diff --git a/doc/v2/api/config/evaluators.rst b/doc/v2/api/config/evaluators.rst index 9ac972fb193a2fb525edc507f7ba1303d2c8eabe..458d892e825a7a9bbe7843ad5c508bd5a31f5f0f 100644 --- a/doc/v2/api/config/evaluators.rst +++ b/doc/v2/api/config/evaluators.rst @@ -101,7 +101,7 @@ value_printer :noindex: Detection -===== +========== detection_map ------------- diff --git a/doc/v2/api/config/layer.rst b/doc/v2/api/config/layer.rst index 1a6496968cae1fef88142ba9ca3f9e63a81b196d..5a0cfadfce84df41defdf518b7c3a6222d5b30a1 100644 --- a/doc/v2/api/config/layer.rst +++ b/doc/v2/api/config/layer.rst @@ -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: diff --git a/doc/v2/api/index_en.rst b/doc/v2/api/index_en.rst index b11cd449affd1dcd9d3f42492961469331350942..70c5c524aaf0a9ae003bf4340c3f268c225d4419 100644 --- a/doc/v2/api/index_en.rst +++ b/doc/v2/api/index_en.rst @@ -8,4 +8,3 @@ API model_configs.rst data.rst run_logic.rst - fluid/index.rst diff --git a/doc/v2/build_and_install/pip_install_cn.rst b/doc/v2/build_and_install/pip_install_cn.rst index 853bdb21bbcf07ae1742d2196dbcfe4668828b7b..095da19cd41d29bfa72ab23abd24bec45f925a86 100644 --- a/doc/v2/build_and_install/pip_install_cn.rst +++ b/doc/v2/build_and_install/pip_install_cn.rst @@ -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 `__", "`paddlepaddle-latest-cp27-cp27m-linux_x86_64.whl `_" "cuda8.0_cudnn5_avx_mkl", "`paddlepaddle_gpu-latest-cp27-cp27mu-linux_x86_64.whl `__", "`paddlepaddle_gpu-latest-cp27-cp27m-linux_x86_64.whl `__" "cuda8.0_cudnn7_avx_mkl", "`paddlepaddle_gpu-latest-cp27-cp27mu-linux_x86_64.whl `__", "`paddlepaddle_gpu-latest-cp27-cp27m-linux_x86_64.whl `__" + "cuda9.0_cudnn7_avx_mkl", "`paddlepaddle_gpu-latest-cp27-cp27mu-linux_x86_64.whl `__", "`paddlepaddle_gpu-latest-cp27-cp27m-linux_x86_64.whl `__" .. _pip_dependency: diff --git a/doc/v2/build_and_install/pip_install_en.rst b/doc/v2/build_and_install/pip_install_en.rst index fecf6d3712feac3265100a6121901ba784f7d5cc..8406e4aa1fbb953c3b615b10d1bcb2c45974dde0 100644 --- a/doc/v2/build_and_install/pip_install_en.rst +++ b/doc/v2/build_and_install/pip_install_en.rst @@ -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 `__", "`paddlepaddle-latest-cp27-cp27m-linux_x86_64.whl `__" "cuda8.0_cudnn5_avx_mkl", "`paddlepaddle_gpu-latest-cp27-cp27mu-linux_x86_64.whl `__", "`paddlepaddle_gpu-latest-cp27-cp27m-linux_x86_64.whl `__" "cuda8.0_cudnn7_avx_mkl", "`paddlepaddle_gpu-latest-cp27-cp27mu-linux_x86_64.whl `__", "`paddlepaddle_gpu-latest-cp27-cp27m-linux_x86_64.whl `__" + "cuda9.0_cudnn7_avx_mkl", "`paddlepaddle_gpu-latest-cp27-cp27mu-linux_x86_64.whl `__", "`paddlepaddle_gpu-latest-cp27-cp27m-linux_x86_64.whl `__" .. _pip_dependency: diff --git a/paddle/fluid/operators/conv_mkldnn_op.cc b/paddle/fluid/operators/conv_mkldnn_op.cc index 63d371310d2a26a1460e527fc51923dfd6e0b8bc..6b06913d1c83f4534238ac3dd22ac4035c0f0fbf 100644 --- a/paddle/fluid/operators/conv_mkldnn_op.cc +++ b/paddle/fluid/operators/conv_mkldnn_op.cc @@ -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 class ConvMKLDNNOpKernel : public paddle::framework::OpKernel { public: @@ -25,6 +36,10 @@ class ConvMKLDNNOpKernel : public paddle::framework::OpKernel { 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(); const auto& mkldnn_engine = dev_ctx.GetEngine(); @@ -33,10 +48,12 @@ class ConvMKLDNNOpKernel : public paddle::framework::OpKernel { auto* filter = ctx.Input("Filter"); auto* output = ctx.Output("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 strides = ctx.Attr>("strides"); std::vector paddings = ctx.Attr>("paddings"); @@ -63,60 +80,86 @@ class ConvMKLDNNOpKernel : public paddle::framework::OpKernel { paddle::framework::vectorize2int(filter->dims()); std::vector 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(const_cast(input_data))); - auto weights_memory = - mkldnn::memory({weights_md, mkldnn_engine}, - reinterpret_cast(const_cast(filter_data))); - auto dst_memory = mkldnn::memory({dst_md, mkldnn_engine}, output_data); - - std::shared_ptr 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_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 pipeline{conv_prim}; - mkldnn::stream(mkldnn::stream::kind::eager).submit(pipeline).wait(); + std::vector 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 - ConvFwdPrimitiveDesc(const mkldnn::memory::desc& src, - const mkldnn::memory::desc& weights, - const mkldnn::memory::desc& dst, - const std::vector& strides, - const std::vector& 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( - p_conv_pd); + std::unique_ptr ConvFwdPrimitiveDesc( + const memory::desc& src, const memory::desc& weights, + const memory::desc& dst, const std::vector& strides, + const std::vector& 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(p_conv_pd); } }; @@ -139,6 +182,19 @@ class ConvMKLDNNGradOpKernel : public paddle::framework::OpKernel { Tensor* input_grad = ctx.Output(framework::GradVarName("Input")); Tensor* filter_grad = ctx.Output(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 { paddle::framework::vectorize2int(filter->dims()); std::vector 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(const_cast(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( - dev_ctx.GetBlob(key_conv_pd)); + auto conv_pd = std::static_pointer_cast( + 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(filter_grad_data)); - auto src_memory = - mkldnn::memory({src_md, mkldnn_engine}, - reinterpret_cast(const_cast(input_data))); + memory(conv_bwd_weights_pd.diff_weights_primitive_desc(), + reinterpret_cast(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 pipeline{conv_bwd_weights_prim}; - mkldnn::stream(mkldnn::stream::kind::eager).submit(pipeline).wait(); + std::vector 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(const_cast(input_grad_data))); - auto weights_memory = - mkldnn::memory({weights_md, mkldnn_engine}, - reinterpret_cast(const_cast(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(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 pipeline{conv_bwd_data_prim}; - mkldnn::stream(mkldnn::stream::kind::eager).submit(pipeline).wait(); + // push primitive and execute it + std::vector 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& strides, - const std::vector& 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& strides, - const std::vector& 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 diff --git a/paddle/fluid/operators/conv_op.cc b/paddle/fluid/operators/conv_op.cc index 27f1313116aad99d34fa8f1d3d6a1e7aced4d394..37153d58439a90190eb2ad82d5dcc145e22dfa48 100644 --- a/paddle/fluid/operators/conv_op.cc +++ b/paddle/fluid/operators/conv_op.cc @@ -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("data_format"); // TODO(pzelazko-intel): enable MKLDNN layout when it's ready + std::string data_format = ctx.Attr("data_format"); framework::DataLayout layout = framework::StringToDataLayout(data_format); #ifdef PADDLE_WITH_CUDA diff --git a/paddle/fluid/operators/gen_nccl_id_op.cc b/paddle/fluid/operators/gen_nccl_id_op.cc index 111e58844c83806af4ebe0aa9e2126a9ddec1d8a..f824eee4e7d1ef19c9a38fd5d3369265f9c549a0 100644 --- a/paddle/fluid/operators/gen_nccl_id_op.cc +++ b/paddle/fluid/operators/gen_nccl_id_op.cc @@ -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..."; } diff --git a/python/paddle/batch.py b/python/paddle/batch.py index d48c54fcbb66487617b1946bc69724870c8f879c..3c6a53db3c2287e8ef5931a06ca5dad455665ee0 100644 --- a/python/paddle/batch.py +++ b/python/paddle/batch.py @@ -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. diff --git a/python/paddle/fluid/layers/nn.py b/python/paddle/fluid/layers/nn.py index 047c3aa2b7acd0e19b05dfae13c28d2f16007e10..3194b72da68ca0567793de29fcac02995136b4c1 100644 --- a/python/paddle/fluid/layers/nn.py +++ b/python/paddle/fluid/layers/nn.py @@ -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) diff --git a/python/paddle/fluid/tests/book/high-level-api/image_classification/test_image_classification_resnet.py b/python/paddle/fluid/tests/book/high-level-api/image_classification/test_image_classification_resnet.py index 2df3da9cca7042222317de626460909f018cb107..8e222d26907e8fe697b596a67e62cc9df84afe0e 100644 --- a/python/paddle/fluid/tests/book/high-level-api/image_classification/test_image_classification_resnet.py +++ b/python/paddle/fluid/tests/book/high-level-api/image_classification/test_image_classification_resnet.py @@ -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): diff --git a/python/paddle/fluid/tests/book/high-level-api/image_classification/test_image_classification_vgg.py b/python/paddle/fluid/tests/book/high-level-api/image_classification/test_image_classification_vgg.py index 224cca417e717bbcc54b58be6ac0219be207dea3..dbc7bc06c93157f271c79e85b6925468e861e57f 100644 --- a/python/paddle/fluid/tests/book/high-level-api/image_classification/test_image_classification_vgg.py +++ b/python/paddle/fluid/tests/book/high-level-api/image_classification/test_image_classification_vgg.py @@ -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): diff --git a/python/paddle/fluid/tests/book/high-level-api/understand_sentiment/test_understand_sentiment_stacked_lstm.py b/python/paddle/fluid/tests/book/high-level-api/understand_sentiment/test_understand_sentiment_stacked_lstm.py index 113dda88ca974c9e6241f127091bd96fb2af4a70..8c74be0f08855c20f5aa3ecd75622a51e94a0304 100644 --- a/python/paddle/fluid/tests/book/high-level-api/understand_sentiment/test_understand_sentiment_stacked_lstm.py +++ b/python/paddle/fluid/tests/book/high-level-api/understand_sentiment/test_understand_sentiment_stacked_lstm.py @@ -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, diff --git a/python/paddle/fluid/tests/book_memory_optimization/test_memopt_fit_a_line.py b/python/paddle/fluid/tests/book_memory_optimization/test_memopt_fit_a_line.py index 8818cf96fa8f08036f9e23aae786f67b5614b2b9..be347cd5315668dde0454d7959dbf9bcfa465b5f 100644 --- a/python/paddle/fluid/tests/book_memory_optimization/test_memopt_fit_a_line.py +++ b/python/paddle/fluid/tests/book_memory_optimization/test_memopt_fit_a_line.py @@ -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( diff --git a/python/paddle/trainer_config_helpers/attrs.py b/python/paddle/trainer_config_helpers/attrs.py index e6f87ce61b1d16d4f98f111626776aa52c2ec35b..4e3beaf639bad9fed2862a5477095b66ef4b9aee 100644 --- a/python/paddle/trainer_config_helpers/attrs.py +++ b/python/paddle/trainer_config_helpers/attrs.py @@ -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 - `_. + details of what dropout is please refer to `JMLRdropout + `_. :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 - `_. + The details allocation in parallel_nn please refer to `use_case + `_. :type device: int """ diff --git a/python/paddle/trainer_config_helpers/layers.py b/python/paddle/trainer_config_helpers/layers.py index ebc31b23e0f5504b4bebccabe996b054c7fbce3b..e6a03759ef431086390e217eabcdff47e610346c 100644 --- a/python/paddle/trainer_config_helpers/layers.py +++ b/python/paddle/trainer_config_helpers/layers.py @@ -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 `_ The example usage is: @@ -5678,8 +5678,8 @@ def warp_ctc_layer(input, `_ library, which is used in `Deep Speech 2: End-toEnd Speech Recognition in English and Mandarin `_, to compute Connectionist Temporal - Classification (CTC) loss. Besides, another `warp-ctc - `_ repository, which is forked from + Classification (CTC) loss. Besides, another `warp-ctc repository + `_ , 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. diff --git a/python/paddle/v2/minibatch.py b/python/paddle/v2/minibatch.py index d48c54fcbb66487617b1946bc69724870c8f879c..3c6a53db3c2287e8ef5931a06ca5dad455665ee0 100644 --- a/python/paddle/v2/minibatch.py +++ b/python/paddle/v2/minibatch.py @@ -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.