Skip to content

P
Paddle

机器未来 / Paddle
与 Fork 源项目一致

Fork自 PaddlePaddle / Paddle

通知 1
Star 1
Fork 0
P
Paddle
提交 29bf727e 编写于 作者: F fengjiayi
浏览文件
操作

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

上级 c6c9c657 d3b0129c
隐藏空白更改
内联 并排
Showing with 379 addition and 261 deletion
doc/v2/api/config/evaluators.rst
浏览文件 @ 29bf727e
......@@ -101,7 +101,7 @@ value_printer
:noindex:
Detection
=====
==========
detection_map
-------------
......
doc/v2/api/config/layer.rst
浏览文件 @ 29bf727e
......@@ -11,7 +11,7 @@ Data layer
data
----
.. autoclass:: paddle.v2.layer.data
.. autofunction:: paddle.v2.layer.data
:noindex:
Fully Connected Layers
......@@ -21,12 +21,12 @@ Fully Connected Layers
fc
--
.. autoclass:: paddle.v2.layer.fc
.. autofunction:: paddle.v2.layer.fc
:noindex:
selective_fc
------------
.. autoclass:: paddle.v2.layer.selective_fc
.. autofunction:: paddle.v2.layer.selective_fc
:noindex:
Conv Layers
......@@ -34,34 +34,34 @@ Conv Layers
conv_operator
-------------
.. autoclass:: paddle.v2.layer.conv_operator
.. autofunction:: paddle.v2.layer.conv_operator
:noindex:
conv_projection
---------------
.. autoclass:: paddle.v2.layer.conv_projection
.. autofunction:: paddle.v2.layer.conv_projection
:noindex:
conv_shift
----------
.. autoclass:: paddle.v2.layer.conv_shift
.. autofunction:: paddle.v2.layer.conv_shift
:noindex:
img_conv
--------
.. autoclass:: paddle.v2.layer.img_conv
.. autofunction:: paddle.v2.layer.img_conv
:noindex:
.. _api_v2.layer_context_projection:
context_projection
------------------
.. autoclass:: paddle.v2.layer.context_projection
.. autofunction:: paddle.v2.layer.context_projection
:noindex:
row_conv
--------
.. autoclass:: paddle.v2.layer.row_conv
.. autofunction:: paddle.v2.layer.row_conv
:noindex:
Image Pooling Layer
......@@ -69,27 +69,27 @@ Image Pooling Layer
img_pool
--------
.. autoclass:: paddle.v2.layer.img_pool
.. autofunction:: paddle.v2.layer.img_pool
:noindex:
spp
---
.. autoclass:: paddle.v2.layer.spp
.. autofunction:: paddle.v2.layer.spp
:noindex:
maxout
------
.. autoclass:: paddle.v2.layer.maxout
.. autofunction:: paddle.v2.layer.maxout
:noindex:
roi_pool
--------
.. autoclass:: paddle.v2.layer.roi_pool
.. autofunction:: paddle.v2.layer.roi_pool
:noindex:
pad
----
.. autoclass:: paddle.v2.layer.pad
.. autofunction:: paddle.v2.layer.pad
:noindex:
Norm Layer
......@@ -97,27 +97,27 @@ Norm Layer
img_cmrnorm
-----------
.. autoclass:: paddle.v2.layer.img_cmrnorm
.. autofunction:: paddle.v2.layer.img_cmrnorm
:noindex:
batch_norm
----------
.. autoclass:: paddle.v2.layer.batch_norm
.. autofunction:: paddle.v2.layer.batch_norm
:noindex:
sum_to_one_norm
---------------
.. autoclass:: paddle.v2.layer.sum_to_one_norm
.. autofunction:: paddle.v2.layer.sum_to_one_norm
:noindex:
cross_channel_norm
------------------
.. autoclass:: paddle.v2.layer.cross_channel_norm
.. autofunction:: paddle.v2.layer.cross_channel_norm
:noindex:
row_l2_norm
-----------
.. autoclass:: paddle.v2.layer.row_l2_norm
.. autofunction:: paddle.v2.layer.row_l2_norm
:noindex:
Recurrent Layers
......@@ -125,22 +125,22 @@ Recurrent Layers
recurrent
---------
.. autoclass:: paddle.v2.layer.recurrent
.. autofunction:: paddle.v2.layer.recurrent
:noindex:
lstmemory
---------
.. autoclass:: paddle.v2.layer.lstmemory
.. autofunction:: paddle.v2.layer.lstmemory
:noindex:
grumemory
---------
.. autoclass:: paddle.v2.layer.grumemory
.. autofunction:: paddle.v2.layer.grumemory
:noindex:
gated_unit
-----------
.. autoclass:: paddle.v2.layer.gated_unit
.. autofunction:: paddle.v2.layer.gated_unit
:noindex:
Recurrent Layer Group
......@@ -148,32 +148,32 @@ Recurrent Layer Group
memory
------
.. autoclass:: paddle.v2.layer.memory
.. autofunction:: paddle.v2.layer.memory
:noindex:
recurrent_group
---------------
.. autoclass:: paddle.v2.layer.recurrent_group
.. autofunction:: paddle.v2.layer.recurrent_group
:noindex:
lstm_step
---------
.. autoclass:: paddle.v2.layer.lstm_step
.. autofunction:: paddle.v2.layer.lstm_step
:noindex:
gru_step
--------
.. autoclass:: paddle.v2.layer.gru_step
.. autofunction:: paddle.v2.layer.gru_step
:noindex:
beam_search
------------
.. autoclass:: paddle.v2.layer.beam_search
.. autofunction:: paddle.v2.layer.beam_search
:noindex:
get_output
----------
.. autoclass:: paddle.v2.layer.get_output
.. autofunction:: paddle.v2.layer.get_output
:noindex:
Mixed Layer
......@@ -183,54 +183,54 @@ Mixed Layer
mixed
-----
.. autoclass:: paddle.v2.layer.mixed
.. autofunction:: paddle.v2.layer.mixed
:noindex:
.. _api_v2.layer_embedding:
embedding
---------
.. autoclass:: paddle.v2.layer.embedding
.. autofunction:: paddle.v2.layer.embedding
:noindex:
scaling_projection
------------------
.. autoclass:: paddle.v2.layer.scaling_projection
.. autofunction:: paddle.v2.layer.scaling_projection
:noindex:
dotmul_projection
-----------------
.. autoclass:: paddle.v2.layer.dotmul_projection
.. autofunction:: paddle.v2.layer.dotmul_projection
:noindex:
dotmul_operator
---------------
.. autoclass:: paddle.v2.layer.dotmul_operator
.. autofunction:: paddle.v2.layer.dotmul_operator
:noindex:
full_matrix_projection
----------------------
.. autoclass:: paddle.v2.layer.full_matrix_projection
.. autofunction:: paddle.v2.layer.full_matrix_projection
:noindex:
identity_projection
-------------------
.. autoclass:: paddle.v2.layer.identity_projection
.. autofunction:: paddle.v2.layer.identity_projection
:noindex:
slice_projection
-------------------
.. autoclass:: paddle.v2.layer.slice_projection
.. autofunction:: paddle.v2.layer.slice_projection
:noindex:
table_projection
----------------
.. autoclass:: paddle.v2.layer.table_projection
.. autofunction:: paddle.v2.layer.table_projection
:noindex:
trans_full_matrix_projection
----------------------------
.. autoclass:: paddle.v2.layer.trans_full_matrix_projection
.. autofunction:: paddle.v2.layer.trans_full_matrix_projection
:noindex:
Aggregate Layers
......@@ -245,51 +245,46 @@ AggregateLevel
pooling
-------
.. autoclass:: paddle.v2.layer.pooling
.. autofunction:: paddle.v2.layer.pooling
:noindex:
.. _api_v2.layer_last_seq:
last_seq
--------
.. autoclass:: paddle.v2.layer.last_seq
.. autofunction:: paddle.v2.layer.last_seq
:noindex:
.. _api_v2.layer_first_seq:
first_seq
---------
.. autoclass:: paddle.v2.layer.first_seq
.. autofunction:: paddle.v2.layer.first_seq
:noindex:
sub_seq
---------
.. autoclass:: paddle.v2.layer.sub_seq
.. autofunction:: paddle.v2.layer.sub_seq
:noindex:
concat
------
.. autoclass:: paddle.v2.layer.concat
.. autofunction:: paddle.v2.layer.concat
:noindex:
seq_concat
----------
.. autoclass:: paddle.v2.layer.seq_concat
.. autofunction:: paddle.v2.layer.seq_concat
:noindex:
seq_slice
---------
.. autoclass:: paddle.v2.layer.seq_slice
:noindex:
kmax_sequence_score
-------------------
.. autoclass:: paddle.v2.layer.kmax_sequence_score
.. autofunction:: paddle.v2.layer.seq_slice
:noindex:
sub_nested_seq
--------------
.. autoclass:: paddle.v2.layer.sub_nested_seq
.. autofunction:: paddle.v2.layer.sub_nested_seq
:noindex:
Reshaping Layers
......@@ -297,7 +292,7 @@ Reshaping Layers
block_expand
------------
.. autoclass:: paddle.v2.layer.block_expand
.. autofunction:: paddle.v2.layer.block_expand
:noindex:
.. _api_v2.layer_expand:
......@@ -309,22 +304,22 @@ ExpandLevel
expand
------
.. autoclass:: paddle.v2.layer.expand
.. autofunction:: paddle.v2.layer.expand
:noindex:
repeat
------
.. autoclass:: paddle.v2.layer.repeat
.. autofunction:: paddle.v2.layer.repeat
:noindex:
rotate
------
.. autoclass:: paddle.v2.layer.rotate
.. autofunction:: paddle.v2.layer.rotate
:noindex:
seq_reshape
-----------
.. autoclass:: paddle.v2.layer.seq_reshape
.. autofunction:: paddle.v2.layer.seq_reshape
:noindex:
Math Layers
......@@ -332,94 +327,94 @@ Math Layers
addto
-----
.. autoclass:: paddle.v2.layer.addto
.. autofunction:: paddle.v2.layer.addto
:noindex:
linear_comb
-----------
.. autoclass:: paddle.v2.layer.linear_comb
.. autofunction:: paddle.v2.layer.linear_comb
:noindex:
interpolation
-------------
.. autoclass:: paddle.v2.layer.interpolation
.. autofunction:: paddle.v2.layer.interpolation
:noindex:
bilinear_interp
---------------
.. autoclass:: paddle.v2.layer.bilinear_interp
.. autofunction:: paddle.v2.layer.bilinear_interp
:noindex:
dropout
--------
.. autoclass:: paddle.v2.layer.dropout
.. autofunction:: paddle.v2.layer.dropout
:noindex:
dot_prod
---------
.. autoclass:: paddle.v2.layer.dot_prod
.. autofunction:: paddle.v2.layer.dot_prod
:noindex:
out_prod
--------
.. autoclass:: paddle.v2.layer.out_prod
.. autofunction:: paddle.v2.layer.out_prod
:noindex:
power
-----
.. autoclass:: paddle.v2.layer.power
.. autofunction:: paddle.v2.layer.power
:noindex:
scaling
-------
.. autoclass:: paddle.v2.layer.scaling
.. autofunction:: paddle.v2.layer.scaling
:noindex:
clip
----
.. autoclass:: paddle.v2.layer.clip
.. autofunction:: paddle.v2.layer.clip
:noindex:
resize
------
.. autoclass:: paddle.v2.layer.resize
.. autofunction:: paddle.v2.layer.resize
:noindex:
slope_intercept
---------------
.. autoclass:: paddle.v2.layer.slope_intercept
.. autofunction:: paddle.v2.layer.slope_intercept
:noindex:
tensor
------
.. autoclass:: paddle.v2.layer.tensor
.. autofunction:: paddle.v2.layer.tensor
:noindex:
.. _api_v2.layer_cos_sim:
cos_sim
-------
.. autoclass:: paddle.v2.layer.cos_sim
.. autofunction:: paddle.v2.layer.cos_sim
:noindex:
l2_distance
-----------
.. autoclass:: paddle.v2.layer.l2_distance
.. autofunction:: paddle.v2.layer.l2_distance
:noindex:
trans
-----
.. autoclass:: paddle.v2.layer.trans
.. autofunction:: paddle.v2.layer.trans
:noindex:
scale_shift
-----------
.. autoclass:: paddle.v2.layer.scale_shift
.. autofunction:: paddle.v2.layer.scale_shift
:noindex:
factorization_machine
---------------------
.. autoclass:: paddle.v2.layer.factorization_machine
.. autofunction:: paddle.v2.layer.factorization_machine
:noindex:
Sampling Layers
......@@ -427,17 +422,17 @@ Sampling Layers
maxid
-----
.. autoclass:: paddle.v2.layer.max_id
.. autofunction:: paddle.v2.layer.max_id
:noindex:
sampling_id
-----------
.. autoclass:: paddle.v2.layer.sampling_id
.. autofunction:: paddle.v2.layer.sampling_id
:noindex:
multiplex
---------
.. autoclass:: paddle.v2.layer.multiplex
.. autofunction:: paddle.v2.layer.multiplex
:noindex:
.. _api_v2.layer_costs:
......@@ -447,97 +442,97 @@ Cost Layers
cross_entropy_cost
------------------
.. autoclass:: paddle.v2.layer.cross_entropy_cost
.. autofunction:: paddle.v2.layer.cross_entropy_cost
:noindex:
cross_entropy_with_selfnorm_cost
--------------------------------
.. autoclass:: paddle.v2.layer.cross_entropy_with_selfnorm_cost
.. autofunction:: paddle.v2.layer.cross_entropy_with_selfnorm_cost
:noindex:
multi_binary_label_cross_entropy_cost
-------------------------------------
.. autoclass:: paddle.v2.layer.multi_binary_label_cross_entropy_cost
.. autofunction:: paddle.v2.layer.multi_binary_label_cross_entropy_cost
:noindex:
classification_cost
-------------------
.. autoclass:: paddle.v2.layer.classification_cost
.. autofunction:: paddle.v2.layer.classification_cost
:noindex:
huber_regression_cost
-------------------------
.. autoclass:: paddle.v2.layer.huber_regression_cost
.. autofunction:: paddle.v2.layer.huber_regression_cost
:noindex:
huber_classification_cost
-------------------------
.. autoclass:: paddle.v2.layer.huber_classification_cost
.. autofunction:: paddle.v2.layer.huber_classification_cost
:noindex:
lambda_cost
-----------
.. autoclass:: paddle.v2.layer.lambda_cost
.. autofunction:: paddle.v2.layer.lambda_cost
:noindex:
square_error_cost
-----------------
.. autoclass:: paddle.v2.layer.square_error_cost
.. autofunction:: paddle.v2.layer.square_error_cost
:noindex:
rank_cost
---------
.. autoclass:: paddle.v2.layer.rank_cost
.. autofunction:: paddle.v2.layer.rank_cost
:noindex:
sum_cost
---------
.. autoclass:: paddle.v2.layer.sum_cost
.. autofunction:: paddle.v2.layer.sum_cost
:noindex:
crf
---
.. autoclass:: paddle.v2.layer.crf
.. autofunction:: paddle.v2.layer.crf
:noindex:
crf_decoding
------------
.. autoclass:: paddle.v2.layer.crf_decoding
.. autofunction:: paddle.v2.layer.crf_decoding
:noindex:
ctc
---
.. autoclass:: paddle.v2.layer.ctc
.. autofunction:: paddle.v2.layer.ctc
:noindex:
warp_ctc
--------
.. autoclass:: paddle.v2.layer.warp_ctc
.. autofunction:: paddle.v2.layer.warp_ctc
:noindex:
nce
---
.. autoclass:: paddle.v2.layer.nce
.. autofunction:: paddle.v2.layer.nce
:noindex:
hsigmoid
---------
.. autoclass:: paddle.v2.layer.hsigmoid
.. autofunction:: paddle.v2.layer.hsigmoid
:noindex:
smooth_l1_cost
--------------
.. autoclass:: paddle.v2.layer.smooth_l1_cost
.. autofunction:: paddle.v2.layer.smooth_l1_cost
:noindex:
multibox_loss
--------------
.. autoclass:: paddle.v2.layer.multibox_loss
.. autofunction:: paddle.v2.layer.multibox_loss
:noindex:
detection_output
----------------
.. autoclass:: paddle.v2.layer.detection_output
.. autofunction:: paddle.v2.layer.detection_output
:noindex:
Check Layer
......@@ -545,7 +540,7 @@ Check Layer
eos
---
.. autoclass:: paddle.v2.layer.eos
.. autofunction:: paddle.v2.layer.eos
:noindex:
Activation
......@@ -553,5 +548,5 @@ Activation
prelu
--------
.. autoclass:: paddle.v2.layer.prelu
.. autofunction:: paddle.v2.layer.prelu
:noindex:
doc/v2/api/index_en.rst
浏览文件 @ 29bf727e
......@@ -8,4 +8,3 @@ API
model_configs.rst
data.rst
run_logic.rst
fluid/index.rst
doc/v2/build_and_install/pip_install_cn.rst
浏览文件 @ 29bf727e
......@@ -60,6 +60,7 @@ paddlepaddle-gpu==0.11.0 使用CUDA 7.5和cuDNN 5编译的0.11.0版
"cpu_noavx_openblas", "`paddlepaddle-latest-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuNoavxOpenblas/.lastSuccessful/paddlepaddle-latest-cp27-cp27mu-linux_x86_64.whl>`__", "`paddlepaddle-latest-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuNoavxOpenblas/.lastSuccessful/paddlepaddle-latest-cp27-cp27m-linux_x86_64.whl>`_"
"cuda8.0_cudnn5_avx_mkl", "`paddlepaddle_gpu-latest-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda80cudnn5cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-latest-cp27-cp27mu-linux_x86_64.whl>`__", "`paddlepaddle_gpu-latest-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda80cudnn5cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-latest-cp27-cp27m-linux_x86_64.whl>`__"
"cuda8.0_cudnn7_avx_mkl", "`paddlepaddle_gpu-latest-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda8cudnn7cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-latest-cp27-cp27mu-linux_x86_64.whl>`__", "`paddlepaddle_gpu-latest-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda8cudnn7cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-latest-cp27-cp27m-linux_x86_64.whl>`__"
"cuda9.0_cudnn7_avx_mkl", "`paddlepaddle_gpu-latest-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda90cudnn7avxMkl/.lastSuccessful/paddlepaddle_gpu-latest-cp27-cp27mu-linux_x86_64.whl>`__", "`paddlepaddle_gpu-latest-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda90cudnn7avxMkl/.lastSuccessful/paddlepaddle_gpu-latest-cp27-cp27m-linux_x86_64.whl>`__"
.. _pip_dependency:
......
doc/v2/build_and_install/pip_install_en.rst
浏览文件 @ 29bf727e
......@@ -63,6 +63,7 @@ If the links below shows up the login form, just click "Log in as guest" to star
"cpu_noavx_openblas", "`paddlepaddle-latest-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuNoavxOpenblas/.lastSuccessful/paddlepaddle-latest-cp27-cp27mu-linux_x86_64.whl>`__", "`paddlepaddle-latest-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuNoavxOpenblas/.lastSuccessful/paddlepaddle-latest-cp27-cp27m-linux_x86_64.whl>`__"
"cuda8.0_cudnn5_avx_mkl", "`paddlepaddle_gpu-latest-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda80cudnn5cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-latest-cp27-cp27mu-linux_x86_64.whl>`__", "`paddlepaddle_gpu-latest-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda80cudnn5cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-latest-cp27-cp27m-linux_x86_64.whl>`__"
"cuda8.0_cudnn7_avx_mkl", "`paddlepaddle_gpu-latest-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda8cudnn7cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-latest-cp27-cp27mu-linux_x86_64.whl>`__", "`paddlepaddle_gpu-latest-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda8cudnn7cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-latest-cp27-cp27m-linux_x86_64.whl>`__"
"cuda9.0_cudnn7_avx_mkl", "`paddlepaddle_gpu-latest-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda90cudnn7avxMkl/.lastSuccessful/paddlepaddle_gpu-latest-cp27-cp27mu-linux_x86_64.whl>`__", "`paddlepaddle_gpu-latest-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda90cudnn7avxMkl/.lastSuccessful/paddlepaddle_gpu-latest-cp27-cp27m-linux_x86_64.whl>`__"
.. _pip_dependency:
......
paddle/fluid/operators/conv_mkldnn_op.cc
浏览文件 @ 29bf727e
......@@ -18,6 +18,17 @@
namespace paddle {
namespace operators {
using conv_bwd_data = mkldnn::convolution_backward_data;
using conv_bwd_weights = mkldnn::convolution_backward_weights;
using conv_fwd = mkldnn::convolution_forward;
using framework::DataLayout;
using mkldnn::memory;
using mkldnn::primitive;
using mkldnn::reorder;
using mkldnn::stream;
using platform::to_void_cast;
using platform::GetMKLDNNFormat;
template <typename T>
class ConvMKLDNNOpKernel : public paddle::framework::OpKernel<T> {
public:
......@@ -25,6 +36,10 @@ class ConvMKLDNNOpKernel : public paddle::framework::OpKernel<T> {
PADDLE_ENFORCE(paddle::platform::is_cpu_place(ctx.GetPlace()),
"It must use CPUPlace.");
// Get unique name for index
const std::string key = ctx.op().Output("Output");
const std::string key_conv_pd = key + "@conv_pd";
auto& dev_ctx =
ctx.template device_context<paddle::platform::MKLDNNDeviceContext>();
const auto& mkldnn_engine = dev_ctx.GetEngine();
......@@ -33,10 +48,12 @@ class ConvMKLDNNOpKernel : public paddle::framework::OpKernel<T> {
auto* filter = ctx.Input<Tensor>("Filter");
auto* output = ctx.Output<Tensor>("Output");
// Get an unique name from "argument" name of "Output" variable
// This name will be used as key when saving info into device context
const std::string key = ctx.op().Output("Output");
const std::string key_conv_pd = key + "@conv_pd";
PADDLE_ENFORCE(input->layout() == DataLayout::kMKLDNN &&
input->format() != memory::format::format_undef,
"Wrong layout/format set for Input tensor");
PADDLE_ENFORCE(filter->layout() == DataLayout::kMKLDNN &&
filter->format() != memory::format::format_undef,
"Wrong layout/format set for Filter tensor");
std::vector<int> strides = ctx.Attr<std::vector<int>>("strides");
std::vector<int> paddings = ctx.Attr<std::vector<int>>("paddings");
......@@ -63,60 +80,86 @@ class ConvMKLDNNOpKernel : public paddle::framework::OpKernel<T> {
paddle::framework::vectorize2int(filter->dims());
std::vector<int> dst_tz = paddle::framework::vectorize2int(output->dims());
// TODO(pzelazko-intel): support more formats
auto src_md = platform::MKLDNNMemDesc(
src_tz, mkldnn::memory::data_type::f32, mkldnn::memory::format::nchw);
auto weights_md =
platform::MKLDNNMemDesc(weights_tz, mkldnn::memory::data_type::f32,
mkldnn::memory::format::oihw);
auto dst_md = platform::MKLDNNMemDesc(
dst_tz, mkldnn::memory::data_type::f32, mkldnn::memory::format::nchw);
auto src_memory =
mkldnn::memory({src_md, mkldnn_engine},
reinterpret_cast<void*>(const_cast<T*>(input_data)));
auto weights_memory =
mkldnn::memory({weights_md, mkldnn_engine},
reinterpret_cast<void*>(const_cast<T*>(filter_data)));
auto dst_memory = mkldnn::memory({dst_md, mkldnn_engine}, output_data);
std::shared_ptr<mkldnn::convolution_forward::primitive_desc> conv_pd =
ConvFwdPrimitiveDesc(src_md, weights_md, dst_md, strides, paddings,
mkldnn_engine);
// save conv_pd into global device context to be referred in backward path
dev_ctx.SetBlob(key_conv_pd, conv_pd);
// create mkldnn memory from input tensors (data/weights)
auto user_src_memory = memory(
{{{src_tz}, memory::data_type::f32, input->format()}, mkldnn_engine},
to_void_cast(input_data));
auto user_weights_memory =
memory({{{weights_tz}, memory::data_type::f32, filter->format()},
mkldnn_engine},
to_void_cast(filter_data));
/* create memory descriptor for convolution without specified format
* ('any') which lets a primitive (convolution in this case) choose
* the memory format preferred for best performance
*/
auto src_md = platform::MKLDNNMemDesc(src_tz, memory::data_type::f32,
memory::format::any);
auto weights_md = platform::MKLDNNMemDesc(
weights_tz, memory::data_type::f32, memory::format::any);
auto dst_md = platform::MKLDNNMemDesc(dst_tz, memory::data_type::f32,
memory::format::any);
// create a conv primitive descriptor and save it for usage in backward
std::shared_ptr<conv_fwd::primitive_desc> conv_pd = ConvFwdPrimitiveDesc(
src_md, weights_md, dst_md, strides, paddings, mkldnn_engine);
// create reorder primitive if the input format is not the preferred one
auto src_memory = user_src_memory;
primitive reorder_src;
bool is_src_reordered = false;
if (memory::primitive_desc(conv_pd->src_primitive_desc()) !=
user_src_memory.get_primitive_desc()) {
src_memory = memory(conv_pd->src_primitive_desc());
reorder_src = reorder(user_src_memory, src_memory);
is_src_reordered = true;
}
auto weights_memory = user_weights_memory;
primitive reorder_weights;
bool is_weights_reordered = false;
if (memory::primitive_desc(conv_pd->weights_primitive_desc()) !=
user_weights_memory.get_primitive_desc()) {
weights_memory = memory(conv_pd->weights_primitive_desc());
reorder_weights = reorder(user_weights_memory, weights_memory);
is_weights_reordered = true;
}
// create memory primitive for conv dst
auto dst_memory = memory(conv_pd->dst_primitive_desc(), output_data);
// create convolution op primitive
auto conv_prim = mkldnn::convolution_forward(*conv_pd, src_memory,
weights_memory, dst_memory);
auto conv_prim = conv_fwd(*conv_pd, src_memory, weights_memory, dst_memory);
// push primitive to stream and wait until it's executed
std::vector<mkldnn::primitive> pipeline{conv_prim};
mkldnn::stream(mkldnn::stream::kind::eager).submit(pipeline).wait();
std::vector<primitive> pipeline;
if (is_src_reordered) pipeline.push_back(reorder_src);
if (is_weights_reordered) pipeline.push_back(reorder_weights);
pipeline.push_back(conv_prim);
stream(stream::kind::eager).submit(pipeline).wait();
// Save conv_pd/src_memory/weights_memory for backward pass
dev_ctx.SetBlob(key_conv_pd, conv_pd);
output->set_layout(DataLayout::kMKLDNN);
output->set_format(GetMKLDNNFormat(dst_memory));
}
private:
std::unique_ptr<mkldnn::convolution_forward::primitive_desc>
ConvFwdPrimitiveDesc(const mkldnn::memory::desc& src,
const mkldnn::memory::desc& weights,
const mkldnn::memory::desc& dst,
const std::vector<int>& strides,
const std::vector<int>& paddings,
const mkldnn::engine& engine) const {
mkldnn::memory::dims stride_dims = {strides[0], strides[1]};
mkldnn::memory::dims padding_dims = {paddings[0], paddings[1]};
auto conv_desc = mkldnn::convolution_forward::desc(
mkldnn::prop_kind::forward, mkldnn::convolution_direct, src, weights,
dst, stride_dims, padding_dims, padding_dims,
mkldnn::padding_kind::zero);
auto p_conv_pd =
new mkldnn::convolution_forward::primitive_desc(conv_desc, engine);
return std::unique_ptr<mkldnn::convolution_forward::primitive_desc>(
p_conv_pd);
std::unique_ptr<conv_fwd::primitive_desc> ConvFwdPrimitiveDesc(
const memory::desc& src, const memory::desc& weights,
const memory::desc& dst, const std::vector<int>& strides,
const std::vector<int>& paddings, const mkldnn::engine& engine) const {
memory::dims stride_dims = {strides[0], strides[1]};
memory::dims padding_dims = {paddings[0], paddings[1]};
auto conv_desc =
conv_fwd::desc(mkldnn::prop_kind::forward, mkldnn::convolution_direct,
src, weights, dst, stride_dims, padding_dims,
padding_dims, mkldnn::padding_kind::zero);
auto p_conv_pd = new conv_fwd::primitive_desc(conv_desc, engine);
return std::unique_ptr<conv_fwd::primitive_desc>(p_conv_pd);
}
};
......@@ -139,6 +182,19 @@ class ConvMKLDNNGradOpKernel : public paddle::framework::OpKernel<T> {
Tensor* input_grad = ctx.Output<Tensor>(framework::GradVarName("Input"));
Tensor* filter_grad = ctx.Output<Tensor>(framework::GradVarName("Filter"));
PADDLE_ENFORCE(input->layout() == DataLayout::kMKLDNN &&
input->format() != memory::format::format_undef,
"Wrong layout/format set for Input tensor");
PADDLE_ENFORCE(filter->layout() == DataLayout::kMKLDNN &&
filter->format() != memory::format::format_undef,
"Wrong layout/format set for Filter tensor");
PADDLE_ENFORCE(output->layout() == DataLayout::kMKLDNN &&
output->format() != memory::format::format_undef,
"Wrong layout/format set for Output tensor");
PADDLE_ENFORCE(output_grad->layout() == DataLayout::kMKLDNN &&
output_grad->format() != memory::format::format_undef,
"Wrong layout/format set for output_grad tensor");
if (!input_grad && !filter_grad) return;
// Get an unique name from "argument" name of "Output" variable
......@@ -167,108 +223,147 @@ class ConvMKLDNNGradOpKernel : public paddle::framework::OpKernel<T> {
paddle::framework::vectorize2int(filter->dims());
std::vector<int> dst_tz = paddle::framework::vectorize2int(output->dims());
// TODO(pzelazko-intel): support more formats
auto src_md = platform::MKLDNNMemDesc(
src_tz, mkldnn::memory::data_type::f32, mkldnn::memory::format::nchw);
auto diff_src_md = platform::MKLDNNMemDesc(
src_tz, mkldnn::memory::data_type::f32, mkldnn::memory::format::nchw);
auto weights_md =
platform::MKLDNNMemDesc(weights_tz, mkldnn::memory::data_type::f32,
mkldnn::memory::format::oihw);
auto diff_weights_md =
platform::MKLDNNMemDesc(weights_tz, mkldnn::memory::data_type::f32,
mkldnn::memory::format::oihw);
auto diff_dst_md = platform::MKLDNNMemDesc(
dst_tz, mkldnn::memory::data_type::f32, mkldnn::memory::format::nchw);
// create memory
auto diff_dst_memory = mkldnn::memory(
{diff_weights_md, mkldnn_engine},
reinterpret_cast<void*>(const_cast<T*>(output_grad_data)));
// create mkldnn memory from input tensors (input/weights/output_grad)
auto user_src_memory = memory(
{{{src_tz}, memory::data_type::f32, input->format()}, mkldnn_engine},
to_void_cast(input_data));
auto user_weights_memory =
memory({{{weights_tz}, memory::data_type::f32, filter->format()},
mkldnn_engine},
to_void_cast(filter_data));
auto user_diff_dst_memory =
memory({{{dst_tz}, memory::data_type::f32, output_grad->format()},
mkldnn_engine},
to_void_cast(output_grad_data));
/* create memory descriptor for conv backward without specified format
* ('any') which lets a primitive (conv backward in this case) choose
* the memory format preferred for best performance
*/
auto src_md = platform::MKLDNNMemDesc(src_tz, memory::data_type::f32,
memory::format::any);
auto diff_src_md = platform::MKLDNNMemDesc(src_tz, memory::data_type::f32,
memory::format::any);
auto weights_md = platform::MKLDNNMemDesc(
weights_tz, memory::data_type::f32, memory::format::any);
auto diff_weights_md = platform::MKLDNNMemDesc(
weights_tz, memory::data_type::f32, memory::format::any);
auto diff_dst_md = platform::MKLDNNMemDesc(dst_tz, memory::data_type::f32,
memory::format::any);
// Retrieve conv_pd from device context
auto conv_pd =
std::static_pointer_cast<mkldnn::convolution_forward::primitive_desc>(
dev_ctx.GetBlob(key_conv_pd));
auto conv_pd = std::static_pointer_cast<conv_fwd::primitive_desc>(
dev_ctx.GetBlob(key_conv_pd));
PADDLE_ENFORCE(conv_pd != nullptr,
"Fail to find conv_pd in device context");
// create backward conv primitive for weights
if (filter_grad) {
// create primitive descriptor
mkldnn::convolution_backward_weights::primitive_desc conv_bwd_weights_pd =
ConvBwdWeightsPrimitiveDesc(src_md, diff_weights_md, diff_dst_md,
strides, paddings, *conv_pd,
mkldnn_engine);
// create memory
// create backward convolution primitive descriptor
auto conv_bwd_weights_desc = conv_bwd_weights::desc(
mkldnn::convolution_direct, src_md, diff_weights_md, diff_dst_md,
strides, paddings, paddings, mkldnn::padding_kind::zero);
auto conv_bwd_weights_pd = conv_bwd_weights::primitive_desc(
conv_bwd_weights_desc, mkldnn_engine, *conv_pd);
// create reorder primitive if the input format is not the preferred one
auto src_memory = user_src_memory;
primitive reorder_src;
bool is_src_reordered = false;
if (memory::primitive_desc(conv_bwd_weights_pd.src_primitive_desc()) !=
user_src_memory.get_primitive_desc()) {
src_memory = memory(conv_bwd_weights_pd.src_primitive_desc());
reorder_src = reorder(user_src_memory, src_memory);
is_src_reordered = true;
}
auto diff_dst_memory_4filter = user_diff_dst_memory;
primitive reorder_diff_dst_4filter;
bool is_diff_dst_reordered_4filter = false;
if (memory::primitive_desc(
conv_bwd_weights_pd.diff_dst_primitive_desc()) !=
user_diff_dst_memory.get_primitive_desc()) {
diff_dst_memory_4filter =
memory(conv_bwd_weights_pd.diff_dst_primitive_desc());
reorder_diff_dst_4filter =
reorder(user_diff_dst_memory, diff_dst_memory_4filter);
is_diff_dst_reordered_4filter = true;
}
// create mkldnn memory for output (i.e. diff weights)
auto diff_weights_memory =
mkldnn::memory({diff_weights_md, mkldnn_engine},
reinterpret_cast<void*>(filter_grad_data));
auto src_memory =
mkldnn::memory({src_md, mkldnn_engine},
reinterpret_cast<void*>(const_cast<T*>(input_data)));
memory(conv_bwd_weights_pd.diff_weights_primitive_desc(),
reinterpret_cast<void*>(filter_grad_data));
// create backward conv primitive for weights
auto conv_bwd_weights_prim = mkldnn::convolution_backward_weights(
conv_bwd_weights_pd, src_memory, diff_dst_memory,
diff_weights_memory);
auto conv_bwd_weights_prim =
conv_bwd_weights(conv_bwd_weights_pd, src_memory,
diff_dst_memory_4filter, diff_weights_memory);
// push primitive and execute it
std::vector<mkldnn::primitive> pipeline{conv_bwd_weights_prim};
mkldnn::stream(mkldnn::stream::kind::eager).submit(pipeline).wait();
std::vector<primitive> pipeline;
if (is_src_reordered) pipeline.push_back(reorder_src);
if (is_diff_dst_reordered_4filter)
pipeline.push_back(reorder_diff_dst_4filter);
pipeline.push_back(conv_bwd_weights_prim);
stream(stream::kind::eager).submit(pipeline).wait();
filter_grad->set_layout(DataLayout::kMKLDNN);
filter_grad->set_format(GetMKLDNNFormat(diff_weights_memory));
}
if (input_grad) {
// create primitive descriptor
mkldnn::convolution_backward_data::primitive_desc conv_bwd_data_pd =
ConvBwdDataPrimitiveDesc(diff_src_md, weights_md, diff_dst_md,
strides, paddings, *conv_pd, mkldnn_engine);
// create memory
auto diff_src_memory = mkldnn::memory(
{diff_src_md, mkldnn_engine},
reinterpret_cast<void*>(const_cast<T*>(input_grad_data)));
auto weights_memory =
mkldnn::memory({weights_md, mkldnn_engine},
reinterpret_cast<void*>(const_cast<T*>(filter_data)));
// create backward convolution primitive descriptor
auto conv_bwd_data_desc = conv_bwd_data::desc(
mkldnn::convolution_direct, diff_src_md, weights_md, diff_dst_md,
strides, paddings, paddings, mkldnn::padding_kind::zero);
auto conv_bwd_data_pd = conv_bwd_data::primitive_desc(
conv_bwd_data_desc, mkldnn_engine, *conv_pd);
// create reorder primitive if the input format is not the preferred one
auto weights_memory = user_weights_memory;
primitive reorder_weights;
bool is_weights_reordered = false;
if (memory::primitive_desc(conv_bwd_data_pd.weights_primitive_desc()) !=
user_weights_memory.get_primitive_desc()) {
weights_memory = memory(conv_bwd_data_pd.weights_primitive_desc());
reorder_weights = reorder(user_weights_memory, weights_memory);
is_weights_reordered = true;
}
auto diff_dst_memory_4data = user_diff_dst_memory;
primitive reorder_diff_dst_4data;
bool is_diff_dst_reordered_4data = false;
if (memory::primitive_desc(conv_bwd_data_pd.diff_dst_primitive_desc()) !=
user_diff_dst_memory.get_primitive_desc()) {
diff_dst_memory_4data =
memory(conv_bwd_data_pd.diff_dst_primitive_desc());
reorder_diff_dst_4data =
reorder(user_diff_dst_memory, diff_dst_memory_4data);
is_diff_dst_reordered_4data = true;
}
// create mkldnn memory for output (i.e. diff src)
auto diff_src_memory = memory(conv_bwd_data_pd.diff_src_primitive_desc(),
reinterpret_cast<void*>(input_grad_data));
// create backward conv primitive for data
auto conv_bwd_data_prim = mkldnn::convolution_backward_data(
conv_bwd_data_pd, diff_dst_memory, weights_memory, diff_src_memory);
auto conv_bwd_data_prim =
conv_bwd_data(conv_bwd_data_pd, diff_dst_memory_4data, weights_memory,
diff_src_memory);
// push primitive to stream and wait until it's executed
std::vector<mkldnn::primitive> pipeline{conv_bwd_data_prim};
mkldnn::stream(mkldnn::stream::kind::eager).submit(pipeline).wait();
// push primitive and execute it
std::vector<primitive> pipeline;
if (is_weights_reordered) pipeline.push_back(reorder_weights);
if (is_diff_dst_reordered_4data)
pipeline.push_back(reorder_diff_dst_4data);
pipeline.push_back(conv_bwd_data_prim);
stream(stream::kind::eager).submit(pipeline).wait();
input_grad->set_layout(DataLayout::kMKLDNN);
input_grad->set_format(GetMKLDNNFormat(diff_src_memory));
}
} // Compute()
private:
mkldnn::convolution_backward_weights::primitive_desc
ConvBwdWeightsPrimitiveDesc(
const mkldnn::memory::desc& src, const mkldnn::memory::desc& diff_weights,
const mkldnn::memory::desc& diff_dst, const std::vector<int>& strides,
const std::vector<int>& paddings,
const mkldnn::convolution_forward::primitive_desc& conv_pd,
const mkldnn::engine& engine) const {
auto conv_bwd_weights_desc = mkldnn::convolution_backward_weights::desc(
mkldnn::convolution_direct, src, diff_weights, diff_dst, strides,
paddings, paddings, mkldnn::padding_kind::zero);
return mkldnn::convolution_backward_weights::primitive_desc(
conv_bwd_weights_desc, engine, conv_pd);
}
mkldnn::convolution_backward_data::primitive_desc ConvBwdDataPrimitiveDesc(
const mkldnn::memory::desc& diff_src, const mkldnn::memory::desc& weights,
const mkldnn::memory::desc& diff_dst, const std::vector<int>& strides,
const std::vector<int>& paddings,
const mkldnn::convolution_forward::primitive_desc& conv_pd,
const mkldnn::engine& engine) const {
auto conv_bwd_data_desc = mkldnn::convolution_backward_data::desc(
mkldnn::convolution_direct, diff_src, weights, diff_dst, strides,
paddings, paddings, mkldnn::padding_kind::zero);
return mkldnn::convolution_backward_data::primitive_desc(conv_bwd_data_desc,
engine, conv_pd);
}
};
} // namespace operators
......
paddle/fluid/operators/conv_op.cc
浏览文件 @ 29bf727e
......@@ -75,9 +75,8 @@ void ConvOp::InferShape(framework::InferShapeContext* ctx) const {
framework::OpKernelType ConvOp::GetExpectedKernelType(
const framework::ExecutionContext& ctx) const {
framework::LibraryType library{framework::LibraryType::kPlain};
std::string data_format = ctx.Attr<std::string>("data_format");
// TODO(pzelazko-intel): enable MKLDNN layout when it's ready
std::string data_format = ctx.Attr<std::string>("data_format");
framework::DataLayout layout = framework::StringToDataLayout(data_format);
#ifdef PADDLE_WITH_CUDA
......
paddle/fluid/operators/gen_nccl_id_op.cc
浏览文件 @ 29bf727e
......@@ -67,6 +67,10 @@ class GenNCCLIdOp : public framework::OperatorBase {
client->AsyncSendVar(ep, dev_ctx, *scope, NCCL_ID_VARNAME);
}
client->Wait();
for (auto& ep : endpoint_list) {
client->AsyncSendBatchBarrier(ep);
}
client->Wait();
VLOG(3) << "sending completed...";
}
......
python/paddle/batch.py
浏览文件 @ 29bf727e
......@@ -15,7 +15,7 @@
__all__ = ['batch']
def batch(reader, batch_size, drop_last=False):
def batch(reader, batch_size, drop_last=True):
"""
Create a batched reader.
......
python/paddle/fluid/layers/nn.py
浏览文件 @ 29bf727e
......@@ -262,9 +262,10 @@ def embedding(input,
return tmp
# TODO(qijun): expose H0 and C0
def dynamic_lstm(input,
size,
h_0=None,
c_0=None,
param_attr=None,
bias_attr=None,
use_peepholes=True,
......@@ -325,6 +326,13 @@ def dynamic_lstm(input,
(T X 4D), where T is the total time steps in this
mini-batch, D is the hidden size.
size(int): 4 * hidden size.
h_0(Variable): The initial hidden state is an optional input, default is zero.
This is a tensor with shape (N x D), where N is the
batch size and D is the hidden size.
c_0(Variable): The initial cell state is an optional input, default is zero.
This is a tensor with shape (N x D), where N is the
batch size. `h_0` and `c_0` can be NULL but only at the same time.
param_attr(ParamAttr|None): The parameter attribute for the learnable
hidden-hidden weights.
......@@ -388,12 +396,20 @@ def dynamic_lstm(input,
cell = helper.create_tmp_variable(dtype)
batch_gate = helper.create_tmp_variable(dtype)
batch_cell_pre_act = helper.create_tmp_variable(dtype)
inputs = {'Input': input, 'Weight': weight, 'Bias': bias}
batch_size = input.shape[0]
if h_0:
assert h_0.shape == (batch_size, size), \
'The shape of h0 should be (batch_size, %d)' % size
inputs['H0'] = h_0
if c_0:
assert c_0.shape == (batch_size, size), \
'The shape of c0 should be (batch_size, %d)' % size
inputs['C0'] = c_0
helper.append_op(
type='lstm',
inputs={'Input': input,
'Weight': weight,
'Bias': bias},
inputs=inputs,
outputs={
'Hidden': hidden,
'Cell': cell,
......@@ -678,11 +694,13 @@ def dynamic_gru(input,
attr=helper.param_attr, shape=[size, 3 * size], dtype=dtype)
bias = helper.create_parameter(
attr=helper.bias_attr, shape=[1, 3 * size], dtype=dtype, is_bias=True)
batch_size = input.shape[0]
inputs = {'Input': input, 'Weight': weight, 'Bias': bias}
if h_0 != None:
assert h_0.shape == (
size, size), 'The shape of h0 should be(%d, %d)' % (size, size)
inputs['h0'] = h_0
batch_size, size
), 'The shape of h0 should be(batch_size, %d)' % size
inputs['H0'] = h_0
hidden = helper.create_tmp_variable(dtype)
batch_gate = helper.create_tmp_variable(dtype)
......
python/paddle/fluid/tests/book/high-level-api/image_classification/test_image_classification_resnet.py
浏览文件 @ 29bf727e
......@@ -96,10 +96,11 @@ def train(use_cuda, train_program, params_dirname):
train_reader = paddle.batch(
paddle.reader.shuffle(
cifar10_small_test_set.train10(batch_size=10), buf_size=128 * 10),
batch_size=BATCH_SIZE)
batch_size=BATCH_SIZE,
drop_last=False)
test_reader = paddle.batch(
paddle.dataset.cifar.test10(), batch_size=BATCH_SIZE)
paddle.dataset.cifar.test10(), batch_size=BATCH_SIZE, drop_last=False)
def event_handler(event):
if isinstance(event, fluid.EndStepEvent):
......
python/paddle/fluid/tests/book/high-level-api/image_classification/test_image_classification_vgg.py
浏览文件 @ 29bf727e
......@@ -73,10 +73,11 @@ def train(use_cuda, train_program, params_dirname):
train_reader = paddle.batch(
paddle.reader.shuffle(
cifar10_small_test_set.train10(batch_size=10), buf_size=128 * 10),
batch_size=BATCH_SIZE)
batch_size=BATCH_SIZE,
drop_last=False)
test_reader = paddle.batch(
paddle.dataset.cifar.test10(), batch_size=BATCH_SIZE)
paddle.dataset.cifar.test10(), batch_size=BATCH_SIZE, drop_last=False)
def event_handler(event):
if isinstance(event, fluid.EndStepEvent):
......
python/paddle/fluid/tests/book/high-level-api/understand_sentiment/test_understand_sentiment_stacked_lstm.py
浏览文件 @ 29bf727e
......@@ -87,7 +87,9 @@ def train(use_cuda, train_program, params_dirname):
def event_handler(event):
if isinstance(event, fluid.EndEpochEvent):
test_reader = paddle.batch(
paddle.dataset.imdb.test(word_dict), batch_size=BATCH_SIZE)
paddle.dataset.imdb.test(word_dict),
batch_size=BATCH_SIZE,
drop_last=False)
avg_cost, acc = trainer.test(
reader=test_reader, feed_order=['words', 'label'])
......@@ -113,7 +115,8 @@ def train(use_cuda, train_program, params_dirname):
train_reader = paddle.batch(
paddle.reader.shuffle(
paddle.dataset.imdb.train(word_dict), buf_size=25000),
batch_size=BATCH_SIZE)
batch_size=BATCH_SIZE,
drop_last=False)
trainer.train(
num_epochs=1,
......
python/paddle/fluid/tests/book_memory_optimization/test_memopt_fit_a_line.py
浏览文件 @ 29bf727e
......@@ -56,7 +56,7 @@ BATCH_SIZE = 200
# fix the order of training data
train_reader = paddle.batch(
paddle.dataset.uci_housing.train(), batch_size=BATCH_SIZE)
paddle.dataset.uci_housing.train(), batch_size=BATCH_SIZE, drop_last=False)
# train_reader = paddle.batch(
# paddle.reader.shuffle(
......
python/paddle/trainer_config_helpers/attrs.py
浏览文件 @ 29bf727e
......@@ -240,14 +240,15 @@ class ExtraLayerAttribute(object):
:type error_clipping_threshold: float
:param drop_rate: Dropout rate. Dropout will create a mask on layer output.
The dropout rate is the zero rate of this mask. The
details of what dropout is please refer to `here
<https://www.cs.toronto.edu/~hinton/absps/
JMLRdropout.pdf>`_.
details of what dropout is please refer to `JMLRdropout
<https://www.cs.toronto.edu/~hinton/absps/JMLRdropout.pdf
>`_.
:type drop_rate: float
:param device: device ID of layer. device=-1, use CPU. device>=0, use GPU.
The details allocation in parallel_nn please refer to `here
<http://www.paddlepaddle.org/doc/ui/cmd_argument/
use_case.html#case-2-specify-layers-in-different-devices>`_.
The details allocation in parallel_nn please refer to `use_case
<https://github.com/PaddlePaddle/Paddle/blob/develop/doc/v2
/howto/cmd_parameter/use_case_en.md#case-2-specify-layers-in
-different-devices>`_.
:type device: int
"""
......
python/paddle/trainer_config_helpers/layers.py
浏览文件 @ 29bf727e
......@@ -2556,7 +2556,7 @@ def img_conv_layer(input,
the output will be obtained by concatenating the two results.
The details of grouped convolution, please refer to:
`ImageNet Classification with Deep Convolutional Neural Networks
`ImageNet Classification With Deep Convolutional Neural Networks
<http://www.cs.toronto.edu/~kriz/imagenet_classification_with_deep_convolutional.pdf>`_
The example usage is:
......@@ -5678,8 +5678,8 @@ def warp_ctc_layer(input,
<https://github.com/baidu-research/warp-ctc>`_ library, which is used in
`Deep Speech 2: End-toEnd Speech Recognition in English and Mandarin
<https://arxiv.org/pdf/1512.02595v1.pdf>`_, to compute Connectionist Temporal
Classification (CTC) loss. Besides, another `warp-ctc
<https://github.com/gangliao/warp-ctc>`_ repository, which is forked from
Classification (CTC) loss. Besides, another `warp-ctc repository
<https://github.com/gangliao/warp-ctc>`_ , which is forked from
the official one, is maintained to enable more compiling options. During the
building process, PaddlePaddle will clone the source codes, build and
install it to :code:`third_party/install/warpctc` directory.
......
python/paddle/v2/minibatch.py
浏览文件 @ 29bf727e
......@@ -15,7 +15,7 @@
__all__ = ['batch']
def batch(reader, batch_size, drop_last=False):
def batch(reader, batch_size, drop_last=True):
"""
Create a batched reader.
......
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