act=act)
def shortcut(input, ch_out, stride):
ch_in = input.shape[1] if args.data_format == 'NCHW' else input.shape[-1]
if ch_in != ch_out:
return conv_bn_layer(input, ch_out, 1, stride, 0, None)
else:
return input
def basicblock(input, ch_out, stride):
short = shortcut(input, ch_out, stride)
conv1 = conv_bn_layer(input, ch_out, 3, stride, 1)
conv2 = conv_bn_layer(conv1, ch_out, 3, 1, 1, act=None)
return fluid.layers.elementwise_add(x=short, y=conv2, act='relu')
def bottleneck(input, ch_out, stride):
short = shortcut(input, ch_out * 4, stride)
conv1 = conv_bn_layer(input, ch_out, 1, stride, 0)
conv2 = conv_bn_layer(conv1, ch_out, 3, 1, 1)
conv3 = conv_bn_layer(conv2, ch_out * 4, 1, 1, 0, act=None)
return fluid.layers.elementwise_add(x=short, y=conv3, act='relu')
def layer_warp(block_func, input, ch_out, count, stride):
res_out = block_func(input, ch_out, stride)
for i in range(1, count):
res_out = block_func(res_out, ch_out, 1)
return res_out
def resnet_imagenet(input, class_dim, depth=50, data_format='NCHW'):
cfg = {
18: ([2, 2, 2, 1], basicblock),
34: ([3, 4, 6, 3], basicblock),
50: ([3, 4, 6, 3], bottleneck),
101: ([3, 4, 23, 3], bottleneck),
152: ([3, 8, 36, 3], bottleneck)
}
stages, block_func = cfg[depth]
conv1 = conv_bn_layer(input, ch_out=64, filter_size=7, stride=2, padding=3)
pool1 = fluid.layers.pool2d(
input=conv1, pool_type='avg', pool_size=3, pool_stride=2)
res1 = layer_warp(block_func, pool1, 64, stages[0], 1)
res2 = layer_warp(block_func, res1, 128, stages[1], 2)
res3 = layer_warp(block_func, res2, 256, stages[2], 2)
res4 = layer_warp(block_func, res3, 512, stages[3], 2)
pool2 = fluid.layers.pool2d(
input=res4,
pool_size=7,
pool_type='avg',
pool_stride=1,
global_pooling=True)
out = fluid.layers.fc(input=pool2, size=class_dim, act='softmax')
return out
def resnet_cifar10(input, class_dim, depth=32, data_format='NCHW'):
assert (depth - 2) % 6 == 0
n = (depth - 2) // 6
conv1 = conv_bn_layer(
input=input, ch_out=16, filter_size=3, stride=1, padding=1)
res1 = layer_warp(basicblock, conv1, 16, n, 1)
res2 = layer_warp(basicblock, res1, 32, n, 2)
res3 = layer_warp(basicblock, res2, 64, n, 2)
pool = fluid.layers.pool2d(
input=res3, pool_size=8, pool_type='avg', pool_stride=1)
out = fluid.layers.fc(input=pool, size=class_dim, act='softmax')
return out
def run_benchmark(model, args):
if args.use_cprof:
pr = cProfile.Profile()
pr.enable()
if args.data_set == "cifar10":
class_dim = 10
if args.data_format == 'NCHW':
dshape = [3, 32, 32]
else:
dshape = [32, 32, 3]
else:
class_dim = 102
if args.data_format == 'NCHW':
dshape = [3, 224, 224]
else:
dshape = [224, 224, 3]
input = fluid.layers.data(name='data', shape=dshape, dtype='float32')
label = fluid.layers.data(name='label', shape=[1], dtype='int64')
predict = model(input, class_dim)
cost = fluid.layers.cross_entropy(input=predict, label=label)
avg_cost = fluid.layers.mean(x=cost)
batch_size_tensor = fluid.layers.create_tensor(dtype='int64')
batch_acc = fluid.layers.accuracy(
input=predict, label=label, total=batch_size_tensor)
inference_program = fluid.default_main_program().clone()
with fluid.program_guard(inference_program):
inference_program = fluid.io.get_inference_program(
target_vars=[batch_acc, batch_size_tensor])
optimizer = fluid.optimizer.Momentum(learning_rate=0.01, momentum=0.9)
opts = optimizer.minimize(avg_cost)
fluid.memory_optimize(fluid.default_main_program())
train_reader = paddle.batch(
paddle.reader.shuffle(
paddle.dataset.cifar.train10()
if args.data_set == 'cifar10' else paddle.dataset.flowers.train(),
buf_size=5120),
batch_size=args.batch_size)
test_reader = paddle.batch(
paddle.dataset.cifar.test10()
if args.data_set == 'cifar10' else paddle.dataset.flowers.test(),
batch_size=args.batch_size)
def test(exe):
test_accuracy = fluid.average.WeightedAverage()
for batch_id, data in enumerate(test_reader()):
img_data = np.array(map(lambda x: x[0].reshape(dshape),
data)).astype("float32")
y_data = np.array(map(lambda x: x[1], data)).astype("int64")
y_data = y_data.reshape([-1, 1])
acc, weight = exe.run(inference_program,
feed={"data": img_data,
"label": y_data},
fetch_list=[batch_acc, batch_size_tensor])
test_accuracy.add(value=acc, weight=weight)
return test_accuracy.eval()
place = core.CPUPlace() if args.device == 'CPU' else core.CUDAPlace(0)
exe = fluid.Executor(place)
exe.run(fluid.default_startup_program())
accuracy = fluid.average.WeightedAverage()
if args.use_fake_data:
data = train_reader().next()
image = np.array(map(lambda x: x[0].reshape(dshape), data)).astype(
'float32')
label = np.array(map(lambda x: x[1], data)).astype('int64')
label = label.reshape([-1, 1])
iters, num_samples, start_time = 0, 0, time.time()
for pass_id in range(args.pass_num):
accuracy.reset()
train_accs = []
train_losses = []
for batch_id, data in enumerate(train_reader()):
if iters == args.skip_batch_num:
start_time = time.time()
num_samples = 0
if iters == args.iterations:
break
if not args.use_fake_data:
image = np.array(map(lambda x: x[0].reshape(dshape),
data)).astype('float32')
label = np.array(map(lambda x: x[1], data)).astype('int64')
label = label.reshape([-1, 1])
loss, acc, weight = exe.run(
fluid.default_main_program(),
feed={'data': image,
'label': label},
fetch_list=[avg_cost, batch_acc, batch_size_tensor])
iters += 1
num_samples += label[0]
accuracy.add(value=acc, weight=weight)
train_losses.append(loss)
train_accs.append(acc)
print("Pass: %d, Iter: %d, Loss: %f, Accuracy: %f" %
(pass_id, iters, loss, acc))
pass_train_acc = accuracy.eval()
# evaluation
if args.with_test:
pass_test_acc = test(exe)
train_elapsed = time.time() - start_time
print("Pass: %d, Loss: %f, Train Accuray: %f\n" %
(pass_id, np.mean(train_losses), np.mean(train_accs)))
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))
if args.use_cprof:
pr.disable()
s = StringIO.StringIO()
sortby = 'cumulative'
ps = pstats.Stats(pr, stream=s).sort_stats(sortby)
ps.print_stats()
print(s.getvalue())
if __name__ == '__main__':
model_map = {
'resnet_imagenet': resnet_imagenet,
'resnet_cifar10': resnet_cifar10
}
args = parse_args()
print_arguments(args)
if args.data_format == 'NHWC':
raise ValueError('Only support NCHW data_format now.')
if args.use_nvprof and args.device == 'GPU':
with profiler.cuda_profiler("cuda_profiler.txt", 'csv') as nvprof:
run_benchmark(model_map[args.model], args)
else:
run_benchmark(model_map[args.model], args)
benchmark/fluid/run.sh 0 → 100644
浏览文件 @ b853ac8c
#!/bin/bash
# This script benchmarking the PaddlePaddle Fluid on
# single thread single GPU.
export CUDNN_PATH=/paddle/cudnn_v5/cuda/lib
# disable openmp and mkl parallel
#https://github.com/PaddlePaddle/Paddle/issues/7199
export MKL_NUM_THREADS=1
export OMP_NUM_THREADS=1
ht=`lscpu |grep "per core"|awk -F':' '{print $2}'|xargs`
if [ $ht -eq 1 ]; then # HT is OFF
if [ -z "$KMP_AFFINITY" ]; then
export KMP_AFFINITY="granularity=fine,compact,0,0"
fi
if [ -z "$OMP_DYNAMIC" ]; then
export OMP_DYNAMIC="FALSE"
fi
else # HT is ON
if [ -z "$KMP_AFFINITY" ]; then
export KMP_AFFINITY="granularity=fine,compact,1,0"
fi
fi
# disable multi-gpu if have more than one
export CUDA_VISIBLE_DEVICES=0
export LD_LIBRARY_PATH=/usr/local/lib:$LD_LIBRARY_PATH
export LD_LIBRARY_PATH=$CUDNN_PATH:$LD_LIBRARY_PATH
# vgg16
# cifar10 gpu cifar10 128
FLAGS_benchmark=true python fluid/vgg.py \
--device=GPU \
--batch_size=128 \
--skip_batch_num=5 \
--iterations=30 \
2>&1 > vgg16_gpu_128.log
# resnet50
# resnet50 gpu cifar10 128
FLAGS_benchmark=true python fluid/resnet.py \
--device=GPU \
--batch_size=128 \
--data_set=cifar10 \
--model=resnet_cifar10 \
--skip_batch_num=5 \
--iterations=30 \
2>&1 > resnet50_gpu_128.log
# lstm
benchmark/fluid/stacked_dynamic_lstm.py 0 → 100644
浏览文件 @ b853ac8c
# 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.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import argparse
import cPickle
import os
import random
import time
import numpy
import paddle.v2 as paddle
import paddle.v2.dataset.imdb as imdb
import paddle.fluid as fluid
from paddle.v2 import batch
import paddle.fluid.profiler as profiler
def parse_args():
parser = argparse.ArgumentParser("Understand Sentiment by Dynamic RNN.")
parser.add_argument(
'--batch_size',
type=int,
default=32,
help='The sequence number of a batch data. (default: %(default)d)')
parser.add_argument(
'--emb_dim',
type=int,
default=512,
help='Dimension of embedding table. (default: %(default)d)')
parser.add_argument(
'--hidden_dim',
type=int,
default=512,
help='Hidden size of lstm unit. (default: %(default)d)')
parser.add_argument(
'--pass_num',
type=int,
default=100,
help='Epoch number to train. (default: %(default)d)')
parser.add_argument(
'--device',
type=str,
default='CPU',
choices=['CPU', 'GPU'],
help='The device type.')
parser.add_argument(
'--crop_size',
type=int,
default=int(os.environ.get('CROP_SIZE', '1500')),
help='The max sentence length of input. Since this model use plain RNN,'
' Gradient could be explored if sentence is too long')
args = parser.parse_args()
return args
word_dict = imdb.word_dict()
def crop_sentence(reader, crop_size):
unk_value = word_dict['<unk>']
def __impl__():
for item in reader():
if len([x for x in item[0] if x != unk_value]) < crop_size:
yield item
return __impl__
def main():
args = parse_args()
lstm_size = args.hidden_dim
data = fluid.layers.data(
name="words", shape=[1], lod_level=1, dtype='int64')
sentence = fluid.layers.embedding(
input=data, size=[len(word_dict), args.emb_dim])
sentence = fluid.layers.fc(input=sentence, size=lstm_size, act='tanh')
rnn = fluid.layers.DynamicRNN()
with rnn.block():
word = rnn.step_input(sentence)
prev_hidden = rnn.memory(value=0.0, shape=[lstm_size])
prev_cell = rnn.memory(value=0.0, shape=[lstm_size])
def gate_common(
ipt,
hidden,
size, ):
gate0 = fluid.layers.fc(input=ipt, size=size, bias_attr=True)
gate1 = fluid.layers.fc(input=hidden, size=size, bias_attr=False)
gate = fluid.layers.sums(input=[gate0, gate1])
return gate
forget_gate = fluid.layers.sigmoid(
x=gate_common(word, prev_hidden, lstm_size))
input_gate = fluid.layers.sigmoid(
x=gate_common(word, prev_hidden, lstm_size))
output_gate = fluid.layers.sigmoid(
x=gate_common(word, prev_hidden, lstm_size))
cell_gate = fluid.layers.tanh(
x=gate_common(word, prev_hidden, lstm_size))
cell = fluid.layers.sums(input=[
fluid.layers.elementwise_mul(
x=forget_gate, y=prev_cell), fluid.layers.elementwise_mul(
x=input_gate, y=cell_gate)
])
hidden = fluid.layers.elementwise_mul(
x=output_gate, y=fluid.layers.tanh(x=cell))
rnn.update_memory(prev_cell, cell)
rnn.update_memory(prev_hidden, hidden)
rnn.output(hidden)
last = fluid.layers.sequence_pool(rnn(), 'last')
logit = fluid.layers.fc(input=last, size=2, act='softmax')
loss = fluid.layers.cross_entropy(
input=logit,
label=fluid.layers.data(
name='label', shape=[1], dtype='int64'))
loss = fluid.layers.mean(x=loss)
# add acc
batch_size_tensor = fluid.layers.create_tensor(dtype='int64')
batch_acc = fluid.layers.accuracy(input=logit, label=fluid.layers.data(name='label', \
shape=[1], dtype='int64'), total=batch_size_tensor)
inference_program = fluid.default_main_program().clone()
with fluid.program_guard(inference_program):
inference_program = fluid.io.get_inference_program(
target_vars=[batch_acc, batch_size_tensor])
adam = fluid.optimizer.Adam()
adam.minimize(loss)
fluid.memory_optimize(fluid.default_main_program())
place = fluid.CPUPlace() if args.device == 'CPU' else fluid.CUDAPlace(0)
exe = fluid.Executor(place)
exe.run(fluid.default_startup_program())
def train_loop(pass_num, crop_size):
with profiler.profiler(args.device, 'total') as prof:
for pass_id in range(pass_num):
train_reader = batch(
paddle.reader.shuffle(
crop_sentence(imdb.train(word_dict), crop_size),
buf_size=25000),
batch_size=args.batch_size)
word_nums = 0
pass_start_time = time.time()
for batch_id, data in enumerate(train_reader()):
tensor_words = to_lodtensor([x[0] for x in data], place)
for x in data:
word_nums += len(x[0])
label = numpy.array([x[1] for x in data]).astype("int64")
label = label.reshape((-1, 1))
loss_np, acc, weight = exe.run(
fluid.default_main_program(),
feed={"words": tensor_words,
"label": label},
fetch_list=[loss, batch_acc, batch_size_tensor])
print("pass_id=%d, batch_id=%d, loss=%f, acc=%f" %
(pass_id, batch_id, loss_np, acc))
pass_end_time = time.time()
time_consumed = pass_end_time - pass_start_time
words_per_sec = word_nums / time_consumed
print("pass_id=%d, sec/pass: %f, words/s: %f" %
(pass_id, time_consumed, words_per_sec))
train_loop(args.pass_num, args.crop_size)
def to_lodtensor(data, place):
seq_lens = [len(seq) for seq in data]
cur_len = 0
lod = [cur_len]
for l in seq_lens:
cur_len += l
lod.append(cur_len)
flattened_data = numpy.concatenate(data, axis=0).astype("int64")
flattened_data = flattened_data.reshape([len(flattened_data), 1])
res = fluid.LoDTensor()
res.set(flattened_data, place)
res.set_lod([lod])
return res
if __name__ == '__main__':
main()
benchmark/fluid/vgg.py 0 → 100644
浏览文件 @ b853ac8c
# 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.
"""VGG16 benchmark in Fluid"""
from __future__ import print_function
import sys
import time
import numpy as np
import paddle.v2 as paddle
import paddle.fluid as fluid
import paddle.fluid.core as core
import argparse
import functools
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument(
'--batch_size', type=int, default=128, help="Batch size for training.")
parser.add_argument(
'--skip_batch_num',
type=int,
default=5,
help='The first num of minibatch num to skip, for better performance test')
parser.add_argument(
'--iterations', type=int, default=80, help='The number of minibatches.')
parser.add_argument(
'--learning_rate',
type=float,
default=1e-3,
help="Learning rate for training.")
parser.add_argument('--pass_num', type=int, default=50, help="No. of passes.")
parser.add_argument(
'--device',
type=str,
default='GPU',
choices=['CPU', 'GPU'],
help="The device type.")
parser.add_argument(
'--data_format',
type=str,
default='NCHW',
choices=['NCHW', 'NHWC'],
help='The data order, now only support NCHW.')
parser.add_argument(
'--data_set',
type=str,
default='cifar10',
choices=['cifar10', 'flowers'],
help='Optional dataset for benchmark.')
parser.add_argument(
'--with_test',
action='store_true',
help='If set, test the testset during training.')
args = parser.parse_args()
def vgg16_bn_drop(input):
def conv_block(input, num_filter, groups, dropouts):
return fluid.nets.img_conv_group(
input=input,
pool_size=2,
pool_stride=2,
conv_num_filter=[num_filter] * groups,
conv_filter_size=3,
conv_act='relu',
conv_with_batchnorm=True,
conv_batchnorm_drop_rate=dropouts,
pool_type='max')
conv1 = conv_block(input, 64, 2, [0.3, 0])
conv2 = conv_block(conv1, 128, 2, [0.4, 0])
conv3 = conv_block(conv2, 256, 3, [0.4, 0.4, 0])
conv4 = conv_block(conv3, 512, 3, [0.4, 0.4, 0])
conv5 = conv_block(conv4, 512, 3, [0.4, 0.4, 0])
drop = fluid.layers.dropout(x=conv5, dropout_prob=0.5)
fc1 = fluid.layers.fc(input=drop, size=512, act=None)
bn = fluid.layers.batch_norm(input=fc1, act='relu')
drop2 = fluid.layers.dropout(x=bn, dropout_prob=0.5)
fc2 = fluid.layers.fc(input=drop2, size=512, act=None)
return fc2
def main():
if args.data_set == "cifar10":
classdim = 10
if args.data_format == 'NCHW':
data_shape = [3, 32, 32]
else:
data_shape = [32, 32, 3]
else:
classdim = 102
if args.data_format == 'NCHW':
data_shape = [3, 224, 224]
else:
data_shape = [224, 224, 3]
# Input data
images = fluid.layers.data(name='pixel', shape=data_shape, dtype='float32')
label = fluid.layers.data(name='label', shape=[1], dtype='int64')
# Train program
net = vgg16_bn_drop(images)
predict = fluid.layers.fc(input=net, size=classdim, act='softmax')
cost = fluid.layers.cross_entropy(input=predict, label=label)
avg_cost = fluid.layers.mean(x=cost)
# Evaluator
batch_size_tensor = fluid.layers.create_tensor(dtype='int64')
batch_acc = fluid.layers.accuracy(
input=predict, label=label, total=batch_size_tensor)
# inference program
inference_program = fluid.default_main_program().clone()
with fluid.program_guard(inference_program):
inference_program = fluid.io.get_inference_program(
target_vars=[batch_acc, batch_size_tensor])
# Optimization
optimizer = fluid.optimizer.Adam(learning_rate=args.learning_rate)
opts = optimizer.minimize(avg_cost)
fluid.memory_optimize(fluid.default_main_program())
# Initialize executor
place = core.CPUPlace() if args.device == 'CPU' else core.CUDAPlace(0)
exe = fluid.Executor(place)
# Parameter initialization
exe.run(fluid.default_startup_program())
# data reader
train_reader = paddle.batch(
paddle.reader.shuffle(
paddle.dataset.cifar.train10()
if args.data_set == 'cifar10' else paddle.dataset.flowers.train(),
buf_size=5120),
batch_size=args.batch_size)
test_reader = paddle.batch(
paddle.dataset.cifar.test10()
if args.data_set == 'cifar10' else paddle.dataset.flowers.test(),
batch_size=args.batch_size)
# test
def test(exe):
test_accuracy = fluid.average.WeightedAverage()
for batch_id, data in enumerate(test_reader()):
img_data = np.array(map(lambda x: x[0].reshape(data_shape),
data)).astype("float32")
y_data = np.array(map(lambda x: x[1], data)).astype("int64")
y_data = y_data.reshape([-1, 1])
acc, weight = exe.run(inference_program,
feed={"pixel": img_data,
"label": y_data},
fetch_list=[batch_acc, batch_size_tensor])
test_accuracy.add(value=acc, weight=weight)
return test_accuracy.eval()
iters, num_samples, start_time = 0, 0, time.time()
accuracy = fluid.average.WeightedAverage()
for pass_id in range(args.pass_num):
accuracy.reset()
train_accs = []
train_losses = []
for batch_id, data in enumerate(train_reader()):
if iters == args.skip_batch_num:
start_time = time.time()
num_samples = 0
if iters == args.iterations:
break
img_data = np.array(map(lambda x: x[0].reshape(data_shape),
data)).astype("float32")
y_data = np.array(map(lambda x: x[1], data)).astype("int64")
y_data = y_data.reshape([-1, 1])
loss, acc, weight = exe.run(
fluid.default_main_program(),
feed={"pixel": img_data,
"label": y_data},
fetch_list=[avg_cost, batch_acc, batch_size_tensor])
accuracy.add(value=acc, weight=weight)
iters += 1
num_samples += len(data)
print(
"Pass = %d, Iter = %d, Loss = %f, Accuracy = %f" %
(pass_id, iters, loss, acc)
) # The accuracy is the accumulation of batches, but not the current batch.
pass_train_acc = accuracy.eval()
train_losses.append(loss)
train_accs.append(acc)
# evaluation
if args.with_test:
pass_test_acc = test(exe)
train_elapsed = time.time() - start_time
print("Pass: %d, Loss: %f, Train Accuray: %f\n" %
(pass_id, np.mean(train_losses), np.mean(train_accs)))
def print_arguments():
print('----------- Configuration Arguments -----------')
for arg, value in sorted(vars(args).iteritems()):
print('%s: %s' % (arg, value))
print('------------------------------------------------')
if __name__ == "__main__":
print_arguments()
main()
doc/fluid/api/layers.rst
浏览文件 @ b853ac8c
......@@ -494,6 +494,12 @@ reshape
.. autofunction:: paddle.fluid.layers.reshape
:noindex:
pad
---
.. autofunction:: paddle.fluid.layers.pad
:noindex:
scale
-----
......
doc/fluid/design/algorithm/parameter_average.md
浏览文件 @ b853ac8c
......@@ -7,7 +7,7 @@ Polyak and Juditsky (1992) showed that the test performance of simple average of
Hence, to accelerate the speed of Stochastic Gradient Descent, Averaged Stochastic Gradient Descent (ASGD) was proposed in Polyak and Juditsky (1992). For ASGD, the running average of parameters obtained by SGD, is used as the estimator for <img src="./images/theta_star.gif"/><br/> . The averaging is done as follows:
<img src="./images/asgd.gif" align="center"/><br/>
![](./images/asgd.gif)
We propose averaging for any optimizer similar to how ASGD performs it, as mentioned above.
......
doc/fluid/design/concepts/README.md
浏览文件 @ b853ac8c
......@@ -6,11 +6,33 @@ Here are some initial thoughts. Your comments are welcome!
I think we need only the following few CMake functions to make a project description mean and clean:
| C++ | CUDA C++ | Go |
|---|---|---|
| cc_library | nv_library | go_library |
| cc_binary | nv_binary | go_binary |
| cc_test | nv_test | go_test |
<table>
<thead>
<tr>
<th>C++</th>
<th>CUDA C++</th>
<th>Go</th>
</tr>
</thead>
<tbody>
<tr>
<td>cc_library </td>
<td>nv_library </td>
<td>go_library </td>
</tr>
<tr>
<td>cc_binary </td>
<td>nv_binary </td>
<td>go_binary </td>
</tr>
<tr>
<td> cc_test </td>
<td> nv_test </td>
<td> go_test </td>
</tr>
</tbody>
</table>
- The `_library` functions generate .a files from source code.
- The `_binary` functions generate executable binary files.
......
doc/fluid/design/concepts/block.md
浏览文件 @ b853ac8c
......@@ -14,11 +14,29 @@ In programming languages, a block is a pair of curly braces that includes local
Blocks work with control flow structures like `if`, `else`, and `for`, which have equivalents in deep learning:
| programming languages | PaddlePaddle |
|-----------------------|-----------------------|
| for, while loop | RNN, WhileOp |
| if, if-else, switch | IfElseOp, SwitchOp |
| sequential execution | a sequence of layers |
<table>
<thead>
<tr>
<th>programming languages</th>
<th>PaddlePaddle</th>
</tr>
</thead>
<tbody>
<tr>
<td>for, while loop </td>
<td>RNN, WhileOp </td>
</tr>
<tr>
<td>if, if-else, switch </td>
<td>IfElseOp, SwitchOp </td>
</tr>
<tr>
<td>sequential execution </td>
<td>a sequence of layers </td>
</tr>
</tbody>
</table>
A key difference is that a C++ program describes a one pass computation, whereas a deep learning program describes both the forward and backward passes.
......@@ -26,12 +44,33 @@ A key difference is that a C++ program describes a one pass computation, whereas
The existence of the backward pass makes the execution of a block of PaddlePaddle different from traditional programs:
| programming languages | PaddlePaddle |
|-----------------------|---------------------------------|
| stack | scope hierarchy |
| stack frame | scope |
| push at entering block| push at entering block |
| pop at leaving block | destroy when minibatch completes|
<table>
<thead>
<tr>
<th>programming languages</th>
<th>PaddlePaddle</th>
</tr>
</thead>
<tbody>
<tr>
<td>stack </td>
<td>scope hierarchy </td>
</tr>
<tr>
<td>stack frame </td>
<td>scope </td>
</tr>
<tr>
<td>push at entering block </td>
<td>push at entering block </td>
</tr>
<tr>
<td>pop at leaving block </td>
<td>destroy when minibatch completes </td>
</tr>
</tbody>
</table>
1. In traditional programs:
......
doc/fluid/design/concepts/functions_operators_layers.md
浏览文件 @ b853ac8c
......@@ -86,12 +86,40 @@ def layer.fc(X):
We'd like to have Python bindings to operators in package `paddle.operator`, and Python compositions of operators in package `paddle.layer`. So we have the following concepts in above illustrative example:
| C++ functions/functors | mul | add | | |
|------------------------|--------------|--------------|-------------|----------|
| C++ operator class | mulOp | addOp | FCOp | |
| Python binding | operator.mul | operator.add | operator.fc | |
| Python function | | | | layer.fc |
<table>
<thead>
<tr>
<th>C++ functions/functors</th>
<th>mul</th>
<th>add</th>
<th></th>
<th></th>
</tr>
</thead>
<tbody>
<tr>
<td>C++ operator class </td>
<td>mulOp</td>
<td>addOp </td>
<td>FCOp </td>
<td></td>
</tr>
<tr>
<td>Python binding </td>
<td>operator.mul</td>
<td> operator.add </td>
<td>operator.fc </td>
<td></td>
</tr>
<tr>
<td>Python function </td>
<td></td>
<td></td>
<td> </td>
<td>layer.fc</td>
</tr>
</tbody>
</table>
This is how we differentiate layer and operators in PaddlePaddle:
......
doc/fluid/design/concepts/lod_tensor.md
浏览文件 @ b853ac8c
......@@ -2,12 +2,38 @@
Like other deep learning systems, PaddlePaddle supports training models from sequence data. Also, like other systems, PaddlePaddle represent a mini-batch of sequences as a Tensor. What is different is that PaddlePaddle doesn't require all sequences in a mini-batch to be of the same length. Thus no need for padding zeros.
| | TensorFlow | PaddlePaddle |
|-----------------------|------------|--------------|
| RNN | Support | Support |
| recursive RNN | Support | Support |
| padding zeros | Must | No need |
| blob data type | Tensor | LoDTensor |
<table>
<thead>
<tr>
<th></th>
<th>TensorFlow</th>
<th>PaddlePaddle</th>
</tr>
</thead>
<tbody>
<tr>
<td>RNN </td>
<td>Support </td>
<td>Support </td>
</tr>
<tr>
<td>recursive RNN </td>
<td>Support </td>
<td>Support </td>
</tr>
<tr>
<td>padding zeros </td>
<td> Must </td>
<td>No need </td>
</tr>
<tr>
<td> blob data type </td>
<td> Tensor</td>
<td> LoDTensor </td>
</tr>
</tbody>
</table>
PaddlePaddle achieves this flexibility by passing through a new data type, *LoD Tensor*, which is a Tensor attached with segmentation index known as *LoD*, between operators. The LoD index doesn't only segment a tensor, but also recursively segments sub-sequences. This document presents the design of LoD and LoDTensor.
......
doc/fluid/design/concepts/var_desc.md
浏览文件 @ b853ac8c
......@@ -10,10 +10,27 @@ PaddlePaddle uses proto message to describe compile time program because :
The computation `Program` consists of nested `Blocks`. Each `Block` will consist of data(i.e. `Variable`) and `Operations`. The concept to represent them is in the table below.
| |compile time|runtime|
|---|---|---|
|Data|VarDesc(proto)|Variable(cpp)|
|Operation|OpDesc(proto)|Operator(cpp)|
<table>
<thead>
<tr>
<th></th>
<th>compile time</th>
<th>runtime</th>
</tr>
</thead>
<tbody>
<tr>
<td>Data </td>
<td>VarDesc(proto) </td>
<td>Variable(cpp) </td>
</tr>
<tr>
<td>Operation </td>
<td>OpDesc(proto) </td>
<td>Operator(cpp) </td>
</tr>
</tbody>
</table>
## Definition of VarType
......
doc/fluid/design/concurrent/concurrent_programming.md
浏览文件 @ b853ac8c
......@@ -10,12 +10,38 @@ The answer relies on the fact that a `ProgramDesc` is similar to an abstract syn
The following table compares concepts in Fluid and Go
| Go | Fluid |
|----|-------|
|user-defined functions | [layers](https://github.com/PaddlePaddle/Paddle/tree/develop/python/paddle/fluid) |
| control-flow and built-in functions | [intrinsics/operators](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/operators) |
| goroutines, channels | [class ThreadPool](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/framework/thread_pool.h) |
| runtime | [class Executor](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/executor.h) |
<table>
<thead>
<tr>
<th></th>
<th>Go</th>
<th>Fluid</th>
</tr>
</thead>
<tbody>
<tr>
<td>user-defined functions </td>
<td>
<a href="https://github.com/PaddlePaddle/Paddle/tree/develop/python/paddle/fluid">layers</a></td>
</tr>
<tr>
<td>control-flow and built-in functions </td>
<td>
<a href="https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/operators">intrinsics/operators</a></td>
</tr>
<tr>
<td>goroutines, channels </td>
<td>
<a href="https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/framework/thread_pool.h">class ThreadPool</a></td>
</tr>
<tr>
<td>runtime </td>
<td>
<a href="https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/executor.h">class Executor</a></td>
</tr>
</tbody>
</table>
## An Example Concurrent Program
......@@ -77,11 +103,11 @@ message ProgramDesc {
read(output = X)
kube_get_workers_addrs(output = L)
Y = tensor_array(len(L))
parallel_for(input = X, output = Y,
parallel_for(input = X, output = Y,
attrs = {L, block_id(1)}) # referring to block 1
]
}
block[1] = Block {
parent = 0,
vars = [x, y, index],
......@@ -102,7 +128,7 @@ func main() { //// block 0
X = fluid.read(...)
L = fluid.k8s.get_worker_addrs()
Y = fluid.tensor_array(len(L))
fluid.parallel_for(X, L,
fluid.parallel_for(X, L,
func(index int) { //// block 1
x = X[index]
fluid.send(L[index], x)
......@@ -116,7 +142,7 @@ An explanation of the above program:
- `fluid.k8s` is a package that provides access to Kubernetes API.
- `fluid.k8s.get_worker_addrs` returns the list of IP and ports of all pods of the current job except for the current one (the master pod).
- `fluid.tensor_array` creates a [tensor array](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/lod_tensor_array.h). `fluid.parallel_for` creates a `ParallelFor` intrinsic, which, when executed,
- `fluid.tensor_array` creates a [tensor array](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/lod_tensor_array.h). `fluid.parallel_for` creates a `ParallelFor` intrinsic, which, when executed,
1. creates `len(L)` scopes, each for the concurrent running of the sub-block (block 1 in this case), and initializes a variable named "index" in the scope to an integer value in the range `[0, len(L)-1]`, and
2. creates `len(L)` threads by calling into the `ThreadPool` singleton, each thread
......
doc/fluid/design/concurrent/csp.md
浏览文件 @ b853ac8c
......@@ -13,14 +13,41 @@ Most DL systems, including TensorFlow, Caffe2, and MxNet, can asynchronously exe
There were many concurrent programming models, implemented in various forms:
| concurrent programming model | implementation |
|-----|-----|
| mutex | types and functions in standard libraries |
| semaphore | types and functions in standard libraries |
| communicating sequential processes (CSP) | Go programming language |
| actor model | Erlang programming language |
| message passing | MPI |
| bulk synchronous parallel (BSP) | Pregel distributed programming framework |
<table>
<thead>
<tr>
<th>concurrent programming model</th>
<th>implementation</th>
</tr>
</thead>
<tbody>
<tr>
<td>mutex </td>
<td>types and functions in standard libraries </td>
</tr>
<tr>
<td>semaphore </td>
<td> types and functions in standard libraries </td>
</tr>
<tr>
<td> communicating sequential processes (CSP) </td>
<td> Go programming language </td>
</tr>
<tr>
<td> actor model </td>
<td> Erlang programming language </td>
</tr>
<tr>
<td> message passing </td>
<td> MPI </td>
</tr>
<tr>
<td> bulk synchronous parallel (BSP) </td>
<td> Pregel distributed programming framework </td>
</tr>
</tbody>
</table>
Since Fluid was designed to be a programming language, we would like to implement CSP in Fluid.
......@@ -118,9 +145,9 @@ There are four types of actions with a channel:
```go
close(ch)
```
Please be aware that a closed channel is not a nil channel, which is `var ch chan int`.
There are some [axioms with channels](https://dave.cheney.net/2014/03/19/channel-axioms):
1. A send to a nil channel blocks forever
......
doc/fluid/design/modules/python_api.md
浏览文件 @ b853ac8c
......@@ -2,12 +2,33 @@
Due to the refactorization of the PaddlePaddle core, we need Python classes to construct corresponding protobuf messages that describe a DL program.
| Python classes | Protobuf messages |
| --- | --- |
| Program | ProgramDesc |
| Block | BlockDesc |
| Operator | OpDesc |
| Variable | VarDesc |
<table>
<thead>
<tr>
<th>Python classes</th>
<th>Protobuf messages</th>
</tr>
</thead>
<tbody>
<tr>
<td>Program </td>
<td>ProgramDesc </td>
</tr>
<tr>
<td>Block </td>
<td>BlockDesc </td>
</tr>
<tr>
<td>Operator </td>
<td>OpDesc </td>
</tr>
<tr>
<td>Variable </td>
<td>VarDesc </td>
</tr>
</tbody>
</table>
Please be aware that these Python classes need to maintain some construction-time information, which are not part of the protobuf messages.
......
doc/fluid/design/motivation/fluid.md
浏览文件 @ b853ac8c
......@@ -10,11 +10,37 @@ Fluid is the answer. Fluid is similar to PyTorch and TensorFlow Eager Execution
Deep learning infrastructure is one of the fastest evolving technologies. Within four years, there have already been three generations of technologies invented.
| Existed since | model as sequence of layers | model as graph of operators | No model |
|--|--|--|--|
| 2013 | Caffe, Theano, Torch, PaddlePaddle | | |
| 2015 | | TensorFlow, MxNet, Caffe2, ONNX, n-graph | |
| 2016 | | | PyTorch, TensorFlow Eager Execution, PaddlePaddle Fluid |
<table>
<thead>
<tr>
<th>Existed since</th>
<th>model as sequence of layers</th>
<th>model as graph of operators</th>
<th>No model</th>
</tr>
</thead>
<tbody>
<tr>
<td>2013 </td>
<td>Caffe, Theano, Torch, PaddlePaddle </td>
<td> </td>
<td> </td>
</tr>
<tr>
<td>2015 </td>
<td> </td>
<td>TensorFlow, MxNet, Caffe2, ONNX, n-graph </td>
<td> </td>
</tr>
<tr>
<td>2016 </td>
<td> </td>
<td> </td>
<td> PyTorch, TensorFlow Eager Execution, PaddlePaddle Fluid</td>
</tr>
</tbody>
</table>
From the above table, we see that the deep learning technology is evolving towards getting rid of the concept of a model. To understand the reasons behind this direction, a comparison of the *programming paradigms* or the ways to program deep learning applications using these systems, would be helpful. The following section goes over these.
......
doc/fluid/design/motivation/refactorization.md
浏览文件 @ b853ac8c
......@@ -36,11 +36,37 @@ At compile time, the Python program generates a protobuf message representation
At runtime, the C++ program realizes the graph and runs it.
| | Representation (protobuf messages) | Realization (C++ class objects) |
|---|---|---|
|Data|[VarDesc](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/framework.proto#L107)|[Variable](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/variable.h#L24)|
|Operation|[OpDesc](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/framework.proto#L35)|[Operator](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/operator.h#L64)|
|Block|BlockDesc|Block|
<table>
<thead>
<tr>
<th></th>
<th>Representation (protobuf messages)</th>
<th>Realization (C++ class objects) </th>
</tr>
</thead>
<tbody>
<tr>
<td>Data</td>
<td>
<a href="https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/framework.proto#L107">VarDesc</a></td>
<td>
<a href="https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/variable.h#L24">Variable</a></td>
</tr>
<tr>
<td>Operation </td>
<td>
<a href="https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/framework.proto#L35">OpDesc</a></td>
<td>
<a href="https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/operator.h#L64">Operator</a></td>
</tr>
<tr>
<td>Block </td>
<td>BlockDesc </td>
<td>Block </td>
</tbody>
</table>
The word *graph* is interchangeable with *block* in this document. A graph consists of computation steps and local variables similar to a C++/Java program block, or a pair of parentheses(`{` and `}`).
......
doc/fluid/design/network/deep_speech_2.md
浏览文件 @ b853ac8c
# DeepSpeech2 on PaddlePaddle: Design Doc
# DeepSpeech2 on PaddlePaddle: Design Doc
We are planning to build Deep Speech 2 (DS2) \[[1](#references)\], a powerful Automatic Speech Recognition (ASR) engine, on PaddlePaddle. For the first-stage plan, we have the following short-term goals:
......@@ -68,11 +68,33 @@ We roughly break down the project into 14 tasks:
Tasks parallelizable within phases:
Roadmap | Description | Parallelizable Tasks
----------- | :------------------------------------ | :--------------------
Phase I | Simplified model & components | *Task 1* ~ *Task 8*
Phase II | Standard model & benchmarking & profiling | *Task 9* ~ *Task 12*
Phase III | Documentations | *Task13* ~ *Task14*
<table>
<thead>
<tr>
<th>Roadmap</th>
<th>Description</th>
<th> Parallelizable Tasks</th>
</tr>
</thead>
<tbody>
<tr>
<td>Phase I </td>
<td>Simplified model & components </td>
<td>Task 1 ~ Task 8</td>
</tr>
<tr>
<td>Phase II </td>
<td> Standard model & benchmarking & profiling</td>
<td>Task 9 ~ Task 12 </td>
</tr>
<tr>
<td>Phase III </td>
<td> Documentations</td>
<td> Task13 ~ Task14 </td>
</tr>
</tbody>
</table>
Issue for each task will be created later. Contributions, discussions and comments are all highly appreciated and welcomed!
......@@ -102,37 +124,82 @@ We don't have to persist on this 2-3-7-1-1-1 depth \[[2](#references)\]. Similar
Key ingredients about the layers:
- **Data Layers**:
- **Data Layers**:
- Frame sequences data of audio **spectrogram** (with FFT).
- Token sequences data of **transcription** text (labels).
- Token sequences data of **transcription** text (labels).
- These two type of sequences do not have the same lengthes, thus a CTC-loss layer is required.
- **2D Convolution Layers**:
- **2D Convolution Layers**:
- Not only temporal convolution, but also **frequency convolution**. Like a 2D image convolution, but with a variable dimension (i.e. temporal dimension).
- With striding for only the first convlution layer.
- No pooling for all convolution layers.
- **Uni-directional RNNs**
- **Uni-directional RNNs**
- Uni-directional + row convolution: for low-latency inference.
- Bi-direcitional + without row convolution: if we don't care about the inference latency.
- **Row convolution**:
- For looking only a few steps ahead into the feature, instead of looking into a whole sequence in bi-directional RNNs.
- Not nessesary if with bi-direcitional RNNs.
- Not nessesary if with bi-direcitional RNNs.
- "**Row**" means convolutions are done within each frequency dimension (row), and no convolution kernels shared across.
- **Batch Normalization Layers**:
- Added to all above layers (except for data and loss layer).
- Sequence-wise normalization for RNNs: BatchNorm only performed on input-state projection and not state-state projection, for efficiency consideration.
Required Components | PaddlePaddle Support | Need to Develop
:------------------------------------- | :-------------------------------------- | :-----------------------
Data Layer I (Spectrogram) | Not supported yet. | TBD (Task 3)
Data Layer II (Transcription) | `paddle.data_type.integer_value_sequence` | -
2D Convolution Layer | `paddle.layer.image_conv_layer` | -
DataType Converter (vec2seq) | `paddle.layer.block_expand` | -
Bi-/Uni-directional RNNs | `paddle.layer.recurrent_group` | -
Row Convolution Layer | Not supported yet. | TBD (Task 4)
CTC-loss Layer | `paddle.layer.warp_ctc` | -
Batch Normalization Layer | `paddle.layer.batch_norm` | -
CTC-Beam search | Not supported yet. | TBD (Task 6)
<table>
<thead>
<tr>
<th>Required Components</th>
<th> PaddlePaddle Support</th>
<th> Need to Develop</th>
</tr>
</thead>
<tbody>
<tr>
<td>Data Layer I (Spectrogram) </td>
<td>Not supported yet.</td>
<td>TBD (Task 3)</td>
</tr>
<tr>
<td>Data Layer II (Transcription) </td>
<td> paddle.data_type.integer_value_sequence</td>
<td> - </td>
</tr>
<tr>
<td>2D Convolution Layer </td>
<td> paddle.layer.image_conv_layer</td>
<td> - </td>
</tr>
<tr>
<td>DataType Converter (vec2seq)</td>
<td> paddle.layer.block_expand</td>
<td> - </td>
</tr>
<tr>
<td>Bi-/Uni-directional RNNs </td>
<td>paddle.layer.recurrent_group</td>
<td> - </td>
</tr>
<tr>
<td>Row Convolution Layer </td>
<td>Not supported yet.</td>
<td>TBD (Task 4)</td>
</tr>
<tr>
<td>CTC-loss Layer </td>
<td>paddle.layer.warp_ctc</td>
<td> - </td>
</tr>
<tr>
<td>Batch Normalization Layer </td>
<td>paddle.layer.batch_norm</td>
<td> - </td>
</tr>
<tr>
<td>CTC-Beam search </td>
<td>Not supported yet.</td>
<td> TBD (Task 6) </td>
</tr>
</tbody>
</table>
### Row Convolution
......@@ -145,14 +212,14 @@ TODO by Assignees
Figure 2. Algorithm for CTC Beam Search Decoder.
</div>
- The **Beam Search Decoder** for DS2 CTC-trained network follows the similar approach in \[[3](#references)\] as shown in Figure 2, with two important modifications for the ambiguous parts:
- 1) in the iterative computation of probabilities, the assignment operation is changed to accumulation for one prefix may comes from different paths;
- The **Beam Search Decoder** for DS2 CTC-trained network follows the similar approach in \[[3](#references)\] as shown in Figure 2, with two important modifications for the ambiguous parts:
- 1) in the iterative computation of probabilities, the assignment operation is changed to accumulation for one prefix may comes from different paths;
- 2) the if condition ```if l^+ not in A_prev then``` after probabilities' computation is deprecated for it is hard to understand and seems unnecessary.
- An **external scorer** would be passed into the decoder to evaluate a candidate prefix during decoding whenever a white space appended in English decoding and any character appended in Mandarin decoding.
- Such external scorer consists of language model, word count or any other custom scorers.
- The **language model** is built from Task 5, with parameters should be carefully tuned to achieve minimum WER/CER (c.f. Task 7)
- This decoder needs to perform with **high efficiency** for the convenience of parameters tuning and speech recognition in reality.
- This decoder needs to perform with **high efficiency** for the convenience of parameters tuning and speech recognition in reality.
## Future Work
......
doc/fluid/dev/index_cn.rst
浏览文件 @ b853ac8c
......@@ -4,9 +4,9 @@
.. toctree::
:maxdepth: 1
new_op_en.md
new_op_kernel_en.md
use_eigen_en.md
new_op_cn.md
new_op_kernel.md
use_eigen_cn.md
name_convention.md
support_new_device.md
releasing_process.md
......
doc/fluid/dev/index_en.rst
浏览文件 @ b853ac8c
......@@ -5,7 +5,7 @@ Development
:maxdepth: 1
new_op_en.md
new_op_kernel_en.md
new_op_kernel.md
use_eigen_en.md
name_convention.md
support_new_device.md
......
doc/fluid/dev/new_op_cn.md
浏览文件 @ b853ac8c
......@@ -26,13 +26,32 @@
依据是否包含kernel，可以将Op分为两种：包含Kernel的Op和不包含kernel的Op，前者Op的定义继承自`OperatorWithKernel`，后者继承自`OperatorBase`。本教程主要介绍带Kernel的Op如何写，简单总结Op需要包含的内容如下：
内容 | 定义位置
-------------- | :----------------------
OpProtoMake定义 | `.cc`文件，Backward Op不需要定义OpProtoMake
Op定义 | `.cc`文件
Kernel实现 | CPU、CUDA共享Kernel实现在`.h`文件中，否则，CPU 实现在`.cc`文件中，CUDA 实现在`.cu`文件中。
注册Op | Op注册实现在`.cc`文件；Kernel注册CPU实现在`.cc`文件中，CUDA实现在`.cu`文件中
<table>
<thead>
<tr>
<th>内容</th>
<th>定义位置</th>
</tr>
</thead>
<tbody>
<tr>
<td>OpProtoMake定义 </td>
<td>`.cc`文件，Backward Op不需要定义OpProtoMake </td>
</tr>
<tr>
<td>Op定义 </td>
<td> `.cc`文件</td>
</tr>
<tr>
<td>Kernel实现 </td>
<td> CPU、CUDA共享Kernel实现在`.h`文件中，否则，CPU 实现在`.cc`文件中，CUDA 实现在`.cu`文件中。</td>
</tr>
<tr>
<td>注册Op </td>
<td> Op注册实现在`.cc`文件；Kernel注册CPU实现在`.cc`文件中，CUDA实现在`.cu`文件中</td>
</tr>
</tbody>
</table>
实现新的op都添加至目录[paddle/operators](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/operators)下，文件命名以`*_op.h`（如有） 、 `*_op.cc` 、`*_op.cu`（如有）结尾。**系统会根据文件名自动构建op和其对应的Python扩展。**
......
doc/fluid/dev/new_op_en.md
浏览文件 @ b853ac8c
......@@ -33,6 +33,33 @@ Op definition | `.cc` files
Kernel implementation | The kernel methods shared between CPU and CUDA are defined in `.h` files. CPU-specific kernels live in `.cc` files, while CUDA-specific kernels are implemented in `.cu`files.
Registering the Op | Ops are registered in `.cc` files; For Kernel registration, `.cc` files contain the CPU implementation, while `.cu` files contain the CUDA implementation.
<table>
<thead>
<tr>
<th>Information</th>
<th> Where is it defined</th>
</tr>
</thead>
<tbody>
<tr>
<td>OpProtoMake definition </td>
<td> `.cc`files, Backward Op does not need an OpProtoMake interface. </td>
</tr>
<tr>
<td>Op definition </td>
<td> `.cc` files</td>
</tr>
<tr>
<td>Kernel implementation </td>
<td> The kernel methods shared between CPU and CUDA are defined in `.h` files. CPU-specific kernels live in `.cc` files, while CUDA-specific kernels are implemented in `.cu`files.</td>
</tr>
<tr>
<td>Registering the Op </td>
<td> Ops are registered in `.cc` files; For Kernel registration, `.cc` files contain the CPU implementation, while `.cu` files contain the CUDA implementation.</td>
</tr>
</tbody>
</table>
New Operator implementations are added to the list [paddle/operators](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/operators), with file names in the format `*_op.h` (if applicable), `*_op.cc`, `*_op.cu` (if applicable).** The system will use the naming scheme to automatically build operators and their corresponding Python extensions.**
......@@ -279,7 +306,7 @@ A forward operator unit test inherits `unittest.TestCase` and defines metaclass
def test_check_output(self):
self.check_output()
def test_check_grad_normal(self):
self.check_grad(['X', 'Y'], 'Out', max_relative_error=0.5)
......
doc/fluid/dev/releasing_process.md
浏览文件 @ b853ac8c
......@@ -66,7 +66,7 @@ PaddlePaddle开发过程使用[git-flow](http://nvie.com/posts/a-successful-git-
* 建议，开发者fork的版本库使用`develop`分支同步主版本库的`develop`分支
* 建议，开发者fork的版本库中，再基于`develop`版本fork出自己的功能分支。
* 当功能分支开发完毕后，向PaddlePaddle的主版本库提交`Pull Reuqest`，进而进行代码评审。
* 在评审过程中，开发者修改自己的代码，可以继续在自己的功能分支提交代码。
* 在评审过程中，开发者修改自己的代码，可以继续在自己的功能分支提交代码。
* BugFix分支也是在开发者自己的fork版本库维护，与功能分支不同的是，BugFix分支需要分别给主版本库的`master`、`develop`与可能有的`release/版本号`分支，同时提起`Pull Request`。
......@@ -78,13 +78,116 @@ PaddlePaddle开发过程使用[git-flow](http://nvie.com/posts/a-successful-git-
PaddlePaddle每次发版本首先要保证PaddlePaddle Book中所有章节功能的正确性。功能的正确性包括验证PaddlePaddle目前的`paddle_trainer`训练和纯使用`Python`训练模型正确性。
| | 新手入门章节 | 识别数字 | 图像分类 | 词向量 | 情感分析 | 语意角色标注 | 机器翻译 | 个性化推荐 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| API.V2 + Docker + GPU | | | | | | | | |
| API.V2 + Docker + CPU | | | | | | | | |
| `paddle_trainer` + Docker + GPU | | | | | | | | |
| `paddle_trainer` + Docker + CPU | | | | | | | | |
| API.V2 + Ubuntu + GPU | | | | | | | | |
| API.V2 + Ubuntu + CPU | | | | | | | | |
| `paddle_trainer` + Ubuntu + GPU | | | | | | | | |
| `paddle_trainer` + Ubuntu + CPU | | | | | | | | |
<table>
<thead>
<tr>
<th></th>
<th>新手入门章节 </th>
<th> 识别数字</th>
<th> 图像分类</th>
<th>词向量</th>
<th> 情感分析</th>
<th>语意角色标注</th>
<th> 机器翻译</th>
<th>个性化推荐</th>
</tr>
</thead>
<tbody>
<tr>
<td>API.V2 + Docker + GPU </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
</tr>
<tr>
<td> API.V2 + Docker + CPU </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
</tr>
<tr>
<td>`paddle_trainer` + Docker + GPU </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
</tr>
<tr>
<td>`paddle_trainer` + Docker + CPU </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
</tr>
<tr>
<td> API.V2 + Ubuntu + GPU</td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
</tr>
<tr>
<td>API.V2 + Ubuntu + CPU </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
</tr>
<tr>
<td> `paddle_trainer` + Ubuntu + GPU</td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
</tr>
<tr>
<td> `paddle_trainer` + Ubuntu + CPU</td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
<td> </td>
</tr>
</tbody>
</table>
doc/fluid/getstarted/concepts/save_model/model_format.md
浏览文件 @ b853ac8c
......@@ -4,30 +4,70 @@
A model is an output of the training process. One complete model consists of two parts, the **topology** and the **parameters**. In order to support industrial deployment, the model format must be self-complete and must not expose any training source code.
As a result, In PaddlePaddle, the **topology** is represented as a [ProgramDesc](https://github.com/PaddlePaddle/Paddle/blob/1c0a4c901c9fc881d120249c703b15d1c50dae7d/doc/design/program.md), which describes the model structure. The **parameters** contain all the trainable weights in the model. We must support large size parameters and efficient serialization/deserialization of parameters.
As a result, In PaddlePaddle, the **topology** is represented as a [ProgramDesc](https://github.com/PaddlePaddle/Paddle/blob/1c0a4c901c9fc881d120249c703b15d1c50dae7d/doc/design/program.md), which describes the model structure. The **parameters** contain all the trainable weights in the model. We must support large size parameters and efficient serialization/deserialization of parameters.
## Implementation
The topology is saved as a plain text in a detailed self-contain protobuf file.
The topology is saved as a plain text in a detailed self-contain protobuf file.
The parameters are saved as a binary file. As we all know, the protobuf message has a limit of [64M size](https://developers.google.com/protocol-buffers/docs/reference/cpp/google.protobuf.io.coded_stream#CodedInputStream.SetTotalBytesLimit.details). We have done a [benchmark experiment](https://github.com/PaddlePaddle/Paddle/pull/4610), which shows that protobuf is not fit for the task.
As a result, we design a particular format for tensor serialization. By default, an arbitrary tensor in Paddle is a [LoDTensor](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/lod_tensor.md), and has a description information proto of [LoDTensorDesc](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/framework.proto#L99). We save the DescProto as the byte string header. It contains all the necessary information, such as the `dims`, and the `LoD` information in [LoDTensor](https://github.com/PaddlePaddle/Paddle/blob/1c0a4c901c9fc881d120249c703b15d1c50dae7d/paddle/framework/lod_tensor.md). A tensor stores values in a continuous memory buffer. For speed we dump the raw memory to disk and save it as the byte string content. So, the binary format of one tensor is,
As a result, we design a particular format for tensor serialization. By default, an arbitrary tensor in Paddle is a [LoDTensor](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/lod_tensor.md), and has a description information proto of [LoDTensorDesc](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/framework.proto#L99). We save the DescProto as the byte string header. It contains all the necessary information, such as the `dims`, and the `LoD` information in [LoDTensor](https://github.com/PaddlePaddle/Paddle/blob/1c0a4c901c9fc881d120249c703b15d1c50dae7d/paddle/framework/lod_tensor.md). A tensor stores values in a continuous memory buffer. For speed we dump the raw memory to disk and save it as the byte string content. So, the binary format of one tensor is,
The table below shows a tensor's byte view in detail. Note that all the signed values are written in the little-endian format.
|field name | type | description |
| --- | --- | --- |
| version | uint32_t | Version of saved file. Always 0 now. |
| tensor desc length | uint32_t | TensorDesc(Protobuf message) length in bytes. |
| tensor desc | void* | TensorDesc protobuf binary message |
| tensor data | void* | Tensor's data in binary format. The length of `tensor_data` is decided by `TensorDesc.dims()` and `TensorDesc.data_type()` |
| lod_level | uint64_t | Level of LoD |
| length of lod[0] | uint64_t | [Optional] length of lod[0] in bytes. |
| data of lod[0] | uint64_t* | [Optional] lod[0].data() |
| ... | ... | ... |
<table>
<thead>
<tr>
<th>field name</th>
<th>type </th>
<th>description </th>
</tr>
</thead>
<tbody>
<tr>
<td> version</td>
<td> uint32_t </td>
<td> Version of saved file. Always 0 now.</td>
</tr>
<tr>
<td> tensor desc length </td>
<td> uint32_t </td>
<td> TensorDesc(Protobuf message) length in bytes. </td>
</tr>
<tr>
<td>tensor desc </td>
<td> void*</td>
<td> TensorDesc protobuf binary message </td>
</tr>
<tr>
<td> tensor data </td>
<td> void* </td>
<td> Tensor's data in binary format. The length of `tensor_data` is decided by `TensorDesc.dims()` and `TensorDesc.data_type()` </td>
</tr>
<tr>
<td> lod_level</td>
<td> uint64_t </td>
<td> Level of LoD </td>
</tr>
<tr>
<td> length of lod[0] </td>
<td> uint64_t </td>
<td> [Optional] length of lod[0] in bytes. </td>
</tr>
<tr>
<td> data of lod[0] </td>
<td> uint64_t* </td>
<td> [Optional] lod[0].data() </td>
</tr>
<tr>
<td>... </td>
<td> ... </td>
<td> ... </td>
</tr>
</tbody>
</table>
## Summary
......
doc/fluid/howto/cluster/fluid_cluster_train_cn.md
浏览文件 @ b853ac8c
......@@ -65,10 +65,10 @@ exit(1)
**因此，在分布式的Fluid环境中，我们有两个角色需要创建，分别是Parameter Server和Trainer。**
### 分布式训练
### 分布式训练
Fliud专门提供了工具[Distributed Transpiler](https://github.com/PaddlePaddle/Paddle/blob/ba65d54d9d3b41cd3c5171b00f476d4e60133ddb/doc/fluid/design/dist_train/distributed_architecture.md#distributed-transpiler)用于将单机版的训练程序转换为分布式版本的训练程序。工具背后的理念是找出程序的优化算子和梯度参数，将他们分隔为两部分，通过send/recv 操作算子进行连接,优化算子和梯度参数可以在优化器的minimize函数的返回值中获取到。
```python
optimize_ops, params_grads = sgd_optimizer.minimize(avg_cost)
optimize_ops, params_grads = sgd_optimizer.minimize(avg_cost)
```
将Distributed Transpiler、优化算子和梯度函数放在一个代码中如下：
```python
......@@ -99,15 +99,51 @@ for pass_id in range(100):
### 分布式训练脚本运行说明
分布式任务的运行需要将表格中说明的多个参数进行赋值:
| 参数名 | 值类型 | 说明 | 示例 |
|:-------------|:------|:---------------------------------------|:-------------|
| trainer_id | int | 当前训练节点的ID，训练节点ID编号为0 - n-1， n为trainers的值 | 0/1/2/3 |
| pservers | str | parameter server 列表 | 127.0.0.1:6710,127.0.0.1:6711 |
| trainers | int | 训练节点的总个数，>0的数字 | 4 |
| server_endpoint | str | 当前所起的服务节点的IP:PORT | 127.0.0.1:8789 |
| training_role | str | 节点角色， TRAINER/PSERVER | PSERVER |
**注意：** ```training_role```是用来区分当前所起服务的角色的，用于训练程序中，用户可根据需要自行定义，其他参数为fluid.DistributeTranspiler的transpile函数所需要，需要在调用函数前进行定义，样例如下：
<table>
<thead>
<tr>
<th>参数名</th>
<th> 值类型</th>
<th>说明</th>
<th> 示例</th>
</tr>
</thead>
<tbody>
<tr>
<td>trainer_id </td>
<td> int</td>
<td> 当前训练节点的ID，训练节点ID编号为0 - n-1， n为trainers的值 </td>
<td> 0/1/2/3 </td>
</tr>
<tr>
<td>pservers </td>
<td> str</td>
<td> parameter server 列表 </td>
<td> 127.0.0.1:6710,127.0.0.1:6711 </td>
</tr>
<tr>
<td>trainers </td>
<td>int </td>
<td> 训练节点的总个数，>0的数字 </td>
<td> 4 </td>
</tr>
<tr>
<td> server_endpoint</td>
<td> str </td>
<td> 当前所起的服务节点的IP:PORT </td>
<td> 127.0.0.1:8789 </td>
</tr>
<tr>
<td> training_role</td>
<td>str </td>
<td> 节点角色， TRAINER/PSERVER </td>
<td> PSERVER </td>
</tr>
</tbody>
</table>
**注意：** ```training_role```是用来区分当前所起服务的角色的，用于训练程序中，用户可根据需要自行定义，其他参数为fluid.DistributeTranspiler的transpile函数所需要，需要在调用函数前进行定义，样例如下：
```python
t = fluid.DistributeTranspiler()
......
doc/fluid/howto/optimization/cpu_profiling_cn.md
浏览文件 @ b853ac8c
......@@ -42,14 +42,40 @@ cprofilev -a 0.0.0.0 -p 3214 -f profile.out main.py
每一列的含义是:
| 列名 | 含义 |
| --- | --- |
| ncalls | 函数的调用次数 |
| tottime | 函数实际使用的总时间。该时间去除掉本函数调用其他函数的时间 |
| percall | tottime的每次调用平均时间 |
| cumtime | 函数总时间。包含这个函数调用其他函数的时间 |
| percall | cumtime的每次调用平均时间 |
| filename:lineno(function) | 文件名, 行号，函数名 |
<table>
<thead>
<tr>
<th>列名</th>
<th>含义 </th>
</tr>
</thead>
<tbody>
<tr>
<td> ncalls</td>
<td> 函数的调用次数</td>
</tr>
<tr>
<td>tottime</td>
<td> 函数实际使用的总时间。该时间去除掉本函数调用其他函数的时间</td>
</tr>
<tr>
<td> percall </td>
<td> tottime的每次调用平均时间</td>
</tr>
<tr>
<td> cumtime</td>
<td> 函数总时间。包含这个函数调用其他函数的时间</td>
</tr>
<tr>
<td> percall</td>
<td> cumtime的每次调用平均时间</td>
</tr>
<tr>
<td> filename:lineno(function) </td>
<td> 文件名, 行号，函数名 </td>
</tr>
</tbody>
</table>
### 寻找性能瓶颈
......
doc/fluid/howto/optimization/cpu_profiling_en.md
浏览文件 @ b853ac8c
......@@ -57,14 +57,40 @@ port, we will see the output like the following:
where each line corresponds to Python function, and the meaning of
each column is as follows:
| column | meaning |
| --- | --- |
| ncalls | the number of calls into a function |
| tottime | the total execution time of the function, not including the execution time of other functions called by the function |
| percall | tottime divided by ncalls |
| cumtime | the total execution time of the function, including the execution time of other functions being called |
| percall | cumtime divided by ncalls |
| filename:lineno(function) | where the function is defined |
<table>
<thead>
<tr>
<th>column</th>
<th>meaning </th>
</tr>
</thead>
<tbody>
<tr>
<td> ncalls</td>
<td> the number of calls into a function</td>
</tr>
<tr>
<td>tottime</td>
<td> the total execution time of the function, not including the execution time of other functions called by the function</td>
</tr>
<tr>
<td> percall </td>
<td> tottime divided by ncalls</td>
</tr>
<tr>
<td> cumtime</td>
<td> the total execution time of the function, including the execution time of other functions being called</td>
</tr>
<tr>
<td> percall</td>
<td> cumtime divided by ncalls</td>
</tr>
<tr>
<td> filename:lineno(function) </td>
<td> where the function is define </td>
</tr>
</tbody>
</table>
### Identify Performance Bottlenecks
......
doc/v2/faq/build_and_install/index_en.rst
浏览文件 @ b853ac8c
############################
Install, Build and Unit test
############################
.. _install_faq:
TBD
###############################
Compile, Install, and Unit Test
###############################
.. contents::
1. Insufficient CUDA driver version
----------------------------------------------------------------
Many users usually face issues like `Cuda Error: CUDA driver version is insufficient for CUDA runtime version` when running the PaddlePaddle GPU Docker image. The cause is that you may not map the local CUDA driver to a container directory.
You can solve the issue by running the following commands:
.. code-block:: bash
$ export CUDA_SO="$(\ls usr/lib64/libcuda* | xargs -I{} echo '-v {}:{}') $(\ls /usr/lib64/libnvidia* | xargs -I{} echo '-v {}:{}')"
$ export DEVICES=$(\ls /dev/nvidia* | xargs -I{} echo '--device {}:{}')
$ docker run ${CUDA_SO} ${DEVICES} -it paddlepaddle/paddle:latest-gpu
For more infomation about Docker's installation and usage, please refer to `PaddlePaddle Docker documentation <http://www.paddlepaddle.org/docs/0.11.0/documentation/zh/getstarted/build_and_install/docker_install_en.html>`_ .
2. Version mismatch between PythonLibs and PythonInterpreter
----------------------------------------------------------------
It is a common bug when CMake looks up Python. If you install multiple versions of Python, Cmake may find the version mismatch between PythonLibs and PythonInterpreter . You are forced to specify a Python version, as follows.
.. code-block:: bash
cmake .. -DPYTHON_EXECUTABLE=<exc_path> -DPYTHON_LIBRARY=<lib_path> -DPYTHON_INCLUDE_DIR=<inc_path>
You should specify ``<exc_path>``, ``<lib_path>``, ``<inc_path>`` to your local paths.
3. PaddlePaddle version is 0.0.0
------------------------------------------------
This issue would happen when you run the code `paddle version` or `cmake ..`
.. code-block:: bash
CMake Warning at cmake/version.cmake:20 (message):
Cannot add paddle version from git tag
You should pull all remote branches to your local machine with the command :code:`git fetch upstream` and then run :code:`cmake`
4. paddlepaddle\*.whl is not a supported wheel on this platform.
------------------------------------------------------------------------
The primary cause for this issue is that it can not find the correct PaddlePaddle installation package that matches your current system.The latest PaddlePaddle Python installation package supports Linux x86_64 and MacOS 10.12 os including Python2.7 and Pip 9.0.1.
You can upgrade Pip with the following command\:
.. code-block:: bash
pip install --upgrade pip
If it does not work for you, you can run the command :code:`python -c "import pip; print(pip.pep425tags.get_supported())"` to get the suffix of Python package which your system may support and then compare it with the suffix of your installation.
If the system supports :code:`linux_x86_64` and the installation package is :code:`manylinux1_x86_64`, you should upgrade pip to the latest
if the system supports :code:`manylinux_x86_64` and the local installation package is :code:`linux1_x86_64`, you can rename the whl package to :code:`manylinux1_x86_64` and then try again.
5. ImportError: No module named v2
----------------------------------
Please uninstall Paddle V1 if you have installed it before.
.. code-block:: bash
pip uninstall py_paddle paddle
Then install Python for PaddlePaddle , enter the build directory and run the following commands
pip install python/dist/paddle*.whl && pip install ../paddle/dist/py_paddle*.whl
6. Illegal instruction
-----------------------
This issue may be caused by the wrong usage of PaddlePaddle binary version which uses avx SIMD instructions to increase the performance of cpu. Please choose the correct version.
7. Python unittest fails
--------------------------------
If the following python unittest testcases fail:
.. code-block:: bash
24 - test_PyDataProvider (Failed)
26 - test_RecurrentGradientMachine (Failed)
27 - test_NetworkCompare (Failed)
28 - test_PyDataProvider2 (Failed)
32 - test_Prediction (Failed)
33 - test_Compare (Failed)
34 - test_Trainer (Failed)
35 - test_TrainerOnePass (Failed)
36 - test_CompareTwoNets (Failed)
37 - test_CompareTwoOpts (Failed)
38 - test_CompareSparse (Failed)
39 - test_recurrent_machine_generation (Failed)
40 - test_PyDataProviderWrapper (Failed)
41 - test_config_parser (Failed)
42 - test_swig_api (Failed)
43 - layers_test (Failed)
Please check the PaddlePaddle unittest logs which may suggest the following:
.. code-block:: bash
paddle package is already in your PYTHONPATH. But unittest need a clean environment.
Please uninstall paddle package before start unittest. Try to 'pip uninstall paddle'.
The solution is:
* Remove old PaddlePaddle to make a clean environment for the unit tests. If PaddlePaddle package is already in Python's site-packages, unit tests would refer Python package in site-packages instead of Python package in the :code:`/python` directory of the source directory. Setting :code:`PYTHONPATH` to :code:`/python` is also useless because Python's search path would give the priority to the installed Python package.
8. Failed to download the MKLML library
----------------------------------------------
.. code-block:: bash
make[2]: *** [third_party/mklml/src/extern_mklml-stamp/extern_mklml-download] error 4
make[1]: *** [CMakeFiles/extern_mklml.dir/all] error 2
make[1]: *** waiting for the unfinished jobs....
Cause: The network speed or SSL link causes the MKLML library to download unsuccessfully.
The solution is: manually download and install, the specific steps are as follows.
.. code-block:: bash
// 1. enter the directory
cd build/third_party/mklml/src/extern_mklml
// 2. check the size of the package, normally 75M, if less than 75M, the download fails
du -sh mklml_lnx_2018.0.1.20171007.tgz
// 3. manually download and unzip and make the download success tag:
wget --no-check-certificate https://github.com/01org/mkl-dnn/releases/download/v0.11/mklml_lnx_2018.0.1.20171007.tgz -c -O mklml_lnx_2018.0.1.20171007.tgz
tar zxf mklml_lnx_2018.0.1.20171007.tgz
touch ../extern_mklml-stamp/extern_mklml-download
// 4. then compile
paddle/fluid/framework/details/multi_devices_graph_builder.cc
浏览文件 @ b853ac8c
......@@ -55,6 +55,7 @@ std::unique_ptr<SSAGraph> MultiDevSSAGraphBuilder::Build(
const ProgramDesc &program) const {
auto graph = new SSAGraph();
SSAGraph &result = *graph;
std::unordered_set<std::string> og_has_been_broadcast;
result.vars_.resize(places_.size());
bool is_forwarding = true;
......@@ -122,9 +123,15 @@ std::unique_ptr<SSAGraph> MultiDevSSAGraphBuilder::Build(
if (!is_forwarding) {
auto var_names = op->OutputArgumentNames();
// Currently, we assume that once gradient is generated, it can be
// broadcast, and each gradient is only broadcast once. But there are no
// other cases, for example, we need to adjust the gradient according to
// the input when we get the gradient, which is not considered at present.
for (auto &og : var_names) {
if (grad_names_.count(og) != 0) { // is param grad
// Insert NCCL AllReduce Op
if (grad_names_.count(og) != 0 &&
og_has_been_broadcast.count(og) == 0) { // is param grad
// Insert NCCL AllReduce Op
og_has_been_broadcast.insert(og);
#ifdef PADDLE_WITH_CUDA
result.ops_.emplace_back(
new NCCLAllReduceOpHandle(local_scopes_, places_, *nccl_ctxs_));
......
paddle/fluid/framework/selected_rows.h
浏览文件 @ b853ac8c
......@@ -10,6 +10,9 @@ See the License for the specific language governing permissions and
limitations under the License. */
#pragma once
#include <vector>
#include "paddle/fluid/framework/lod_tensor.h"
#include "paddle/fluid/framework/tensor.h"
......@@ -52,7 +55,7 @@ class SelectedRows {
private:
// Notice: rows can be duplicate. We can have {0, 4, 7, 0, 5, 7, 9} here.
// SelectedRows are simplely concated when adding together. Until a
// SelectedRows are simply concated when adding together. Until a
// SelectedRows add a Tensor, will the duplicate rows be handled.
Vector<int64_t> rows_;
std::unique_ptr<Tensor> value_{nullptr};
......
paddle/fluid/operators/lookup_table_op.cc
浏览文件 @ b853ac8c
......@@ -18,6 +18,22 @@ limitations under the License. */
namespace paddle {
namespace operators {
static inline framework::OpKernelType ExpectedKernelType(
const framework::ExecutionContext& ctx) {
auto* table_var = ctx.InputVar("W");
if (table_var->IsType<LoDTensor>()) {
return framework::OpKernelType(
framework::ToDataType(table_var->Get<LoDTensor>().type()),
ctx.device_context());
} else if (table_var->IsType<SelectedRows>()) {
return framework::OpKernelType(
framework::ToDataType(table_var->Get<SelectedRows>().value().type()),
ctx.device_context());
} else {
PADDLE_THROW("W should be LoDTensor or SelectedRows");
}
}
class LookupTableOp : public framework::OperatorWithKernel {
public:
using framework::OperatorWithKernel::OperatorWithKernel;
......@@ -51,9 +67,7 @@ class LookupTableOp : public framework::OperatorWithKernel {
protected:
framework::OpKernelType GetExpectedKernelType(
const framework::ExecutionContext& ctx) const override {
return framework::OpKernelType(
framework::ToDataType(ctx.Input<LoDTensor>("W")->type()),
ctx.device_context());
return ExpectedKernelType(ctx);
}
};
......@@ -84,7 +98,7 @@ class LookupTableOpMaker : public framework::OpProtoAndCheckerMaker {
"If the value is -1, it makes no effect to lookup. "
"Otherwise the given value indicates padding the output "
"with zeros whenever lookup encounters it in Ids.")
.SetDefault(-1);
.SetDefault(kNoPadding);
AddComment(R"DOC(
Lookup Table Operator.
......@@ -124,9 +138,7 @@ class LookupTableOpGrad : public framework::OperatorWithKernel {
protected:
framework::OpKernelType GetExpectedKernelType(
const framework::ExecutionContext& ctx) const override {
return framework::OpKernelType(
framework::ToDataType(ctx.Input<LoDTensor>("W")->type()),
ctx.device_context());
return ExpectedKernelType(ctx);
}
};
......
paddle/fluid/operators/lookup_table_op.h
浏览文件 @ b853ac8c
......@@ -14,6 +14,9 @@ limitations under the License. */
#pragma once
#include <string>
#include <vector>
#include "paddle/fluid/framework/eigen.h"
#include "paddle/fluid/framework/lod_tensor.h"
#include "paddle/fluid/framework/op_registry.h"
......@@ -25,16 +28,37 @@ namespace operators {
using Tensor = framework::Tensor;
using LoDTensor = framework::LoDTensor;
using SelectedRows = framework::SelectedRows;
using DDim = framework::DDim;
static constexpr int64_t kNoPadding = -1;
inline size_t getIndex(const std::vector<int64_t> &rows, int64_t value) {
auto it = std::find(rows.begin(), rows.end(), value);
PADDLE_ENFORCE(it != rows.end(), "id should be in rows");
return static_cast<size_t>(std::distance(rows.begin(), it));
}
template <typename T>
class LookupTableKernel : public framework::OpKernel<T> {
public:
void Compute(const framework::ExecutionContext& context) const override {
auto* table_t = context.Input<LoDTensor>("W");
auto* ids_var = context.InputVar("Ids");
Tensor* output_t = context.Output<Tensor>("Out");
void Compute(const framework::ExecutionContext &context) const override {
auto *table_var = context.InputVar("W");
auto *ids_var = context.InputVar("Ids");
Tensor *output_t = context.Output<Tensor>("Out");
int64_t padding_idx = context.Attr<int64_t>("padding_idx");
DDim table_dim;
int64_t* ids;
if (table_var->IsType<LoDTensor>()) {
table_dim = context.Input<LoDTensor>("W")->dims();
} else if (table_var->IsType<SelectedRows>()) {
auto *table_t = context.Input<SelectedRows>("W");
table_dim = table_t->value().dims();
} else {
PADDLE_THROW("table only support LoDTensor and SelectedRows");
}
int64_t *ids;
int64_t ids_numel;
// The type of Ids(Input) is SelectedRows or LoDTensor, when Ids's type
......@@ -42,39 +66,50 @@ class LookupTableKernel : public framework::OpKernel<T> {
// when Ids's type is SelectedRows, the rows of Ids contains the
// ids to be looked up in W.
if (ids_var->IsType<LoDTensor>()) {
auto* ids_t = context.Input<LoDTensor>("Ids");
ids = const_cast<int64_t*>(ids_t->data<int64_t>());
auto *ids_t = context.Input<LoDTensor>("Ids");
ids = const_cast<int64_t *>(ids_t->data<int64_t>());
ids_numel = ids_t->numel();
} else if (ids_var->IsType<SelectedRows>()) {
auto* ids_t = context.Input<SelectedRows>("Ids");
ids = const_cast<int64_t*>(ids_t->rows().data());
auto *ids_t = context.Input<SelectedRows>("Ids");
ids = const_cast<int64_t *>(ids_t->rows().data());
ids_numel = ids_t->rows().size();
output_t->Resize({ids_numel, table_t->dims()[1]});
output_t->Resize({ids_numel, table_dim[1]});
} else {
PADDLE_THROW("Unsupported Variable Type of Ids");
}
int64_t padding_idx = context.Attr<int64_t>("padding_idx");
if (table_var->IsType<LoDTensor>()) {
auto *table_t = context.Input<LoDTensor>("W");
int64_t row_number = table_t->dims()[0];
int64_t row_width = table_t->dims()[1];
int N = table_t->dims()[0];
int D = table_t->dims()[1];
auto* table = table_t->data<T>();
auto* output = output_t->mutable_data<T>(context.GetPlace());
auto *table = table_t->data<T>();
auto *output = output_t->mutable_data<T>(context.GetPlace());
if (padding_idx == -1) {
for (int64_t i = 0; i < ids_numel; ++i) {
PADDLE_ENFORCE_LT(ids[i], N);
PADDLE_ENFORCE_GE(ids[i], 0);
memcpy(output + i * D, table + ids[i] * D, D * sizeof(T));
if (padding_idx != kNoPadding && ids[i] == padding_idx) {
memset(output + i * row_width, 0, row_width * sizeof(T));
} else {
PADDLE_ENFORCE_LT(ids[i], row_number);
PADDLE_ENFORCE_GE(ids[i], 0);
memcpy(output + i * row_width, table + ids[i] * row_width,
row_width * sizeof(T));
}
}
} else {
} else if (table_var->IsType<SelectedRows>()) {
const auto &table_t = table_var->Get<SelectedRows>();
int64_t row_width = table_t.value().dims()[1];
const auto *table = table_t.value().data<T>();
auto *output = output_t->mutable_data<T>(context.GetPlace());
for (int64_t i = 0; i < ids_numel; ++i) {
if (ids[i] == padding_idx) {
memset(output + i * D, 0, D * sizeof(T));
if (padding_idx != kNoPadding && ids[i] == padding_idx) {
memset(output + i * row_width, 0, row_width * sizeof(T));
} else {
PADDLE_ENFORCE_LT(ids[i], N);
PADDLE_ENFORCE_GE(ids[i], 0);
memcpy(output + i * D, table + ids[i] * D, D * sizeof(T));
auto id_index = getIndex(table_t.rows(), ids[i]);
memcpy(output + i * row_width, table + id_index * row_width,
row_width * sizeof(T));
}
}
}
......@@ -84,17 +119,27 @@ class LookupTableKernel : public framework::OpKernel<T> {
template <typename T>
class LookupTableGradKernel : public framework::OpKernel<T> {
public:
void Compute(const framework::ExecutionContext& context) const override {
void Compute(const framework::ExecutionContext &context) const override {
auto *table_var = context.InputVar("W");
DDim table_dim;
if (table_var->IsType<LoDTensor>()) {
table_dim = context.Input<LoDTensor>("W")->dims();
} else if (table_var->IsType<SelectedRows>()) {
auto *table_t = context.Input<SelectedRows>("W");
table_dim = table_t->value().dims();
} else {
PADDLE_THROW("table only support LoDTensor and SelectedRows");
}
bool is_sparse = context.Attr<bool>("is_sparse");
// Since paddings are not trainable and fixed in forward, the gradient of
// paddings makes no sense and we don't deal with it in backward.
if (is_sparse) {
auto* ids = context.Input<LoDTensor>("Ids");
auto* table = context.Input<LoDTensor>("W");
auto* d_output = context.Input<LoDTensor>(framework::GradVarName("Out"));
auto* d_table = context.Output<SelectedRows>(framework::GradVarName("W"));
auto *ids = context.Input<LoDTensor>("Ids");
auto *d_output = context.Input<LoDTensor>(framework::GradVarName("Out"));
auto *d_table = context.Output<SelectedRows>(framework::GradVarName("W"));
auto* ids_data = ids->data<int64_t>();
auto *ids_data = ids->data<int64_t>();
auto ids_dim = ids->dims();
framework::Vector<int64_t> new_rows;
......@@ -104,31 +149,30 @@ class LookupTableGradKernel : public framework::OpKernel<T> {
}
d_table->set_rows(new_rows);
auto* d_table_value = d_table->mutable_value();
d_table_value->Resize({ids_dim[0], table->dims()[1]});
auto *d_table_value = d_table->mutable_value();
d_table_value->Resize({ids_dim[0], table_dim[1]});
d_table_value->mutable_data<T>(context.GetPlace());
d_table->set_height(table->dims()[0]);
d_table->set_height(table_dim[0]);
auto* d_output_data = d_output->data<T>();
auto* d_table_data = d_table_value->data<T>();
auto *d_output_data = d_output->data<T>();
auto *d_table_data = d_table_value->data<T>();
PADDLE_ENFORCE_EQ(d_table_value->dims(), d_output->dims());
memcpy(d_table_data, d_output_data, sizeof(T) * d_output->numel());
} else {
auto* ids = context.Input<LoDTensor>("Ids");
auto* d_output = context.Input<LoDTensor>(framework::GradVarName("Out"));
auto* d_table = context.Output<LoDTensor>(framework::GradVarName("W"));
auto* table = context.Input<LoDTensor>("W");
auto *ids = context.Input<LoDTensor>("Ids");
auto *d_output = context.Input<LoDTensor>(framework::GradVarName("Out"));
auto *d_table = context.Output<LoDTensor>(framework::GradVarName("W"));
auto* ids_data = ids->data<int64_t>();
auto *ids_data = ids->data<int64_t>();
auto ids_dim = ids->dims();
int N = table->dims()[0];
int N = table_dim[0];
int D = d_output->dims()[1];
auto* d_output_data = d_output->data<T>();
auto* d_table_data = d_table->mutable_data<T>(context.GetPlace());
auto *d_output_data = d_output->data<T>();
auto *d_table_data = d_table->mutable_data<T>(context.GetPlace());
memset(d_table_data, 0, d_table->numel() * sizeof(T));
......
paddle/fluid/operators/nccl_op_test.cu.cc
浏览文件 @ b853ac8c
......@@ -137,6 +137,8 @@ class NCCLTester : public ::testing::Test {
TEST_F(NCCLTester, ncclInitOp) {}
// ncclAllReduceOp with desc
// TODO(helin): https://github.com/PaddlePaddle/Paddle/issues/9367
/*
TEST_F(NCCLTester, ncclAllReduceOp) {
std::unique_ptr<f::OpDesc> op2(new f::OpDesc);
op2->SetType("ncclAllReduce");
......@@ -184,6 +186,7 @@ TEST_F(NCCLTester, ncclAllReduceOp) {
}
}
}
*/
// ncclReduceOp with desc
TEST_F(NCCLTester, ncclReduceOp) {
......@@ -236,6 +239,8 @@ TEST_F(NCCLTester, ncclReduceOp) {
}
// ncclBcastOp with desc
// TODO(helin): https://github.com/PaddlePaddle/Paddle/issues/9540
/*
TEST_F(NCCLTester, ncclBcastOp) {
std::unique_ptr<f::OpDesc> op2(new f::OpDesc);
const int kRoot = 0;
......@@ -281,3 +286,4 @@ TEST_F(NCCLTester, ncclBcastOp) {
ASSERT_NEAR(ct[j], result, 1e-5);
}
}
*/
paddle/fluid/operators/reader/create_double_buffer_reader_op.cc
浏览文件 @ b853ac8c
......@@ -20,12 +20,29 @@ namespace paddle {
namespace operators {
namespace reader {
static constexpr size_t kDoubleBufferSize = 2;
// 'Double buffer' means we shall maintain two batches of input data at the same
// time. So the kCacheSize shoul be at least 2.
static constexpr size_t kCacheSize = 2;
// There will be two bacthes out of the channel during training:
// 1. the one waiting to be sent to the channel
// 2. the one just be received from the channel, which is also being used by
// subsequent operators.
// So the channel size should be kChacheSize - 2
static constexpr size_t kChannelSize = 0; // kCacheSize - 2
class DoubleBufferReader : public framework::DecoratedReader {
public:
struct Item {
Item() : ctx_(nullptr) {}
Item(Item&& b) {
payloads_ = std::move(b.payloads_);
ctx_ = std::move(b.ctx_);
}
Item& operator=(Item&& b) {
payloads_ = std::move(b.payloads_);
ctx_ = std::move(b.ctx_);
return *this;
}
std::vector<framework::LoDTensor> payloads_;
platform::DeviceContext* ctx_;
......@@ -34,42 +51,44 @@ class DoubleBufferReader : public framework::DecoratedReader {
explicit DoubleBufferReader(
ReaderBase* reader, platform::Place target_place = platform::CPUPlace())
: DecoratedReader(reader), place_(target_place) {
for (size_t i = 0; i < kDoubleBufferSize; ++i) {
if (platform::is_gpu_place(place_)) {
#ifdef PADDLE_WITH_CUDA
for (size_t i = 0; i < kCacheSize; ++i) {
if (platform::is_gpu_place(place_)) {
ctxs_.emplace_back(new platform::CUDADeviceContext(
boost::get<platform::CUDAPlace>(place_)));
#endif
}
}
start_thread();
}
void start_thread() {
buffer_ = framework::MakeChannel<Item>(kDoubleBufferSize);
prefetcher_ = std::thread([this] { PrefetchThreadFunc(); });
#endif
StartPrefetcher();
}
bool HasNext() const override;
void ReadNext(std::vector<framework::LoDTensor>* out) override;
void ReInit() override;
~DoubleBufferReader() {
buffer_->Close();
prefetcher_.join();
delete buffer_;
~DoubleBufferReader() { EndPrefetcher(); }
private:
void StartPrefetcher() {
channel_ = framework::MakeChannel<Item>(kChannelSize);
prefetcher_ = std::thread([this] { PrefetchThreadFunc(); });
}
bool HasNext() const override;
void EndPrefetcher() {
channel_->Close();
if (prefetcher_.joinable()) {
prefetcher_.join();
}
delete channel_;
channel_ = nullptr;
}
private:
void PrefetchThreadFunc();
std::thread prefetcher_;
framework::Channel<Item>* buffer_;
framework::Channel<Item>* channel_;
platform::Place place_;
std::vector<std::unique_ptr<platform::DeviceContext>> ctxs_;
mutable Item local_buffer_;
};
class CreateDoubleBufferReaderOp : public framework::OperatorBase {
......@@ -123,70 +142,70 @@ class CreateDoubleBufferReaderOpMaker : public DecoratedReaderMakerBase {
}
};
bool DoubleBufferReader::HasNext() const {
while (!channel_->IsClosed() && !channel_->CanReceive()) {
}
return channel_->CanReceive();
}
void DoubleBufferReader::ReadNext(std::vector<framework::LoDTensor>* out) {
if (!HasNext()) {
PADDLE_THROW("There is no next data!");
}
if (local_buffer_.payloads_.empty()) {
buffer_->Receive(&local_buffer_);
}
*out = local_buffer_.payloads_;
local_buffer_.payloads_.clear();
if (local_buffer_.ctx_) {
local_buffer_.ctx_->Wait();
Item batch;
channel_->Receive(&batch);
*out = batch.payloads_;
if (batch.ctx_) {
batch.ctx_->Wait();
}
}
void DoubleBufferReader::ReInit() {
reader_->ReInit();
buffer_->Close();
prefetcher_.join();
delete buffer_;
start_thread();
EndPrefetcher();
StartPrefetcher();
}
void DoubleBufferReader::PrefetchThreadFunc() {
VLOG(5) << "A new prefetch thread starts.";
size_t gpu_ctx_offset = 0;
std::vector<std::vector<framework::LoDTensor>> cpu_tensor_cache(kCacheSize);
std::vector<std::vector<framework::LoDTensor>> gpu_tensor_cache(kCacheSize);
size_t cached_tensor_id = 0;
while (reader_->HasNext()) {
Item batch;
reader_->ReadNext(&batch.payloads_);
auto& cpu_batch = cpu_tensor_cache[cached_tensor_id];
reader_->ReadNext(&cpu_batch);
if (platform::is_gpu_place(place_)) {
std::vector<framework::LoDTensor> gpu_batch;
auto& gpu_ctx = this->ctxs_[gpu_ctx_offset++];
gpu_ctx_offset %= this->ctxs_.size();
gpu_batch.resize(batch.payloads_.size());
for (size_t i = 0; i < batch.payloads_.size(); ++i) {
framework::TensorCopy(batch.payloads_[i], place_, *gpu_ctx,
&gpu_batch[i]);
gpu_batch[i].set_lod(batch.payloads_[i].lod());
auto& gpu_batch = gpu_tensor_cache[cached_tensor_id];
auto* gpu_ctx = ctxs_[cached_tensor_id].get();
gpu_batch.resize(cpu_batch.size());
for (size_t i = 0; i < cpu_batch.size(); ++i) {
framework::TensorCopy(cpu_batch[i], place_, *gpu_ctx, &gpu_batch[i]);
gpu_batch[i].set_lod(cpu_batch[i].lod());
}
batch.ctx_ = gpu_ctx.get();
std::swap(gpu_batch, batch.payloads_);
batch.payloads_ = gpu_batch;
batch.ctx_ = gpu_ctx;
} else {
// CPUPlace
batch.payloads_ = cpu_batch;
}
++cached_tensor_id;
cached_tensor_id %= kCacheSize;
try {
buffer_->Send(&batch);
channel_->Send(&batch);
} catch (paddle::platform::EnforceNotMet e) {
VLOG(5) << "WARNING: The double buffer channel has been closed. The "
"prefetch thread will terminate.";
break;
}
}
buffer_->Close();
channel_->Close();
VLOG(5) << "Prefetch thread terminates.";
}
bool DoubleBufferReader::HasNext() const {
if (local_buffer_.payloads_.empty()) {
bool ok = buffer_->Receive(&local_buffer_);
return ok;
} else {
return true;
}
}
} // namespace reader
} // namespace operators
} // namespace paddle
......
paddle/scripts/docker/build.sh
浏览文件 @ b853ac8c
......@@ -104,7 +104,9 @@ EOF
# make install should also be test when unittest
make install -j `nproc`
pip install /usr/local/opt/paddle/share/wheels/*.whl
paddle version
if [[ ${WITH_FLUID_ONLY:-OFF} == "OFF" ]] ; then
paddle version
fi
fi
}
......@@ -183,6 +185,14 @@ EOF
NCCL_DEPS=""
fi
if [[ ${WITH_FLUID_ONLY:-OFF} == "OFF" ]]; then
PADDLE_VERSION="paddle version"
CMD='"paddle", "version"'
else
PADDLE_VERSION="true"
CMD='"true"'
fi
cat >> /paddle/build/Dockerfile <<EOF
ADD python/dist/*.whl /
# run paddle version to install python packages first
......@@ -192,7 +202,7 @@ EOF
pip install /*.whl; apt-get install -f -y && \
apt-get clean -y && \
rm -f /*.whl && \
paddle version && \
${PADDLE_VERSION} && \
ldconfig
${DOCKERFILE_CUDNN_DSO}
${DOCKERFILE_GPU_ENV}
......@@ -200,7 +210,7 @@ EOF
ADD go/cmd/pserver/pserver /usr/bin/
ADD go/cmd/master/master /usr/bin/
# default command shows the paddle version and exit
CMD ["paddle", "version"]
CMD [${CMD}]
EOF
}
......
python/CMakeLists.txt
浏览文件 @ b853ac8c
......@@ -81,6 +81,7 @@ if (WITH_TESTING)
# enable v2 API unittest only when paddle swig api is compiled
add_subdirectory(paddle/v2/tests)
add_subdirectory(paddle/v2/plot/tests)
add_subdirectory(paddle/v2/reader/tests)
endif()
endif()
add_subdirectory(paddle/fluid/tests)
......
python/paddle/dataset/__init__.py
浏览文件 @ b853ac8c
......@@ -37,7 +37,7 @@ __all__ = [
'cifar',
'movielens',
'conll05',
'sentiment'
'sentiment',
'uci_housing',
'wmt14',
'wmt16',
......
python/paddle/fluid/layers/nn.py
浏览文件 @ b853ac8c
......@@ -76,6 +76,7 @@ __all__ = [
'reshape',
'lod_reset',
'lrn',
'pad',
]
......@@ -3379,6 +3380,7 @@ def reshape(x, shape, actual_shape=None, act=None, inplace=True, name=None):
Examples:
.. code-block:: python
data = fluid.layers.data(
name='data', shape=[2, 4, 6], dtype='float32')
reshaped = fluid.layers.reshape(
......@@ -3580,3 +3582,62 @@ def lrn(input, n=5, k=1.0, alpha=1e-4, beta=0.75, name=None):
"beta": beta})
return lrn_out
def pad(x, paddings, pad_value=0., name=None):
"""
Pads a tensor with a constant value given by :attr:`pad_value`, and the
padded width is specified by :attr:`paddings`.
Specifically, the number of values padded before the contents of :attr:`x`
in dimension :attr:`i` is indicated by :attr:`paddings[i]`, and the number
of values padded after the contents of :attr:`x` in dimension :attr:`i` is
indicated by :attr:`paddings[i+1]`.
See below for an example.
.. code-block:: text
Given:
x = [[1, 2], [3, 4]]
paddings = [0, 1, 1, 2]
pad_value = 0
Return:
out = [[0, 1, 2, 0, 0]
[0, 3, 4, 0, 0]
[0, 0, 0, 0, 0]]
Args:
x (Variable): The input tensor variable.
paddings (list): A list of integers. Its elements specify the padded
width before and after for each dimension in turn.
The length of :attr:paddings must be
:math:`rank(x) \\times 2`.
pad_value (float): The constant value used to pad.
name(str|None): A name for this layer(optional). If set None, the layer
will be named automatically.
Returns:
Variable: The padded tensor variable.
Examples:
.. code-block:: python
# x is a rank 2 tensor variable.
out = fluid.layers.pad(
x=x, paddings=[0, 1, 1, 2], pad_value=0.)
"""
helper = LayerHelper('pad', input=x, **locals())
dtype = helper.input_dtype()
out = helper.create_tmp_variable(dtype)
helper.append_op(
type='pad',
inputs={'X': x},
outputs={'Out': out},
attrs={'paddings': paddings,
'pad_value': float(pad_value)})
return out
python/paddle/fluid/tests/unittests/test_lookup_table_op.py
浏览文件 @ b853ac8c
......@@ -96,5 +96,47 @@ class TestLookupTableIdsIsSelectedRows(OpTest):
self.check_with_place(place)
class TestLookupTableWIsSelectedRows(OpTest):
def check_with_place(self, place):
scope = core.Scope()
# create and initialize Id Variable
ids_tensor = scope.var('Ids').get_tensor()
ids_array = np.array([[0], [4], [3], [5]]).astype("int64")
ids_tensor.set(ids_array, place)
# create and initialize W Variable
rows = [0, 1, 2, 3, 4, 5, 6]
row_numel = 12
w_selected_rows = scope.var('W').get_selected_rows()
w_selected_rows.set_height(len(rows))
w_selected_rows.set_rows(rows)
w_array = np.ones((len(rows), row_numel)).astype("float32")
for i in range(len(rows)):
w_array[i] *= i
ids_tensor = w_selected_rows.get_tensor()
ids_tensor.set(w_array, place)
# create Out Variable
Out_tensor = scope.var('Out').get_tensor()
# create and run lookup_table operator
lookup_table = Operator("lookup_table", W='W', Ids='Ids', Out='Out')
lookup_table.run(scope, place)
# get result from Out
result_array = np.array(Out_tensor)
# all(): return True if all elements of the iterable are true (or if the iterable is empty)
for idx, row in enumerate(ids_array):
assert (row[0] == result_array[idx]).all()
def test_w_is_selected_rows(self):
places = [core.CPUPlace()]
# currently only support CPU
for place in places:
self.check_with_place(place)
if __name__ == "__main__":
unittest.main()
python/paddle/fluid/tests/unittests/test_recv_op.py
浏览文件 @ b853ac8c
......@@ -23,7 +23,7 @@ import time
class TestRecvOp(unittest.TestCase):
def test_send(self):
def no_test_send(self):
# Run init_serv in a thread
place = fluid.CPUPlace()
p = Process(target=self.init_serv, args=(place, ))
......
python/paddle/v2/__init__.py
浏览文件 @ b853ac8c
......@@ -22,13 +22,17 @@ import data_type
import topology
import networks
import evaluator
from . import dataset
from . import reader
from . import plot
import attr
import op
import pooling
import inference
import networks
import minibatch
import plot
import image
import paddle.trainer.config_parser as cp
__all__ = [
......@@ -44,11 +48,14 @@ __all__ = [
'data_type',
'attr',
'pooling',
'dataset',
'reader',
'topology',
'networks',
'infer',
'plot',
'evaluator',
'image',
'master',
]
......@@ -146,3 +153,4 @@ def init(**kwargs):
infer = inference.infer
batch = minibatch.batch
python/paddle/v2/dataset/__init__.py 0 → 100644
浏览文件 @ b853ac8c
# Copyright (c) 2016 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.
"""
Dataset package.
"""
import mnist
import imikolov
import imdb
import cifar
import movielens
import conll05
import uci_housing
import sentiment
import wmt14
import wmt16
import mq2007
import flowers
import voc2012
__all__ = [
'mnist',
'imikolov',
'imdb',
'cifar',
'movielens',
'conll05',
'sentiment',
'uci_housing',
'wmt14',
'wmt16',
'mq2007',
'flowers',
'voc2012',
]
python/paddle/v2/dataset/cifar.py 0 → 100644
浏览文件 @ b853ac8c
# Copyright (c) 2016 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.
"""
CIFAR dataset.
This module will download dataset from
https://www.cs.toronto.edu/~kriz/cifar.html and parse train/test set into
paddle reader creators.
The CIFAR-10 dataset consists of 60000 32x32 colour images in 10 classes,
with 6000 images per class. There are 50000 training images and 10000 test
images.
The CIFAR-100 dataset is just like the CIFAR-10, except it has 100 classes
containing 600 images each. There are 500 training images and 100 testing
images per class.
"""
import cPickle
import itertools
import numpy
import paddle.v2.dataset.common
import tarfile
__all__ = ['train100', 'test100', 'train10', 'test10', 'convert']
URL_PREFIX = 'https://www.cs.toronto.edu/~kriz/'
CIFAR10_URL = URL_PREFIX + 'cifar-10-python.tar.gz'
CIFAR10_MD5 = 'c58f30108f718f92721af3b95e74349a'
CIFAR100_URL = URL_PREFIX + 'cifar-100-python.tar.gz'
CIFAR100_MD5 = 'eb9058c3a382ffc7106e4002c42a8d85'
def reader_creator(filename, sub_name):
def read_batch(batch):
data = batch['data']
labels = batch.get('labels', batch.get('fine_labels', None))
assert labels is not None
for sample, label in itertools.izip(data, labels):
yield (sample / 255.0).astype(numpy.float32), int(label)
def reader():
with tarfile.open(filename, mode='r') as f:
names = (each_item.name for each_item in f
if sub_name in each_item.name)
for name in names:
batch = cPickle.load(f.extractfile(name))
for item in read_batch(batch):
yield item
return reader
def train100():
"""
CIFAR-100 training set creator.
It returns a reader creator, each sample in the reader is image pixels in
[0, 1] and label in [0, 99].
:return: Training reader creator
:rtype: callable
"""
return reader_creator(
paddle.v2.dataset.common.download(CIFAR100_URL, 'cifar', CIFAR100_MD5),
'train')
def test100():
"""
CIFAR-100 test set creator.
It returns a reader creator, each sample in the reader is image pixels in
[0, 1] and label in [0, 9].
:return: Test reader creator.
:rtype: callable
"""
return reader_creator(
paddle.v2.dataset.common.download(CIFAR100_URL, 'cifar', CIFAR100_MD5),
'test')
def train10():
"""
CIFAR-10 training set creator.
It returns a reader creator, each sample in the reader is image pixels in
[0, 1] and label in [0, 9].
:return: Training reader creator
:rtype: callable
"""
return reader_creator(
paddle.v2.dataset.common.download(CIFAR10_URL, 'cifar', CIFAR10_MD5),
'data_batch')
def test10():
"""
CIFAR-10 test set creator.
It returns a reader creator, each sample in the reader is image pixels in
[0, 1] and label in [0, 9].
:return: Test reader creator.
:rtype: callable
"""
return reader_creator(
paddle.v2.dataset.common.download(CIFAR10_URL, 'cifar', CIFAR10_MD5),
'test_batch')
def fetch():
paddle.v2.dataset.common.download(CIFAR10_URL, 'cifar', CIFAR10_MD5)
paddle.v2.dataset.common.download(CIFAR100_URL, 'cifar', CIFAR100_MD5)
def convert(path):
"""
Converts dataset to recordio format
"""
paddle.v2.dataset.common.convert(path, train100(), 1000, "cifar_train100")
paddle.v2.dataset.common.convert(path, test100(), 1000, "cifar_test100")
paddle.v2.dataset.common.convert(path, train10(), 1000, "cifar_train10")
paddle.v2.dataset.common.convert(path, test10(), 1000, "cifar_test10")
python/paddle/v2/dataset/common.py 0 → 100644
浏览文件 @ b853ac8c
# Copyright (c) 2016 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 requests
import hashlib
import os
import errno
import shutil
import sys
import importlib
import paddle.v2.dataset
import cPickle
import glob
import cPickle as pickle
__all__ = [
'DATA_HOME',
'download',
'md5file',
'split',
'cluster_files_reader',
'convert',
]
DATA_HOME = os.path.expanduser('~/.cache/paddle/dataset')
# When running unit tests, there could be multiple processes that
# trying to create DATA_HOME directory simultaneously, so we cannot
# use a if condition to check for the existence of the directory;
# instead, we use the filesystem as the synchronization mechanism by
# catching returned errors.
def must_mkdirs(path):
try:
os.makedirs(DATA_HOME)
except OSError as exc:
if exc.errno != errno.EEXIST:
raise
pass
must_mkdirs(DATA_HOME)
def md5file(fname):
hash_md5 = hashlib.md5()
f = open(fname, "rb")
for chunk in iter(lambda: f.read(4096), b""):
hash_md5.update(chunk)
f.close()
return hash_md5.hexdigest()
def download(url, module_name, md5sum, save_name=None):
dirname = os.path.join(DATA_HOME, module_name)
if not os.path.exists(dirname):
os.makedirs(dirname)
filename = os.path.join(dirname,
url.split('/')[-1]
if save_name is None else save_name)
retry = 0
retry_limit = 3
while not (os.path.exists(filename) and md5file(filename) == md5sum):
if os.path.exists(filename):
print "file md5", md5file(filename), md5sum
if retry < retry_limit:
retry += 1
else:
raise RuntimeError("Cannot download {0} within retry limit {1}".
format(url, retry_limit))
print "Cache file %s not found, downloading %s" % (filename, url)
r = requests.get(url, stream=True)
total_length = r.headers.get('content-length')
if total_length is None:
with open(filename, 'w') as f:
shutil.copyfileobj(r.raw, f)
else:
with open(filename, 'w') as f:
dl = 0
total_length = int(total_length)
for data in r.iter_content(chunk_size=4096):
dl += len(data)
f.write(data)
done = int(50 * dl / total_length)
sys.stdout.write("\r[%s%s]" % ('=' * done,
' ' * (50 - done)))
sys.stdout.flush()
return filename
def fetch_all():
for module_name in filter(lambda x: not x.startswith("__"),
dir(paddle.v2.dataset)):
if "fetch" in dir(
importlib.import_module("paddle.v2.dataset.%s" % module_name)):
getattr(
importlib.import_module("paddle.v2.dataset.%s" % module_name),
"fetch")()
def fetch_all_recordio(path):
for module_name in filter(lambda x: not x.startswith("__"),
dir(paddle.v2.dataset)):
if "convert" in dir(
importlib.import_module("paddle.v2.dataset.%s" % module_name)) and \
not module_name == "common":
ds_path = os.path.join(path, module_name)
must_mkdirs(ds_path)
getattr(
importlib.import_module("paddle.v2.dataset.%s" % module_name),
"convert")(ds_path)
def split(reader, line_count, suffix="%05d.pickle", dumper=cPickle.dump):
"""
you can call the function as:
split(paddle.v2.dataset.cifar.train10(), line_count=1000,
suffix="imikolov-train-%05d.pickle")
the output files as:
|-imikolov-train-00000.pickle
|-imikolov-train-00001.pickle
|- ...
|-imikolov-train-00480.pickle
:param reader: is a reader creator
:param line_count: line count for each file
:param suffix: the suffix for the output files, should contain "%d"
means the id for each file. Default is "%05d.pickle"
:param dumper: is a callable function that dump object to file, this
function will be called as dumper(obj, f) and obj is the object
will be dumped, f is a file object. Default is cPickle.dump.
"""
if not callable(dumper):
raise TypeError("dumper should be callable.")
lines = []
indx_f = 0
for i, d in enumerate(reader()):
lines.append(d)
if i >= line_count and i % line_count == 0:
with open(suffix % indx_f, "w") as f:
dumper(lines, f)
lines = []
indx_f += 1
if lines:
with open(suffix % indx_f, "w") as f:
dumper(lines, f)
def cluster_files_reader(files_pattern,
trainer_count,
trainer_id,
loader=cPickle.load):
"""
Create a reader that yield element from the given files, select
a file set according trainer count and trainer_id
:param files_pattern: the files which generating by split(...)
:param trainer_count: total trainer count
:param trainer_id: the trainer rank id
:param loader: is a callable function that load object from file, this
function will be called as loader(f) and f is a file object.
Default is cPickle.load
"""
def reader():
if not callable(loader):
raise TypeError("loader should be callable.")
file_list = glob.glob(files_pattern)
file_list.sort()
my_file_list = []
for idx, fn in enumerate(file_list):
if idx % trainer_count == trainer_id:
print "append file: %s" % fn
my_file_list.append(fn)
for fn in my_file_list:
with open(fn, "r") as f:
lines = loader(f)
for line in lines:
yield line
return reader
def convert(output_path, reader, line_count, name_prefix):
import recordio
"""
Convert data from reader to recordio format files.
:param output_path: directory in which output files will be saved.
:param reader: a data reader, from which the convert program will read
data instances.
:param name_prefix: the name prefix of generated files.
:param max_lines_to_shuffle: the max lines numbers to shuffle before
writing.
"""
assert line_count >= 1
indx_f = 0
def write_data(indx_f, lines):
filename = "%s/%s-%05d" % (output_path, name_prefix, indx_f)
writer = recordio.writer(filename)
for l in lines:
# FIXME(Yancey1989):
# dumps with protocol: pickle.HIGHEST_PROTOCOL
writer.write(cPickle.dumps(l))
writer.close()
lines = []
for i, d in enumerate(reader()):
lines.append(d)
if i % line_count == 0 and i >= line_count:
write_data(indx_f, lines)
lines = []
indx_f += 1
continue
write_data(indx_f, lines)
python/paddle/v2/dataset/conll05.py 0 → 100644
浏览文件 @ b853ac8c
# Copyright (c) 2016 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.
"""
Conll05 dataset.
Paddle semantic role labeling Book and demo use this dataset as an example.
Because Conll05 is not free in public, the default downloaded URL is test set
of Conll05 (which is public). Users can change URL and MD5 to their Conll
dataset. And a pre-trained word vector model based on Wikipedia corpus is used
to initialize SRL model.
"""
import tarfile
import gzip
import itertools
import paddle.v2.dataset.common
__all__ = ['test, get_dict', 'get_embedding', 'convert']
DATA_URL = 'http://www.cs.upc.edu/~srlconll/conll05st-tests.tar.gz'
DATA_MD5 = '387719152ae52d60422c016e92a742fc'
WORDDICT_URL = 'http://paddlepaddle.bj.bcebos.com/demo/srl_dict_and_embedding/wordDict.txt'
WORDDICT_MD5 = 'ea7fb7d4c75cc6254716f0177a506baa'
VERBDICT_URL = 'http://paddlepaddle.bj.bcebos.com/demo/srl_dict_and_embedding/verbDict.txt'
VERBDICT_MD5 = '0d2977293bbb6cbefab5b0f97db1e77c'
TRGDICT_URL = 'http://paddlepaddle.bj.bcebos.com/demo/srl_dict_and_embedding/targetDict.txt'
TRGDICT_MD5 = 'd8c7f03ceb5fc2e5a0fa7503a4353751'
EMB_URL = 'http://paddlepaddle.bj.bcebos.com/demo/srl_dict_and_embedding/emb'
EMB_MD5 = 'bf436eb0faa1f6f9103017f8be57cdb7'
UNK_IDX = 0
def load_label_dict(filename):
d = dict()
tag_dict = set()
with open(filename, 'r') as f:
for i, line in enumerate(f):
line = line.strip()
if line.startswith("B-"):
tag_dict.add(line[2:])
elif line.startswith("I-"):
tag_dict.add(line[2:])
index = 0
for tag in tag_dict:
d["B-" + tag] = index
index += 1
d["I-" + tag] = index
index += 1
d["O"] = index
return d
def load_dict(filename):
d = dict()
with open(filename, 'r') as f:
for i, line in enumerate(f):
d[line.strip()] = i
return d
def corpus_reader(data_path, words_name, props_name):
"""
Read one corpus. It returns an iterator. Each element of
this iterator is a tuple including sentence and labels. The sentence is
consist of a list of word IDs. The labels include a list of label IDs.
:return: a iterator of data.
:rtype: iterator
"""
def reader():
tf = tarfile.open(data_path)
wf = tf.extractfile(words_name)
pf = tf.extractfile(props_name)
with gzip.GzipFile(fileobj=wf) as words_file, gzip.GzipFile(
fileobj=pf) as props_file:
sentences = []
labels = []
one_seg = []
for word, label in itertools.izip(words_file, props_file):
word = word.strip()
label = label.strip().split()
if len(label) == 0: # end of sentence
for i in xrange(len(one_seg[0])):
a_kind_lable = [x[i] for x in one_seg]
labels.append(a_kind_lable)
if len(labels) >= 1:
verb_list = []
for x in labels[0]:
if x != '-':
verb_list.append(x)
for i, lbl in enumerate(labels[1:]):
cur_tag = 'O'
is_in_bracket = False
lbl_seq = []
verb_word = ''
for l in lbl:
if l == '*' and is_in_bracket == False:
lbl_seq.append('O')
elif l == '*' and is_in_bracket == True:
lbl_seq.append('I-' + cur_tag)
elif l == '*)':
lbl_seq.append('I-' + cur_tag)
is_in_bracket = False
elif l.find('(') != -1 and l.find(')') != -1:
cur_tag = l[1:l.find('*')]
lbl_seq.append('B-' + cur_tag)
is_in_bracket = False
elif l.find('(') != -1 and l.find(')') == -1:
cur_tag = l[1:l.find('*')]
lbl_seq.append('B-' + cur_tag)
is_in_bracket = True
else:
raise RuntimeError('Unexpected label: %s' %
l)
yield sentences, verb_list[i], lbl_seq
sentences = []
labels = []
one_seg = []
else:
sentences.append(word)
one_seg.append(label)
pf.close()
wf.close()
tf.close()
return reader
def reader_creator(corpus_reader,
word_dict=None,
predicate_dict=None,
label_dict=None):
def reader():
for sentence, predicate, labels in corpus_reader():
sen_len = len(sentence)
verb_index = labels.index('B-V')
mark = [0] * len(labels)
if verb_index > 0:
mark[verb_index - 1] = 1
ctx_n1 = sentence[verb_index - 1]
else:
ctx_n1 = 'bos'
if verb_index > 1:
mark[verb_index - 2] = 1
ctx_n2 = sentence[verb_index - 2]
else:
ctx_n2 = 'bos'
mark[verb_index] = 1
ctx_0 = sentence[verb_index]
if verb_index < len(labels) - 1:
mark[verb_index + 1] = 1
ctx_p1 = sentence[verb_index + 1]
else:
ctx_p1 = 'eos'
if verb_index < len(labels) - 2:
mark[verb_index + 2] = 1
ctx_p2 = sentence[verb_index + 2]
else:
ctx_p2 = 'eos'
word_idx = [word_dict.get(w, UNK_IDX) for w in sentence]
ctx_n2_idx = [word_dict.get(ctx_n2, UNK_IDX)] * sen_len
ctx_n1_idx = [word_dict.get(ctx_n1, UNK_IDX)] * sen_len
ctx_0_idx = [word_dict.get(ctx_0, UNK_IDX)] * sen_len
ctx_p1_idx = [word_dict.get(ctx_p1, UNK_IDX)] * sen_len
ctx_p2_idx = [word_dict.get(ctx_p2, UNK_IDX)] * sen_len
pred_idx = [predicate_dict.get(predicate)] * sen_len
label_idx = [label_dict.get(w) for w in labels]
yield word_idx, ctx_n2_idx, ctx_n1_idx, \
ctx_0_idx, ctx_p1_idx, ctx_p2_idx, pred_idx, mark, label_idx
return reader
def get_dict():
"""
Get the word, verb and label dictionary of Wikipedia corpus.
"""
word_dict = load_dict(
paddle.v2.dataset.common.download(WORDDICT_URL, 'conll05st',
WORDDICT_MD5))
verb_dict = load_dict(
paddle.v2.dataset.common.download(VERBDICT_URL, 'conll05st',
VERBDICT_MD5))
label_dict = load_label_dict(
paddle.v2.dataset.common.download(TRGDICT_URL, 'conll05st',
TRGDICT_MD5))
return word_dict, verb_dict, label_dict
def get_embedding():
"""
Get the trained word vector based on Wikipedia corpus.
"""
return paddle.v2.dataset.common.download(EMB_URL, 'conll05st', EMB_MD5)
def test():
"""
Conll05 test set creator.
Because the training dataset is not free, the test dataset is used for
training. It returns a reader creator, each sample in the reader is nine
features, including sentence sequence, predicate, predicate context,
predicate context flag and tagged sequence.
:return: Training reader creator
:rtype: callable
"""
word_dict, verb_dict, label_dict = get_dict()
reader = corpus_reader(
paddle.v2.dataset.common.download(DATA_URL, 'conll05st', DATA_MD5),
words_name='conll05st-release/test.wsj/words/test.wsj.words.gz',
props_name='conll05st-release/test.wsj/props/test.wsj.props.gz')
return reader_creator(reader, word_dict, verb_dict, label_dict)
def fetch():
paddle.v2.dataset.common.download(WORDDICT_URL, 'conll05st', WORDDICT_MD5)
paddle.v2.dataset.common.download(VERBDICT_URL, 'conll05st', VERBDICT_MD5)
paddle.v2.dataset.common.download(TRGDICT_URL, 'conll05st', TRGDICT_MD5)
paddle.v2.dataset.common.download(EMB_URL, 'conll05st', EMB_MD5)
paddle.v2.dataset.common.download(DATA_URL, 'conll05st', DATA_MD5)
def convert(path):
"""
Converts dataset to recordio format
"""
paddle.v2.dataset.common.convert(path, test(), 1000, "conl105_train")
paddle.v2.dataset.common.convert(path, test(), 1000, "conl105_test")
python/paddle/v2/dataset/flowers.py 0 → 100644
浏览文件 @ b853ac8c
# Copyright (c) 2016 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.
"""
This module will download dataset from
http://www.robots.ox.ac.uk/~vgg/data/flowers/102/index.html
and parse train/test set intopaddle reader creators.
This set contains images of flowers belonging to 102 different categories.
The images were acquired by searching the web and taking pictures. There are a
minimum of 40 images for each category.
The database was used in:
Nilsback, M-E. and Zisserman, A. Automated flower classification over a large
number of classes.Proceedings of the Indian Conference on Computer Vision,
Graphics and Image Processing (2008)
http://www.robots.ox.ac.uk/~vgg/publications/papers/nilsback08.{pdf,ps.gz}.
"""
import cPickle
import itertools
import functools
from common import download
import tarfile
import scipy.io as scio
from paddle.v2.image import *
from paddle.v2.reader import *
import os
import numpy as np
from multiprocessing import cpu_count
__all__ = ['train', 'test', 'valid']
DATA_URL = 'http://www.robots.ox.ac.uk/~vgg/data/flowers/102/102flowers.tgz'
LABEL_URL = 'http://www.robots.ox.ac.uk/~vgg/data/flowers/102/imagelabels.mat'
SETID_URL = 'http://www.robots.ox.ac.uk/~vgg/data/flowers/102/setid.mat'
DATA_MD5 = '33bfc11892f1e405ca193ae9a9f2a118'
LABEL_MD5 = 'e0620be6f572b9609742df49c70aed4d'
SETID_MD5 = 'a5357ecc9cb78c4bef273ce3793fc85c'
# In official 'readme', tstid is the flag of test data
# and trnid is the flag of train data. But test data is more than train data.
# So we exchange the train data and test data.
TRAIN_FLAG = 'tstid'
TEST_FLAG = 'trnid'
VALID_FLAG = 'valid'
def default_mapper(is_train, sample):
'''
map image bytes data to type needed by model input layer
'''
img, label = sample
img = load_image_bytes(img)
img = simple_transform(
img, 256, 224, is_train, mean=[103.94, 116.78, 123.68])
return img.flatten().astype('float32'), label
train_mapper = functools.partial(default_mapper, True)
test_mapper = functools.partial(default_mapper, False)
def reader_creator(data_file,
label_file,
setid_file,
dataset_name,
mapper,
buffered_size=1024,
use_xmap=True):
'''
1. read images from tar file and
merge images into batch files in 102flowers.tgz_batch/
2. get a reader to read sample from batch file
:param data_file: downloaded data file
:type data_file: string
:param label_file: downloaded label file
:type label_file: string
:param setid_file: downloaded setid file containing information
about how to split dataset
:type setid_file: string
:param dataset_name: data set name (tstid|trnid|valid)
:type dataset_name: string
:param mapper: a function to map image bytes data to type
needed by model input layer
:type mapper: callable
:param buffered_size: the size of buffer used to process images
:type buffered_size: int
:return: data reader
:rtype: callable
'''
labels = scio.loadmat(label_file)['labels'][0]
indexes = scio.loadmat(setid_file)[dataset_name][0]
img2label = {}
for i in indexes:
img = "jpg/image_%05d.jpg" % i
img2label[img] = labels[i - 1]
file_list = batch_images_from_tar(data_file, dataset_name, img2label)
def reader():
for file in open(file_list):
file = file.strip()
batch = None
with open(file, 'r') as f:
batch = cPickle.load(f)
data = batch['data']
labels = batch['label']
for sample, label in itertools.izip(data, batch['label']):
yield sample, int(label) - 1
if use_xmap:
return xmap_readers(mapper, reader, cpu_count(), buffered_size)
else:
return map_readers(mapper, reader)
def train(mapper=train_mapper, buffered_size=1024, use_xmap=True):
'''
Create flowers training set reader.
It returns a reader, each sample in the reader is
image pixels in [0, 1] and label in [1, 102]
translated from original color image by steps:
1. resize to 256*256
2. random crop to 224*224
3. flatten
:param mapper: a function to map sample.
:type mapper: callable
:param buffered_size: the size of buffer used to process images
:type buffered_size: int
:return: train data reader
:rtype: callable
'''
return reader_creator(
download(DATA_URL, 'flowers', DATA_MD5),
download(LABEL_URL, 'flowers', LABEL_MD5),
download(SETID_URL, 'flowers', SETID_MD5), TRAIN_FLAG, mapper,
buffered_size, use_xmap)
def test(mapper=test_mapper, buffered_size=1024, use_xmap=True):
'''
Create flowers test set reader.
It returns a reader, each sample in the reader is
image pixels in [0, 1] and label in [1, 102]
translated from original color image by steps:
1. resize to 256*256
2. random crop to 224*224
3. flatten
:param mapper: a function to map sample.
:type mapper: callable
:param buffered_size: the size of buffer used to process images
:type buffered_size: int
:return: test data reader
:rtype: callable
'''
return reader_creator(
download(DATA_URL, 'flowers', DATA_MD5),
download(LABEL_URL, 'flowers', LABEL_MD5),
download(SETID_URL, 'flowers', SETID_MD5), TEST_FLAG, mapper,
buffered_size, use_xmap)
def valid(mapper=test_mapper, buffered_size=1024, use_xmap=True):
'''
Create flowers validation set reader.
It returns a reader, each sample in the reader is
image pixels in [0, 1] and label in [1, 102]
translated from original color image by steps:
1. resize to 256*256
2. random crop to 224*224
3. flatten
:param mapper: a function to map sample.
:type mapper: callable
:param buffered_size: the size of buffer used to process images
:type buffered_size: int
:return: test data reader
:rtype: callable
'''
return reader_creator(
download(DATA_URL, 'flowers', DATA_MD5),
download(LABEL_URL, 'flowers', LABEL_MD5),
download(SETID_URL, 'flowers', SETID_MD5), VALID_FLAG, mapper,
buffered_size, use_xmap)
def fetch():
download(DATA_URL, 'flowers', DATA_MD5)
download(LABEL_URL, 'flowers', LABEL_MD5)
download(SETID_URL, 'flowers', SETID_MD5)
python/paddle/v2/dataset/imdb.py 0 → 100644
浏览文件 @ b853ac8c
# Copyright (c) 2016 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.
"""
IMDB dataset.
This module downloads IMDB dataset from
http://ai.stanford.edu/%7Eamaas/data/sentiment/. This dataset contains a set
of 25,000 highly polar movie reviews for training, and 25,000 for testing.
Besides, this module also provides API for building dictionary.
"""
import paddle.v2.dataset.common
import collections
import tarfile
import re
import string
__all__ = ['build_dict', 'train', 'test', 'convert']
URL = 'http://ai.stanford.edu/%7Eamaas/data/sentiment/aclImdb_v1.tar.gz'
MD5 = '7c2ac02c03563afcf9b574c7e56c153a'
def tokenize(pattern):
"""
Read files that match the given pattern. Tokenize and yield each file.
"""
with tarfile.open(paddle.v2.dataset.common.download(URL, 'imdb',
MD5)) as tarf:
# Note that we should use tarfile.next(), which does
# sequential access of member files, other than
# tarfile.extractfile, which does random access and might
# destroy hard disks.
tf = tarf.next()
while tf != None:
if bool(pattern.match(tf.name)):
# newline and punctuations removal and ad-hoc tokenization.
yield tarf.extractfile(tf).read().rstrip("\n\r").translate(
None, string.punctuation).lower().split()
tf = tarf.next()
def build_dict(pattern, cutoff):
"""
Build a word dictionary from the corpus. Keys of the dictionary are words,
and values are zero-based IDs of these words.
"""
word_freq = collections.defaultdict(int)
for doc in tokenize(pattern):
for word in doc:
word_freq[word] += 1
# Not sure if we should prune less-frequent words here.
word_freq = filter(lambda x: x[1] > cutoff, word_freq.items())
dictionary = sorted(word_freq, key=lambda x: (-x[1], x[0]))
words, _ = list(zip(*dictionary))
word_idx = dict(zip(words, xrange(len(words))))
word_idx['<unk>'] = len(words)
return word_idx
def reader_creator(pos_pattern, neg_pattern, word_idx):
UNK = word_idx['<unk>']
INS = []
def load(pattern, out, label):
for doc in tokenize(pattern):
out.append(([word_idx.get(w, UNK) for w in doc], label))
load(pos_pattern, INS, 0)
load(neg_pattern, INS, 1)
def reader():
for doc, label in INS:
yield doc, label
return reader
def train(word_idx):
"""
IMDB training set creator.
It returns a reader creator, each sample in the reader is an zero-based ID
sequence and label in [0, 1].
:param word_idx: word dictionary
:type word_idx: dict
:return: Training reader creator
:rtype: callable
"""
return reader_creator(
re.compile("aclImdb/train/pos/.*\.txt$"),
re.compile("aclImdb/train/neg/.*\.txt$"), word_idx)
def test(word_idx):
"""
IMDB test set creator.
It returns a reader creator, each sample in the reader is an zero-based ID
sequence and label in [0, 1].
:param word_idx: word dictionary
:type word_idx: dict
:return: Test reader creator
:rtype: callable
"""
return reader_creator(
re.compile("aclImdb/test/pos/.*\.txt$"),
re.compile("aclImdb/test/neg/.*\.txt$"), word_idx)
def word_dict():
"""
Build a word dictionary from the corpus.
:return: Word dictionary
:rtype: dict
"""
return build_dict(
re.compile("aclImdb/((train)|(test))/((pos)|(neg))/.*\.txt$"), 150)
def fetch():
paddle.v2.dataset.common.download(URL, 'imdb', MD5)
def convert(path):
"""
Converts dataset to recordio format
"""
w = word_dict()
paddle.v2.dataset.common.convert(path, lambda: train(w), 1000, "imdb_train")
paddle.v2.dataset.common.convert(path, lambda: test(w), 1000, "imdb_test")
python/paddle/v2/dataset/imikolov.py 0 → 100644
浏览文件 @ b853ac8c
# Copyright (c) 2016 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.
"""
imikolov's simple dataset.
This module will download dataset from
http://www.fit.vutbr.cz/~imikolov/rnnlm/ and parse training set and test set
into paddle reader creators.
"""
import paddle.v2.dataset.common
import collections
import tarfile
__all__ = ['train', 'test', 'build_dict', 'convert']
URL = 'http://www.fit.vutbr.cz/~imikolov/rnnlm/simple-examples.tgz'
MD5 = '30177ea32e27c525793142b6bf2c8e2d'
class DataType(object):
NGRAM = 1
SEQ = 2
def word_count(f, word_freq=None):
if word_freq is None:
word_freq = collections.defaultdict(int)
for l in f:
for w in l.strip().split():
word_freq[w] += 1
word_freq['<s>'] += 1
word_freq['<e>'] += 1
return word_freq
def build_dict(min_word_freq=50):
"""
Build a word dictionary from the corpus, Keys of the dictionary are words,
and values are zero-based IDs of these words.
"""
train_filename = './simple-examples/data/ptb.train.txt'
test_filename = './simple-examples/data/ptb.valid.txt'
with tarfile.open(
paddle.v2.dataset.common.download(
paddle.v2.dataset.imikolov.URL, 'imikolov',
paddle.v2.dataset.imikolov.MD5)) as tf:
trainf = tf.extractfile(train_filename)
testf = tf.extractfile(test_filename)
word_freq = word_count(testf, word_count(trainf))
if '<unk>' in word_freq:
# remove <unk> for now, since we will set it as last index
del word_freq['<unk>']
word_freq = filter(lambda x: x[1] > min_word_freq, word_freq.items())
word_freq_sorted = sorted(word_freq, key=lambda x: (-x[1], x[0]))
words, _ = list(zip(*word_freq_sorted))
word_idx = dict(zip(words, xrange(len(words))))
word_idx['<unk>'] = len(words)
return word_idx
def reader_creator(filename, word_idx, n, data_type):
def reader():
with tarfile.open(
paddle.v2.dataset.common.download(
paddle.v2.dataset.imikolov.URL, 'imikolov',
paddle.v2.dataset.imikolov.MD5)) as tf:
f = tf.extractfile(filename)
UNK = word_idx['<unk>']
for l in f:
if DataType.NGRAM == data_type:
assert n > -1, 'Invalid gram length'
l = ['<s>'] + l.strip().split() + ['<e>']
if len(l) >= n:
l = [word_idx.get(w, UNK) for w in l]
for i in range(n, len(l) + 1):
yield tuple(l[i - n:i])
elif DataType.SEQ == data_type:
l = l.strip().split()
l = [word_idx.get(w, UNK) for w in l]
src_seq = [word_idx['<s>']] + l
trg_seq = l + [word_idx['<e>']]
if n > 0 and len(src_seq) > n: continue
yield src_seq, trg_seq
else:
assert False, 'Unknow data type'
return reader
def train(word_idx, n, data_type=DataType.NGRAM):
"""
imikolov training set creator.
It returns a reader creator, each sample in the reader is a word ID
tuple.
:param word_idx: word dictionary
:type word_idx: dict
:param n: sliding window size if type is ngram, otherwise max length of sequence
:type n: int
:param data_type: data type (ngram or sequence)
:type data_type: member variable of DataType (NGRAM or SEQ)
:return: Training reader creator
:rtype: callable
"""
return reader_creator('./simple-examples/data/ptb.train.txt', word_idx, n,
data_type)
def test(word_idx, n, data_type=DataType.NGRAM):
"""
imikolov test set creator.
It returns a reader creator, each sample in the reader is a word ID
tuple.
:param word_idx: word dictionary
:type word_idx: dict
:param n: sliding window size if type is ngram, otherwise max length of sequence
:type n: int
:param data_type: data type (ngram or sequence)
:type data_type: member variable of DataType (NGRAM or SEQ)
:return: Test reader creator
:rtype: callable
"""
return reader_creator('./simple-examples/data/ptb.valid.txt', word_idx, n,
data_type)
def fetch():
paddle.v2.dataset.common.download(URL, "imikolov", MD5)
def convert(path):
"""
Converts dataset to recordio format
"""
N = 5
word_dict = build_dict()
paddle.v2.dataset.common.convert(path,
train(word_dict, N), 1000,
"imikolov_train")
paddle.v2.dataset.common.convert(path,
test(word_dict, N), 1000, "imikolov_test")
python/paddle/v2/dataset/mnist.py 0 → 100644
浏览文件 @ b853ac8c
此差异已折叠。
python/paddle/v2/dataset/movielens.py 0 → 100644
浏览文件 @ b853ac8c
此差异已折叠。
python/paddle/v2/dataset/mq2007.py 0 → 100644
浏览文件 @ b853ac8c
此差异已折叠。
python/paddle/v2/dataset/sentiment.py 0 → 100644
浏览文件 @ b853ac8c
此差异已折叠。
python/paddle/v2/dataset/tests/cifar_test.py 0 → 100644
浏览文件 @ b853ac8c
# Copyright (c) 2016 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 paddle.v2.dataset.cifar
import unittest
class TestCIFAR(unittest.TestCase):
def check_reader(self, reader):
sum = 0
label = 0
for l in reader():
self.assertEqual(l[0].size, 3072)
if l[1] > label:
label = l[1]
sum += 1
return sum, label
def test_test10(self):
instances, max_label_value = self.check_reader(
paddle.v2.dataset.cifar.test10())
self.assertEqual(instances, 10000)
self.assertEqual(max_label_value, 9)
def test_train10(self):
instances, max_label_value = self.check_reader(
paddle.v2.dataset.cifar.train10())
self.assertEqual(instances, 50000)
self.assertEqual(max_label_value, 9)
def test_test100(self):
instances, max_label_value = self.check_reader(
paddle.v2.dataset.cifar.test100())
self.assertEqual(instances, 10000)
self.assertEqual(max_label_value, 99)
def test_train100(self):
instances, max_label_value = self.check_reader(
paddle.v2.dataset.cifar.train100())
self.assertEqual(instances, 50000)
self.assertEqual(max_label_value, 99)
if __name__ == '__main__':
unittest.main()
python/paddle/v2/dataset/tests/common_test.py 0 → 100644
浏览文件 @ b853ac8c
此差异已折叠。
python/paddle/v2/dataset/tests/flowers_test.py 0 → 100644
浏览文件 @ b853ac8c
此差异已折叠。
python/paddle/v2/dataset/tests/imdb_test.py 0 → 100644
浏览文件 @ b853ac8c
此差异已折叠。
python/paddle/v2/dataset/tests/imikolov_test.py 0 → 100644
浏览文件 @ b853ac8c
此差异已折叠。
python/paddle/v2/dataset/tests/mnist_test.py 0 → 100644
浏览文件 @ b853ac8c
此差异已折叠。
python/paddle/v2/dataset/tests/mq2007_test.py 0 → 100644
浏览文件 @ b853ac8c
此差异已折叠。
python/paddle/v2/dataset/tests/test_sentiment.py 0 → 100644
浏览文件 @ b853ac8c
此差异已折叠。
python/paddle/v2/dataset/tests/voc2012_test.py 0 → 100644
浏览文件 @ b853ac8c
此差异已折叠。
python/paddle/v2/dataset/tests/wmt16_test.py 0 → 100644
浏览文件 @ b853ac8c
此差异已折叠。
python/paddle/v2/dataset/uci_housing.py 0 → 100644
浏览文件 @ b853ac8c
此差异已折叠。
python/paddle/v2/dataset/voc2012.py 0 → 100644
浏览文件 @ b853ac8c
此差异已折叠。
python/paddle/v2/dataset/wmt14.py 0 → 100644
浏览文件 @ b853ac8c
此差异已折叠。
python/paddle/v2/dataset/wmt16.py 0 → 100644
浏览文件 @ b853ac8c
此差异已折叠。
python/paddle/v2/image.py 0 → 100644
浏览文件 @ b853ac8c
此差异已折叠。
python/paddle/v2/minibatch.py 0 → 100644
浏览文件 @ b853ac8c
此差异已折叠。
python/paddle/v2/reader/__init__.py 0 → 100644
浏览文件 @ b853ac8c
此差异已折叠。
python/paddle/v2/reader/creator.py 0 → 100644
浏览文件 @ b853ac8c
此差异已折叠。
python/paddle/v2/reader/decorator.py 0 → 100644
浏览文件 @ b853ac8c
此差异已折叠。
python/paddle/v2/reader/tests/CMakeLists.txt 0 → 100644
浏览文件 @ b853ac8c
py_test(creator_test SRCS creator_test.py)
py_test(decorator_test SRCS decorator_test.py)
python/paddle/v2/reader/tests/__init__.py 0 → 100644
浏览文件 @ b853ac8c
# 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.
python/paddle/v2/reader/tests/creator_test.py 0 → 100644
浏览文件 @ b853ac8c
此差异已折叠。
python/paddle/v2/reader/tests/decorator_test.py 0 → 100644
浏览文件 @ b853ac8c
此差异已折叠。
python/paddle/v2/tests/cat.jpg 0 → 100644
浏览文件 @ b853ac8c
此差异已折叠。
python/paddle/v2/tests/test_image.py 0 → 100644
浏览文件 @ b853ac8c
此差异已折叠。
python/setup.py.in
浏览文件 @ b853ac8c
此差异已折叠。
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或