From ee48679c0613843855807f6f71b972a01299570e Mon Sep 17 00:00:00 2001 From: SunAhong1993 Date: Wed, 9 Dec 2020 11:59:32 +0800 Subject: [PATCH] add tf static --- x2paddle/convert.py | 15 +- .../dygraph/tf2paddle/tf_op_mapper.py | 22 +- .../static/caffe2paddle/caffe_op_mapper.py | 3 +- .../__pycache__/__init__.cpython-37.pyc | Bin 0 -> 175 bytes .../__pycache__/tf_op_mapper.cpython-37.pyc | Bin 0 -> 38311 bytes .../static/tf2paddle/tf_op_mapper.py | 589 +++++++++--------- .../__pycache__/__init__.cpython-37.pyc | Bin 0 -> 158 bytes .../__pycache__/optimizer.cpython-37.pyc | Bin 0 -> 1897 bytes .../__pycache__/pass_.cpython-37.pyc | Bin 0 -> 746 bytes .../__pycache__/pass_manager.cpython-37.pyc | Bin 0 -> 1106 bytes .../pattern_matcher.cpython-37.pyc | Bin 0 -> 7474 bytes .../code_optimizer/hierachical_tree.py | 10 +- .../code_optimizer/layer_code_generator.py | 32 +- .../code_optimizer/subgraphs_union.py | 16 +- .../__pycache__/__init__.cpython-37.pyc | Bin 0 -> 170 bytes .../__pycache__/__init__.cpython-37.pyc | Bin 0 -> 329 bytes .../transpose_eliminate_pass.cpython-37.pyc | Bin 0 -> 962 bytes .../transpose_elimination.cpython-37.pyc | Bin 0 -> 6076 bytes .../optimizer/elimination/static/__init__.py | 16 + .../__pycache__/__init__.cpython-37.pyc | Bin 0 -> 326 bytes .../transpose_eliminate_pass.cpython-37.pyc | Bin 0 -> 963 bytes .../transpose_elimination.cpython-37.pyc | Bin 0 -> 6132 bytes .../static/transpose_eliminate_pass.py | 33 + .../static/transpose_elimination.py | 291 +++++++++ .../__pycache__/__init__.cpython-37.pyc | Bin 0 -> 165 bytes .../__pycache__/__init__.cpython-37.pyc | Bin 0 -> 1715 bytes .../adaptive_pool2d_fuse_pass.cpython-37.pyc | Bin 0 -> 1007 bytes .../adaptive_pool2d_fuser.cpython-37.pyc | Bin 0 -> 4662 bytes .../batchnorm2d_fuse_pass.cpython-37.pyc | Bin 0 -> 983 bytes .../batchnorm2d_fuser.cpython-37.pyc | Bin 0 -> 4166 bytes .../bn_scale_fuse_pass.cpython-37.pyc | Bin 0 -> 958 bytes .../__pycache__/bn_scale_fuser.cpython-37.pyc | Bin 0 -> 2948 bytes .../constant_fuse_pass.cpython-37.pyc | Bin 0 -> 962 bytes .../__pycache__/constant_fuser.cpython-37.pyc | Bin 0 -> 2187 bytes .../conv2d_add_fuse_pass.cpython-37.pyc | Bin 0 -> 955 bytes .../conv2d_add_fuser.cpython-37.pyc | Bin 0 -> 3710 bytes .../dropout_fuse_pass.cpython-37.pyc | Bin 0 -> 955 bytes .../__pycache__/dropout_fuser.cpython-37.pyc | Bin 0 -> 2168 bytes .../__pycache__/fc_fuse_pass.cpython-37.pyc | Bin 0 -> 920 bytes .../__pycache__/fc_fuser.cpython-37.pyc | Bin 0 -> 4197 bytes ...erpolate_bilinear_fuse_pass.cpython-37.pyc | Bin 0 -> 1042 bytes .../interpolate_bilinear_fuser.cpython-37.pyc | Bin 0 -> 6426 bytes .../prelu_fuse_pass.cpython-37.pyc | Bin 0 -> 929 bytes .../__pycache__/prelu_fuser.cpython-37.pyc | Bin 0 -> 4020 bytes .../reshape_fuse_pass.cpython-37.pyc | Bin 0 -> 955 bytes .../__pycache__/reshape_fuser.cpython-37.pyc | Bin 0 -> 2312 bytes .../tf_batchnorm_fuse_pass.cpython-37.pyc | Bin 0 -> 967 bytes .../tf_batchnorm_fuser.cpython-37.pyc | Bin 0 -> 4811 bytes .../trace_fc_fuse_pass.cpython-37.pyc | Bin 0 -> 907 bytes .../__pycache__/trace_fc_fuser.cpython-37.pyc | Bin 0 -> 3668 bytes .../fusion/dygraph/conv2d_add_fuser.py | 4 - x2paddle/optimizer/fusion/static/__init__.py | 9 +- .../__pycache__/__init__.cpython-37.pyc | Bin 0 -> 672 bytes .../bn_scale_fuse_pass.cpython-37.pyc | Bin 0 -> 955 bytes .../__pycache__/bn_scale_fuser.cpython-37.pyc | Bin 0 -> 3692 bytes .../conv2d_add_fuse_pass.cpython-37.pyc | Bin 0 -> 948 bytes .../conv2d_add_fuser.cpython-37.pyc | Bin 0 -> 4323 bytes .../prelu_fuse_pass.cpython-37.pyc | Bin 0 -> 922 bytes .../__pycache__/prelu_fuser.cpython-37.pyc | Bin 0 -> 4088 bytes ...static_conv2d_add_fuse_pass.cpython-37.pyc | Bin 0 -> 955 bytes .../tf_batchnorm_fuse_pass.cpython-37.pyc | Bin 0 -> 960 bytes .../tf_batchnorm_fuser.cpython-37.pyc | Bin 0 -> 4768 bytes .../optimizer/fusion/static/bn_scale_fuser.py | 1 - .../fusion/static/conv2d_add_fuse_pass.py | 33 + .../fusion/static/conv2d_add_fuser.py | 121 ++++ .../fusion/static/prelu_fuse_pass.py | 33 + .../optimizer/fusion/static/prelu_fuser.py | 139 +++++ .../fusion/static/tf_batchnorm_fuse_pass.py | 33 + .../fusion/static/tf_batchnorm_fuser.py | 227 +++++++ x2paddle/optimizer/optimizer.py | 24 +- .../__pycache__/__init__.cpython-37.pyc | Bin 0 -> 169 bytes .../__pycache__/batch_norm.cpython-37.pyc | Bin 0 -> 3649 bytes .../__pycache__/bias.cpython-37.pyc | Bin 0 -> 1798 bytes .../__pycache__/prelu.cpython-37.pyc | Bin 0 -> 2861 bytes .../__pycache__/transpose.cpython-37.pyc | Bin 0 -> 5628 bytes 75 files changed, 1278 insertions(+), 373 deletions(-) create mode 100644 x2paddle/op_mapper/static/tf2paddle/__pycache__/__init__.cpython-37.pyc create mode 100644 x2paddle/op_mapper/static/tf2paddle/__pycache__/tf_op_mapper.cpython-37.pyc create mode 100644 x2paddle/optimizer/__pycache__/__init__.cpython-37.pyc create mode 100644 x2paddle/optimizer/__pycache__/optimizer.cpython-37.pyc create mode 100644 x2paddle/optimizer/__pycache__/pass_.cpython-37.pyc create mode 100644 x2paddle/optimizer/__pycache__/pass_manager.cpython-37.pyc create mode 100644 x2paddle/optimizer/__pycache__/pattern_matcher.cpython-37.pyc create mode 100644 x2paddle/optimizer/elimination/__pycache__/__init__.cpython-37.pyc create mode 100644 x2paddle/optimizer/elimination/dygraph/__pycache__/__init__.cpython-37.pyc create mode 100644 x2paddle/optimizer/elimination/dygraph/__pycache__/transpose_eliminate_pass.cpython-37.pyc create mode 100644 x2paddle/optimizer/elimination/dygraph/__pycache__/transpose_elimination.cpython-37.pyc create mode 100644 x2paddle/optimizer/elimination/static/__init__.py create mode 100644 x2paddle/optimizer/elimination/static/__pycache__/__init__.cpython-37.pyc create mode 100644 x2paddle/optimizer/elimination/static/__pycache__/transpose_eliminate_pass.cpython-37.pyc create mode 100644 x2paddle/optimizer/elimination/static/__pycache__/transpose_elimination.cpython-37.pyc create mode 100644 x2paddle/optimizer/elimination/static/transpose_eliminate_pass.py create mode 100644 x2paddle/optimizer/elimination/static/transpose_elimination.py create mode 100644 x2paddle/optimizer/fusion/__pycache__/__init__.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/dygraph/__pycache__/__init__.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/dygraph/__pycache__/adaptive_pool2d_fuse_pass.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/dygraph/__pycache__/adaptive_pool2d_fuser.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/dygraph/__pycache__/batchnorm2d_fuse_pass.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/dygraph/__pycache__/batchnorm2d_fuser.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/dygraph/__pycache__/bn_scale_fuse_pass.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/dygraph/__pycache__/bn_scale_fuser.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/dygraph/__pycache__/constant_fuse_pass.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/dygraph/__pycache__/constant_fuser.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/dygraph/__pycache__/conv2d_add_fuse_pass.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/dygraph/__pycache__/conv2d_add_fuser.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/dygraph/__pycache__/dropout_fuse_pass.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/dygraph/__pycache__/dropout_fuser.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/dygraph/__pycache__/fc_fuse_pass.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/dygraph/__pycache__/fc_fuser.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/dygraph/__pycache__/interpolate_bilinear_fuse_pass.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/dygraph/__pycache__/interpolate_bilinear_fuser.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/dygraph/__pycache__/prelu_fuse_pass.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/dygraph/__pycache__/prelu_fuser.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/dygraph/__pycache__/reshape_fuse_pass.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/dygraph/__pycache__/reshape_fuser.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/dygraph/__pycache__/tf_batchnorm_fuse_pass.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/dygraph/__pycache__/tf_batchnorm_fuser.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/dygraph/__pycache__/trace_fc_fuse_pass.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/dygraph/__pycache__/trace_fc_fuser.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/static/__pycache__/__init__.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/static/__pycache__/bn_scale_fuse_pass.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/static/__pycache__/bn_scale_fuser.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/static/__pycache__/conv2d_add_fuse_pass.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/static/__pycache__/conv2d_add_fuser.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/static/__pycache__/prelu_fuse_pass.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/static/__pycache__/prelu_fuser.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/static/__pycache__/static_conv2d_add_fuse_pass.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/static/__pycache__/tf_batchnorm_fuse_pass.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/static/__pycache__/tf_batchnorm_fuser.cpython-37.pyc create mode 100644 x2paddle/optimizer/fusion/static/conv2d_add_fuse_pass.py create mode 100644 x2paddle/optimizer/fusion/static/conv2d_add_fuser.py create mode 100644 x2paddle/optimizer/fusion/static/prelu_fuse_pass.py create mode 100644 x2paddle/optimizer/fusion/static/prelu_fuser.py create mode 100644 x2paddle/optimizer/fusion/static/tf_batchnorm_fuse_pass.py create mode 100644 x2paddle/optimizer/fusion/static/tf_batchnorm_fuser.py create mode 100644 x2paddle/optimizer/tensorflow/__pycache__/__init__.cpython-37.pyc create mode 100644 x2paddle/optimizer/tensorflow/__pycache__/batch_norm.cpython-37.pyc create mode 100644 x2paddle/optimizer/tensorflow/__pycache__/bias.cpython-37.pyc create mode 100644 x2paddle/optimizer/tensorflow/__pycache__/prelu.cpython-37.pyc create mode 100644 x2paddle/optimizer/tensorflow/__pycache__/transpose.cpython-37.pyc diff --git a/x2paddle/convert.py b/x2paddle/convert.py index 7b38f27..3de2951 100644 --- a/x2paddle/convert.py +++ b/x2paddle/convert.py @@ -132,18 +132,9 @@ def tf2paddle(model_path, graph_opt = GraphOptimizer(source_frame="tf", paddle_type=paddle_type) graph_opt.optimize(mapper.paddle_graph) else: - from x2paddle.optimizer.tensorflow.bias import BiasOpt - from x2paddle.optimizer.tensorflow.transpose import TransposeOpt - from x2paddle.optimizer.tensorflow.batch_norm import BatchNormOpt - from x2paddle.optimizer.tensorflow.prelu import PReLUOpt - bias_opt = BiasOpt() - transpose_opt = TransposeOpt() - batch_norm_opt = BatchNormOpt() - prelu_opt = PReLUOpt() - bias_opt.run(mapper.paddle_graph) - batch_norm_opt.run(mapper.paddle_graph) - prelu_opt.run(mapper.paddle_graph) - transpose_opt.run(mapper.paddle_graph) + from x2paddle.optimizer.optimizer import GraphOptimizer + graph_opt = GraphOptimizer(source_frame="tf", paddle_type=paddle_type) + graph_opt.optimize(mapper.paddle_graph) mapper.paddle_graph.gen_model(save_dir) diff --git a/x2paddle/op_mapper/dygraph/tf2paddle/tf_op_mapper.py b/x2paddle/op_mapper/dygraph/tf2paddle/tf_op_mapper.py index 336f3e1..34aee01 100644 --- a/x2paddle/op_mapper/dygraph/tf2paddle/tf_op_mapper.py +++ b/x2paddle/op_mapper/dygraph/tf2paddle/tf_op_mapper.py @@ -284,7 +284,6 @@ class TFOpMapper(OpMapper): inputs["shape"] = dims.name layer_attrs["dtype"] = string(input_value.dtype) layer_attrs["fill_value"] = input_value.value - self.paddle_graph.add_layer( "paddle.full", @@ -578,6 +577,9 @@ class TFOpMapper(OpMapper): inputs={"x": node.name}, outputs=[node.name], perm=[0, 2, 3, 1]) + + def FusedBatchNormV3(self, node): + self.FusedBatchNorm(node) def Mean(self, node): input = self.graph.get_input_node(node, 0) @@ -930,6 +932,23 @@ class TFOpMapper(OpMapper): outputs=[node.name], axis=axis) + def Concat(self, node): + inputs_list = list() + for i in range(1, len(node.inputs)): + inputs_list.append(self.graph.get_input_node(node, i)) + axis = self.graph.get_input_node(node, 0) + assert axis.layer_type == "Const", "axis for ConcatV2 must be type Const" + axis = axis.value + if axis < 0: + axis += len(inputs_list[0].out_shapes[0]) + + input_names = [i.name for i in inputs_list] + self.paddle_graph.add_layer( + kernel="paddle.concat", + inputs={"x": input_names}, + outputs=[node.name], + axis=axis) + def AddN(self, node): inputs_list = list() for i in range(len(node.inputs) - 1): @@ -1400,6 +1419,7 @@ class TFOpMapper(OpMapper): inputs = {"x": x.name, "y": y.name} x_shape = x.out_shapes[0] y_shape = y.out_shapes[0] + # TODO(syf) layer_id = self.paddle_graph.add_layer( "fluid.layers.elementwise_sub", inputs=inputs, outputs=[node.name]) self.paddle_graph.layers[layer_id].input_shapes = {"x": x_shape, "y": y_shape} diff --git a/x2paddle/op_mapper/static/caffe2paddle/caffe_op_mapper.py b/x2paddle/op_mapper/static/caffe2paddle/caffe_op_mapper.py index e99e941..23dcc18 100644 --- a/x2paddle/op_mapper/static/caffe2paddle/caffe_op_mapper.py +++ b/x2paddle/op_mapper/static/caffe2paddle/caffe_op_mapper.py @@ -89,6 +89,7 @@ class CaffeOpMapper(OpMapper): input_shape.append(last_node.output_shape[idx]) node.input_shape = input_shape + func_name = 'shape_' + node.layer_type.lower() if is_fluid_op: node.output_shape = getattr(caffe_shape, func_name)(node.layer, @@ -974,4 +975,4 @@ class CaffeOpMapper(OpMapper): kernel=op_info, inputs={"x": self.get_input_name(input)}, outputs=[node.layer_name]) - + \ No newline at end of file diff --git a/x2paddle/op_mapper/static/tf2paddle/__pycache__/__init__.cpython-37.pyc b/x2paddle/op_mapper/static/tf2paddle/__pycache__/__init__.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..a240985a6ff222bb2c8efbf8c0dc5015589629f4 GIT binary patch literal 175 zcmZ?b<>g`k0v^u`@gVv!h=2h`Aj1KOi&=m~3PUi1CZpdqnB|Nj2(o}r;)0)OTEzh2$(jYQ(xyzu@qIC%qp*87pD zgi1COE6Lg9N@_NharDHXvnss@GOXH!=gSAelZsI(76Ow$#Y%iF# zm6^E|OTu3oKX&;re%4VWD!Gy9C3?wAsr6JZWha;O=01gEFLfx<8sRaU*p*P}*W=Yg ziC#)&b|h?hzAJG)xg){1(A4ZqH~YkT^@(*ei9%mw;v}n6X(u<|Hdku4Q*PB(>eadS zYRj&hPTI0fJj}N{Ro-`uGc{A6WqAGRggJr|UV8MFm6m;rwc0vcYt7eM3;SPp%i9j< zKBFjA_pfeQJJ6A#xCJGy$QayW*)_W{cZ)smJ+kMk{?qBsS;ttYx1`c5P?uFkl1P@5 zMzWY1NscGYFdy+rv1q6LxYV)x-FWV^sDeE+x+=*VDbUO6|lo?qO_t ziJgh{OfRF-%L(%~m0r)-30!6NVC2@by-Y8Q=eb_Cms`+dxPXH3_x-y}0|z~noH-0t zy3w+o3`TO%$;{Oo&8fVTZ?vjbqg!`M%XPC=Z|aL2hPI(Fq^!(k*dGC1a5+?nHT?~kwZ&w4YOT2IU- zRbn=!k_$7Ayv7MTu7;5T#w*4ir?6C!?PK+ z4gcf#->xRq4z*M5QoGe2wO3uC_Ngn?q`K<4k=g81iTR|uT3z#`F`HA@s_U?(^6GkZ z1CEBeQBC1kP&cX9;8;{QtJmUKQq$@d9Ls9IdL52K>VP_kV@2Jnegel~b(?xUjw9+P z)lcC#s%}?5jpLX)r0&3Ro4Qloh2yw-gSs2X?dl$N7{>{9MBR(y4s}$$5yzeCK6O8i zyVL{fO*rmWZ&nZDxJS*Xhj84hj;Xicc!hdc9mjE>I-%Z*B!|@vR4pqhRTJ=sgT2u`jr__>K#_=ZA zR4X{XMzvHM$D370oyPIC%2XDP)5=z>INqYpsK;^Kuh!IA9ABrp>IobVsGeHK@t}H_ z`dJ)rRqs}R2gjdK=hS%|Z&OdI_u%+?^G{>xoZV=*YRzf0 z-dyeO-t^IJ-J<)fcFOAR9hlE*EUdH}YO|;9d3&X{=9g1n>vYR*Ii_{mw7VtuN-hRB z#d^G9E&3VlT5GX;#o)G}woY@^>RvhUakE}qK3ipZg+!N7^`x0Akit%EG=oRc}u zv>?5y&L2GBWKN(fj={WawNrRA8>mY?%KGuTIq&3-pJt7IV}N1&#;C>_psViQ`Q~av zO*d<2>!vkbZ`N1pEgPeT?ya8cR@|0jA{r~JE5UW6CD)_w^Jd+$s&!P>?2fy)7R-7L zB*(qGEq2#lthKr$?#)))cC&dCtcx0UpRcSo?MA10w(FH5LtJ%x;FY3puC$d?YPJ^| zbG2r*)>2a=7+|HGERvH4oczP}TJuQbjAI~~Y0uPJPWt%jDJTD+?jcTe*Epn>buYW* zl;5cPqObDqm1E4bZG9`N59tS2n~ovlFIC9dgJ@9k&ALH%x0KHjkToK8Bc-#x3c^jw z)=QjEZKN(GlAvR~L^a(@E@g1c_7WSpO9|^{JFh_R&LwT*Jg8i?U>CvOq*PjEE~If? zvdibe8LS=aC6=;FLq`&i9RY2``9`IezLZ=VR@vpWc_MNr*LNo`cTyW8d?%SWed3YC z+O@qDO53YTTN>IJk zml8)nkIyGb!IKH(De<^=(cUdN_MA_m#&6k|5o%QIi=Ho1S5+_!Q6YWMy=`x@|I#`zTTkK5Pk z^Yucn(95$e7n7&QA7RTXyh6#Pjq9Xs*SocsHg3RmfmbR4#5p|E-GlkAHzym`B%pfI zTJ3b&rd?N)XY2N~10I3Ak(09L!CC;=O)R#mtUad)hOx6)wa<3ISXk{Wl5B0s}YvP^*4?v2EXvrB$=B&WqmJ$phWt6K%WJM8&FZ-9Gum`W)Fo{t?E=MIFb_ z+KprZy?Ja5W1C!0fj0tQc|P^Q)alG|%>aV+EjFx13(alK)lI&R_NJ!yo5MJDGOQuS z$h=fM)NEMxTzjQ+r^RZ>Kjo6$8v;r)U9Dnm+12Xwp=KM))4CHU-aFlK^48YmT)TCq zZrWgkCa0&TyL*QY2RB$Py2S^R{1a}iJ6epqSy$6jH#%92eBCt3VLFCauaj4GvahCN zfG3slcCz|V1cy4eSf2w&=M?T;o2z$7=bVgc%-K#36IENWOfvACHS+%M>lt>yDK%Q1 zRlCaeS!8eZsZ*4h)w7r*bKC+RYP3{+&B@6dmQ!x8+QF-NePub>jtRcgB#r>jzPbXg z5@nOaMgewx1sK2HT3sOw)SZIe?zF2G;0$@sT7)w6(R8wpo5<^osD@devzur6IyiAu z8ouKUhmk%XM{cmn{9?_j*|zEA(KR_SEcTHSgbTKEUg2FCt2)5bnkqUOtKOV9XV8t= zVPg4C0LmJa(hardBpXga`o~ht2iWOUyW?cYZh*cPs@1t>4Fskd@Xh9f$n`V$Su8w} ztR!<_mG%EfDks-txL3x1C0WF;lEOKjq;Pj4Wq`S!z%!kLxyq8uNRA}C+Xld_zm~X; ze2L^&aA7BabCYWi023sah$ZiYY=-lVbT0+W1leRM0|^8;yq7tjvPK|(B$jfJ_w-p_ z&XOBOFGDo%JNAD7ZY$E-E*()yF0kxOTpD0cO0E61M2Fw zc??5gDkg7Z!nrY0|GK@My(DHUhZ0D)GQIqv#OWhKMaxS=^OX7VmLmLA8mzwp z(Jb>^x|jKd)cY~NLUFQsUIi06$u(fYgPEx9Ofk5;9F3H0E4E_<1J#+B?IqfE{K9rbZUYt zyH~xnwcKhy-kOv#nQYHb$`EP(>g`XgKQ`SRb0rkb*G;RMU7O0qX2{7ZVd}C{j9w>2 z^J%o->@tx_E8@)K7laCafRDJw!c#>RQ)^)=6HJat*fk|{3`O3H2VcO?nn04kXphN= zbIeHsdF)PvMm@x3`XkO2LhKm9jC#Z|YNTzTc%VC=bS}N>0LC*awVbkUwX+H%d`{F+ zz?&I6r!wbB(;iS+ksR_%hRVr&Kq+MxIUiJAK~tCa=7Qy81(k3tKnt2!;yXt49n4{o z&+#^tx)<^;W-Ns;V|VXcThJeX3j!x@HSASz;qBHW)`Re5(DBW8%a9*y_Ki2-{DvuC zWiY?mZ1&Bl9&e7kQ=D%!n^i%M-F;DDm=mEf;D@?X?sF(o$OkuSO(+q~X)w6tv+WzD zlsooM@1L5=$6zR^Q7_tUsy<(XR4fd&ThLTl0~LY6$x2}YU|_;n55dgIN+)A*Wj>2X z<_RP&HP&=DfULjI=TY!-veN8-i6_9EdKOTZX_w4I#YreqPDsmRJFVmCj+BsMjL6JBG# ziKo^Wk^~6lI5B`Qz$BC1@zC)5FZffy;rFND>$r#v_ZCxNXUGdln#JG)2Vgpm>a~yQ zNa;J~J8_oZFfJt^&u3xrDB`++m5*~`f+gqz@Fa&d51NWI@J46>_#!m~P6?|W^s@j; zn$^6JTW;F?T{l-z3C@sdMqa#8A>gR8KiQSO$(}q(9*48dMLl4mL&)~$F(fZ}3P?c` zsemT?YBIB~aXyJRD7Vj&NkLOd^A~x4cpy9o113zg4;1D{kzLlmUa>I}L5c!YMRr;< zU=Us*D7vz##s|`Z_b^8cW->?7+wGdw${qI~v zNiUKFbazK|)1Af|ZEcII^YhJm-)l}~wPQi4RIkq0=Aanc1kIj9xlYbi0|^f{j5}g; z3~=q#P_T-Ge>N{M)5rL3vgITfo#f+Aat=B&t@fcFq~ed2$TYITowD0$eJi%)%11cb zC)&q3*5Ex8)+{Q((v#(8;5Y_s9W1sCgq%!Db8#fBAgjp9g~;=<&^Z+#n)G@(WT}RQ zqHx{*k|n$6BYoQCp&Cq5so^qrhX8US{9?({Y)y49gcp@@mzMyl%3erAvz-A7SlxaEkY# zQ51qSH)KF71}uy)!U3;7iqZ3OKq&K;o4s~hqh z&bZxK2AFpa^zBJYyYCUhJYJ$d4d%RZ$ei#+}0Fh{YNcRd@;zNRq#?&_8 zqVW(e+8(--fRPp6I>2rfl*>Bdw}bH@>q_-;5Q?xu9 zHOi_Jv?zl8OUYYILg!JKmxfp2?4TO+^P1Sh_;$S0Y}nnh_|;XlIvq$y5G4_xMn5D| z-K(N1A82;sTAzMQaFIob?8Zg@>fo(GsP&gaNb9=LN`=ni%z-0+)tJ@%3cF3LVEV{^ zBWR(M0}FKq*z>ZaiZjs1{cAl4Sr{#&e~Ov3!0qI;QPqJ%60EpY+97=A(dEzk%i|<< z_Y6e!x}ly*KKFRNv9M@stRGj~Ihj+9nk8u8F+70O1Yp-dsb3_LfC`BGFWNVP3S}rU zpo&423j2BZ*Vy$IOuyPxoP|xeb5_Ps^m!qMO`uK`ePqTo{82Lm4JI9c#I18`Bu-pJ zfh81#WXAk5lYhkIA2azUO#Uenr|fz{ik86{bFcK!XyNC1<`uf+s~<3g0u%vp%4yhq zf^b2WT=Wp2IQ%D00EZZU&+-%`S(S&y8pms;bIyCqHLP*+w++NW@qgYE((Zo_a%+K`Ym=tF#x#^>;+ z0ZIFPkTmMc@_YlW9%_-Kh2FKxAbDk%q*Z!DLehqpMuen+0033vE=by_J!Wt7uf~O> zjrE2?BrOlpB!n6y=h>G<(!>E3ByAYwvQACXsPUlFs!P(;UT9jb5W+Kz8cJ%P5H$y` zsv+V8vIh2ieZ(zk5+sbaRUwT531cpGl}p4}`;n>CLFG~@^S4m&i&XFQTb6#Cb&6o}J3JHVWFjQY_oI2CysxrtXoUJf4d$qVIcPSs)jtT!V83JEM-kZU%zekGPNG_Qa(mT+Jc!48g$8WJK@z?(W>1%`s`5{w1-L25<)h8lhjaxh&Qq@^R5C_U0p zAa#MrzByC}@`)V9ogmmF0^K0xD5*uDV=PR7d_<*bh)|&qWB9dpsd%bp&n?2|(OlVz zX&;Q_HgU5LiiFzn2~-6Q`f95MjaGG~4p;4xcZF?Bg40Zbn3Jz}tVXll>aXh$@ECUS zV{CQQCB{6*d!lO*E?cZ@9w9-D~HE*`)Ua)mCS^-9i zt%Ax?2%YL4j^GRiaEy!L2rAcxH!g^AUo}j%Y1|6nvAbGj45n5vmTQvR6=GeZ_4q3Z z^@(OiR#F8QaEgX*<&x)CmYjjQ?hkPvXyG&t+9M#=iIHgGClB7vZm{S?vb$%~Mw~q8 zH-pM-s(hsxCQc#?a4S>?pUO=P{p7P@0R)oda;E)Hn-)p}_}KQu{a?zskahh;U{nAm zad`%UclXAJTFtYb>$W!WPm1hFuhzGlb+roDZ1}A{<`kCe^$v6gE0*stLO2!B1q6%q z^e78|2NEZPEe6=>!FTY=CSVx=qsV<9<=f|&#E@;s(CfN*1r0NQi7$MUNo>k852AT@ zaGECYR6`rZq`)r#UO{1#F7x}-#i`sHbA2g$*WF1ZZ% z*GmcWH^K7j>x2j@UTv^s2@tTp9NVJn6wKET2HxJ6h!H>T)1fW-+C5L7_c`df=tm4F zf?#BdgV_xxmW&g26Paf1thn;PVYYR)U*r2;WYf&fMZQq*ef#3Gjm<9NfAA(w_g6*9UlHDWRC3iqu>E?ku8AURlSc_u)b!&?Ca4GFXOJ0wE|=N-D^Swn{kyW21ejX2&sYu!*}A z3B`SpBm5#4o!By+G}KXBky;UU`e`7Aitf&Vsw~{>&;KXLzll>y6gfHhd8k;Sf6|UA zRM=<_H0qTui?3+1psflXD%>&aUInI$Kv(FbkJZ#7Vfc0w>>;fROySoG%o38dj>m&( z!mcKU)cbhn-LTaWMf#9TZrg`sYP&F(T7~1;RrOV%(4kY)v*_VYW0l$7g`Jn)7BqUn;;N+xQ-0snAbWFjE-(#zU^bG(d zt?EpJgqnYfd|^Ypwut=~zAdW881*YW*r3xprv4d1(~Za{7!jJn3E;jk91()ZcKNUq zc9-*u@gayZn@<@Dr7;M?mUZ|{(9H>+4`f}$XF_8eU?gCPJ{!GLwob3s>s{I=z+ZUO z$i1mqj6Q1gEmRGQ!!aDR#r_=60t!R=gNY1SI+4@<2Q$USDvzzBto@fcQsfXhQr;NB zeK0JixmhEl;Y+Xw@M>%{$QgJeL2re$z&dF>T;ih}8v;)SI7X_iPeuw+} z`uu>tW{#~F z6fY-WfUW$HuesY?fJ?+=dke|L-`QO=7vPSt6^1ZE7{W)w8ui8ou@6^#3*C<6ikxoU z*Ep6$*d9{V5F?Ca*qows`=d%o@rA%sxQMMIj}~u2ece%Y{Q|_BjQqY9`h0B%96U2y z3VknUgt4nU{^Z!96c<`g% zzEMBX?!4*bfq%_W*U}tKWOTm|&@`mE@W|^o{E;Wc@<*P+>Xsu<8m2dW^9<%IaMu}? zST@XiAabWf1x{ zQcKk}eK_mGJOZ>JNp#c7X zGqVqw^m?F)k$N0YmqT?Y*^9^w={NluLJ5JMhG@#u(_puP=xCtD*iuJR5*?E+^-tIy zmI(Dsc;`>O75xOMMxu8fHYR{qK?+_4+&Z{;Uw;btE^#}b zs)*rn?*Pxi6nqK8o`WgzCEOPFCA7Lqkky6uJ&k;7r)#2g&#+$0;*AI?c|2jm-5nBkLhZU>h~JHF`Ok-HPD;&4Mbco8 z{r-xiu{*S8^^Tca-Xt-RxQjx3j2TKqTlHroYD%A^`z1_r%*?r&tD>)v9xRK9Ui#HB z_BjS)JPFGwd@)V_;;)zkOnUx`w}t>Y+E=Cf`d98_BV!qDa6V}8t1~o%9C?hO5`-I| zTf0P%3A^;X3||_p(f>4t8XnnE-DzN;baD%38?OG^p^NJOsfs4CG5vl(Y@g=jons>E z{wRl|)&0-1oY3ZFEl5fms1MNE{4euNwE-o^R}J6UcY2oC>^NRZp?Tos0> z;at}iUr9mVeeJvwcV8i`&`J&c4#d;2)_?MH(CDLAl6<0AL9byaYgdiy3ajF5DI_1r z(raI$CbFXGu2uS9t=%7Lo z_v-CLaM)V|=Sgw0cCW?BdWU#oNV{MLqm#AQ9_$T9dyooR>vyu=>7unSVwALN-N$(L z0+Wv;ag5rTg}>>p)dzz`8F8(?`^>^$hikQVM2;e-TSHCU7TxRv7iD+@3JXk(09E*z zg+Av6r$G01y`M>+>hL-Gu8zl7mJ3(C7-7c)lpq3fOw}YPVW79Qc7ux{f__J9cbu3|4MP)Q+9$OPhqfRT9;5hv(YERNUMDep4x+?WuUYPeK`J?gWhl=?=nzInhb7|h%Wi_=I+td6F9B?kqPPm} zCWx`zLYp59LDMxMY*dxo8Ll}fqKP8wk?F_z)I-r;ylaHCs9s6OLCES%Q}cg z78~gn>Ux{eqquL*GC9WNb4&yjV}yxh3zCVit8;g9T2bSey*| zem}=$68RY@WjwjlSL6}a43QW$U5RKu_@)nH)Q*#EaJDQ538osl4xnJskUyR9^qX>~ zr9^kHHb?^rL=FLPvH%or>wF(T2%I?Wb>=yd0&#`==Uf!6bt6F0Ps5{yC>lS}j_`UN zel6?imI0)Xunp(*ejE^)Ck@z*ppO?2&Z9Pj@RHhIP^FC`-Yi{8!Y3l>TLjC{z;M?O z_-63ML%tm9Y0n8AP8uGy+E>O4GlxJmS~I!@taPWgiZ_5T5|eLjVb|X>_83PN16MMc z+)#|Jbuj1~2{a#z;*uMqYbx2z!8yJ2#5x3C3QFb&k>LD(o)a>#Vjkdm6x}vM3nc7S zryDb9iF7lJbKwI#`p#9-XWG`Vw}Y~1*xD&jE!>wN*e+WX0Qtkb+(yPJ=t3y$h{pY# z1d|18ziuMk*&!Wj#Eq3Def0WEKHbUb{=A*N-H#g-zsWKYt(F1)M6(pVd7YC;P-WOX zHS#5k6%ojFS^h$Z;Mz~2Y!}x~2e?)xcLF0K0?=mocDIg{M;t60DY4rbybVoCCjr!A zQo@}A@G$T$BG`5jI|JJ<6daNuh9W(md_DLss-3vdAv+lL@N0`J8VEQ=B)$Ef8CRbK1n{f?DMz!aIpuGr?5(gZ|PgH$B8peKi z2q43O;=J|@1W7NNCZr2AdW%&E!GKCy4Pyd)gHUfECy41uz%>9;p*tZYY>pEaxx_Tg zZ|9|~GJmN#hS&Tv?A~7FCF0#Fh?nX)#2#~_VGsdFqK$cP4u-c+pd=UGZnl9Ied^Ar zW1B(S&P}21L?5)J0Bvh7w1KY%v>|n&4M@2U+Gx_$Ml20&1s~d=%oAwC@Fd(&8ZQUh zeq4a|!4RN@F`pjq|2M&mn z9O>7X^DR8EZ4o?vT+{uFq3I5Dg77`*VL`jo(I1OBI?y!g9BDrXc|Pz_ZjNVhrgLQc9JCl@>^y_u zQ6OQEDD#UhC&++NUximJ^pOZ|3yussA!#uX?=A$5N|=8E1q1lN@Lip6Cp4J|hmRn9 z?`76*woCe5A=-PPxXAS2Uqr7WyniPw7q~veV6b-rl_obp zWYH4_?<*^Weubu%XGNUB9}Q>VGZA`(XJwp~dI&GVlC007e?zi!ltNi$X}DLoBmuJQ z5roVsOFATJMbhE5&)W#IE3NtrLbc|wZKc@D-GzX^Z%C|<_C`@|`eMpV^eP+JdO}eS zZT$TnLJIcfnd}XFdBE@XKfk9}-WUVKpbUiK!FdmU#Lx-D)3~p!R&NM-bzog=i$Pu) zFRy-#QEF+MXu-L+HnB9mw4FQXad#NpHd4&ugc=0|jJcjK%t6(T&Vew+z!li^^JGG1 z2K!+(SKljR6V48IXRNLvZH|R%z?-7h(KR4wudV^J%X3H@7t`jQZvLHUQ7C_}0xdXL zCFC9U^FAGhCS3}=%~H!rl(QtBifsArSV_aWBoTJ^g!SX(?Db9@VZUyUxS6{9_{n^| zjsn7Q(O#(I&Bh#KPBrejXBd0+PSqEnm|m$_%Z`C)vifYO-h}6j)v(-4x>J$Z>dO&p z5kcdYgFD^(Z+z(3?Fi#J1F;dusYp8_WwAB;-;Nk~j3)fpWM8Y`1P%R*Zw{TSp%=7| zG0Lrd@BnbAa9r#UWj|QIU~EmJ&hfU3@A6m2kInyqsmi5e~^$&*E}wEYKgwa64H6At#NlXw9bZ zvU|zKUcqD^lPj6frE|7GV3Jp6^E}0C`U@8vnAwsKTNtGb0~7O88P;-aQk-&l z9mjF)Nqai>FXg^*%72k3UNjc+=9T2$WXe~^MFfL4vi0TB|>_>jKy&pjYVBef30NkUDYn_Saz81E`1AgjV3Uc}dWm zD#k%kfK5P5fHKO(XzfB+2%=in85M_fhSI^tdBn+v3hJ9)4RHX-9D`$08U8aNPX%!& z^9MTSch3C&5PKXd04HHJk?0}zv`_w(_7H)0Sun)VM@d{53^Du+QctGL>UCDUR^>yMFax#Ya%C_JzDU1Ar&dbYtW*4&*yj`k$Z^zH2KEbp!*!pMzm+#H;!M8@bq(G`n9|ly;+`SC26^y&2u0}H^Top#EdZAc zK7~8B5K%;UAn?R|6Ez~wd_rN1G`JIbL=HcP8bbk}NBTGl5?l?J`S$!gLUCJ1G1gz> z-HS*N$4Q9or*Q7mBv*6YPr3Mpk4xsvcBfws3Tya;3~frxgnCgx38+%Pk@b24$S*Jx zc?J0OAU=bowSAxEUBZq|M7ft0j_4-B&aHUj(hwL{xc49bMSe0#&cxWFu|HjH_)qN~ zqC-a@i-Y4s7745`O|fNLK^8wO6~GgaxoqXPp_l^)$(<`{fiS%#{~drZFA^zXNW%$m z&=zn4VzCLF$mS>W_kHN{_7PUR_PF~9htZA^c#G&Aflv*qObE)xd??!%KpEsE5GY7K zS8HfgEHmt(itj{$pQj&)lq!6#8(>iDd_K+@6E<)NA#N91h%JH7WSQ9d)T$%ayk!HX zZpGMwS3txfr>HYY^dd6G0t`+C8R1h}bzd=G0r+~Fdf#hIGu>pr`vLwVxZB0{F2Ul_ zPaJ`Vt(%=T_BbGNT+AUMqUV-XAk>pQmk#9`Fn5DiLEolAMhawiwa>}NSd6gF2e>-8 zC{0-mzsj@NirD_JygH-Dc81|dPPNUqaD_a8?0SW8b&&U=%K-l4YXe$v&uNgcG7Mbn zwX&&;!N5xY|2q1ytRuZzj8lzfgUi8PKIUzNjn^{?md&sWM_%5)mkwlkf)IXl?!iL9GD`b2~0#ga!x~%!YQI1v^XTRa6^ro3+8VfWTyYUAy}Y zh=6Q+Vt7BcqlgGEieorv!*fh#6ih}WhF4EB6ETQpoNZhNB^5lgrSa&K0aKw@>vf41afFJ-?x`H zyyXqq|Mtv*zr*3Lc@D5EU+Xx)-Da>BTL2*0jJ=LLf1z#{)5p(qKcPGv8<$={!gOQ*Ya8Ldg_I>iZ z_f7xcPw(neiD{W4z0$yk^Ed|=k-y&`;BKPjdyv4BGe9;mK_LeH;r%)+HO}V@J&e!! zwO8J%oBDek^hq|>weAz|(0xkymatkO>~9_^!k>r|O5u+aVm%CbIyMnegydbN#fCzK z4t!dyL)Z-pyM~0gmpkUem^LjLrdksy5qgInHUGl#W6Bso*_9KYf%D=|`*`S% z_G;fAx|6U+-Kdld$vy__07H18WPqp+$FJp*RyowN}6S4DdQpY%&2RwWjNTh3Lm!VD-!T`k`7j9{67zkFI< zoP5@4H=n3$#Zz971}$YlZ_q+na4d_FRdj?(>5oWA*Y(NmvH)KjxXZ&V zi@1#Uz`q7Myw3d$EgpB4hg|<%z^M3e_(J{*yz?a{UtvOKsGFbGvG+j6Kf#-neF6uz zSJ6>kdp-+cqIfRq_rK)Q#44gz>Oe> z*6pqO^0c?j>bE%K^v><~*Z*xk+QsB|m=H#tlJ?jVcikyB{^jTRROmwNGZ%qLgz~eK zl`oiUN%$P!5m$a;S7YJE#`rWKE!61fC}m`8P44d%LkS`<2Mn$(svaMS&Qe_dzsyT6 zrdjVZ-|}6i=Xu)~!y$^vWXJsfw+#UGKO&g5@V2M2V#I!avW#BA+_xtwEdCVGd5_^K zx7rCXA@69jMhb=!l0Lr6$Yq`|5BH*dTdzQ$2$C!3?=XJW7m)zDi5+9*Q2 zFz#nz$kw3{=;leKs(j-`)TZ8dUO-^LmE$`vC?6Y~pzNfIv4Me$* z>vw$42=X+y0rf9GE}BiYo(NbTX9G6nc<$rVlS~NS{ha@OXus9~y@#0unMgt`i_br! z&DRj6T#rk=orF`)>4$D7q`lFR)u{Q2xRCIDf94Go;m^E6#V>c}Bl+C0YWmARlG0IE zwta43yYr=_`3^C`@D4WJ0X`Vn6b=UbwvYS$AK1A77qEr7Z(QU#$^ab68=t@#-vDER zHxP6i3Qf(VcyD|aYz5z-M;o+RJZc`dwhe6_T)xUe*+kBhVR3H$mdj1U{b!l zs7j(5rPES5q|(tXPs5qxklQAGu4}_NzH8)ur3fngP|#|eR|H0feW|@)k$*Mk{6m-w ztUu?u!AGa(IsJQ))g_mL*+AMr7Bz|Fd3SG!xb+p` zPcL^a_z3PBxY%-fqw<}1rG4Z$?XVYG4?x-zUp1f$)xsUwp#Ya` z5Iqn`Od|F#z?{zwb2d;G2w`z2;(4mla>Dv0On=^q+EpULAaZ$RuYxYY2YZ6NkJUPt0ej)%Xc))WJY6nM ze^2cwfENIHDfm(!7J<*f$@e&YJjY;t?icw48%2WZ`Fxl}Pk6~wrgqwohtUs2Tz;A) zvQbZv&X)XCgn?p3PELOYEJSKz9BOFiFD84d0A~UottdG7OSs@@8G%A=m)M~>3Bz^ zG`@*I#ar;c0(w_RJW(M-TXrb$DqUJ1!}kB0jjyQ1yQ2fKN%ZpR2ubisuxQ8a&A{HZlNKn<+^oKG&%e%5p z2oiBd$#|o5|Kx#RBN*O>93Fo}K-Ays)ZgAE7=k^O5A1~Hg+1XL7p9Wb0^9vNc*S8 zojq#@Tv5UOEbO#BUv*E-6gx`0;Ourk8{FCI)3DJjK$J3})lJOqh(2p~u=N6;mN!kn zb;22oK3#>v(J{PxCUJ{=X?Lzpi-0KtM!tcsjl&~Xe@ml|EE;@irPPegUY>CVg2~Zm goEpbCq!+=RT*b1RTnBb$kZ0`*i8Pf 0).any(): -# out_shape[out_shape < 0] = 0 -# self.paddle_graph.add_layer( -# kernel="fluid.layers.reshape", -# inputs={"x": node.name}, -# outputs=[node.name], -# shape=out_shape.tolist()) - -# if input.dtype == 'bool': -# self.paddle_graph.add_layer( -# kernel="fluid.layers.cast", -# inputs={"x": node.name}, -# outputs=[node.name], -# dtype="'bool'") def Pad(self, node): input = self.graph.get_node(node.layer.input[0]) @@ -668,37 +624,33 @@ class TFOpMapper(OpMapper): new_padding = paddings[2:6] transpose_name = gen_name("pad", "transpose") self.paddle_graph.add_layer( - kernel="fluid.layers.transpose", + kernel="paddle.transpose", inputs={"x": input.name}, outputs=[transpose_name], perm=[0, 3, 1, 2]) self.paddle_graph.add_layer( - kernel="fluid.layers.pad2d", - inputs={"input": transpose_name}, + kernel="paddle.nn.functional.pad", + inputs={"x": transpose_name}, outputs=[node.name], - paddings=new_padding) + pad=new_padding) self.paddle_graph.add_layer( - kernel="fluid.layers.transpose", + kernel="paddle.transpose", inputs={"x": node.name}, outputs=[node.name], perm=[0, 2, 3, 1]) return self.paddle_graph.add_layer( - kernel="fluid.layers.pad", + kernel="paddle.nn.functional.pad", inputs={"x": input.name}, outputs=[node.name], - paddings=paddings) + pad=paddings) def MirrorPad(self, node): - op_name = name_generator("pad", self.nn_name2id) - output_name = node.name - layer_outputs = [op_name, output_name] input = self.graph.get_input_node(node, 0) paddings = self.graph.get_input_node(node, 1) assert paddings.layer_type == "Const", "Padding should be Const" paddings = np.flip(paddings.value, 0).flatten().tolist() - dim = int(len(paddings) / 2) transpose_name = gen_name("pad", "transpose") self.paddle_graph.add_layer( kernel="paddle.transpose", @@ -706,9 +658,9 @@ class TFOpMapper(OpMapper): outputs=[transpose_name], perm=[0, 3, 1, 2]) self.paddle_graph.add_layer( - kernel="paddle.nn.Pad{}D".format(dim), + kernel="paddle.nn.functional.pad".format(dim), inputs={"x": transpose_name}, - outputs=layer_outputs, + outputs=[node.name], pad=new_padding) self.paddle_graph.add_layer( kernel="paddle.transpose", @@ -717,22 +669,13 @@ class TFOpMapper(OpMapper): perm=[0, 2, 3, 1]) def Squeeze(self, node): - input = self.graph.get_node(node.layer.input[0]) + input = self.graph.get_input_node(node, 0) squeeze_dims = node.get_attr('squeeze_dims') self.paddle_graph.add_layer( - kernel="fluid.layers.squeeze", - inputs={"input": input.name}, - outputs=[node.name], - axes=squeeze_dims) - - def Softmax(self, node): - input = self.graph.get_node(node.layer.input[0]) - axis = node.get_attr("axis") - self.paddle_graph.add_layer( - kernel="fluid.layers.softmax", - inputs={"input": input.name}, + kernel="paddle.squeeze", + inputs={"x": input.name}, outputs=[node.name], - axis=axis) + axis=squeeze_dims) def Shape(self, node): input = self.graph.get_input_node(node, 0) @@ -762,12 +705,12 @@ class TFOpMapper(OpMapper): outputs=[node.name]) def ArgMax(self, node): - input = self.graph.get_node(node.layer.input[0]) - axis = self.graph.get_node(node.layer.input[1]) + input = self.graph.get_input_node(node, 0) + axis = self.graph.get_input_node(node, 1) assert axis.layer_type == "Const", "ArgMax only support Const parameter" axis = axis.value self.paddle_graph.add_layer( - kernel="fluid.layers.argmax", + kernel="paddle.argmax", inputs={"x": input.name}, outputs=[node.name], axis=axis) @@ -786,8 +729,8 @@ class TFOpMapper(OpMapper): sorted=sort) def MatMul(self, node): - x = self.graph.get_node(node.layer.input[0]) - y = self.graph.get_node(node.layer.input[1]) + x = self.graph.get_input_node(node, 0) + y = self.graph.get_input_node(node, 1) transpose_a = node.get_attr('transpose_a') transpose_b = node.get_attr('transpose_b') if transpose_a is None: @@ -795,7 +738,7 @@ class TFOpMapper(OpMapper): if transpose_b is None: transpose_b = node.get_attr('adj_y') self.paddle_graph.add_layer( - kernel="fluid.layers.matmul", + kernel="paddle.matmul", inputs={"x": x.name, "y": y.name}, outputs=[node.name], @@ -820,8 +763,11 @@ class TFOpMapper(OpMapper): data_format = node.get_attr("data_format").decode() pad_mode = node.get_attr("padding").decode() - self.params[kernel.layer_name.replace( - '/', '_')] = numpy.transpose(kernel.value, (2, 3, 0, 1)) + self.paddle_graph.add_layer( + kernel="paddle.transpose", + inputs={"x": kernel.name}, + outputs=[kernel.name], + perm=[2, 3, 0, 1]) input_name = input.name if data_format == "NHWC": @@ -830,34 +776,32 @@ class TFOpMapper(OpMapper): dilations = [dilations[i] for i in [0, 3, 1, 2]] transpose_name = gen_name('depthwise_conv2d', 'transpose') self.paddle_graph.add_layer( - kernel="fluid.layers.transpose", + kernel="paddle.transpose", inputs={"x": input.name}, outputs=[transpose_name], perm=[0, 3, 1, 2]) input_name = transpose_name self.paddle_graph.add_layer( - kernel="fluid.layers.conv2d", - inputs={"input": input_name}, + kernel="paddle.nn.functional.conv2d", + inputs={"x": input_name, + "weight": kernel.name}, outputs=[node.name], - num_filters=in_shape[1], - filter_size=k_size[0:2], stride=strides[2:4], dilation=dilations[2:4], groups=k_size[3] * in_shape[1], padding=string(pad_mode), - param_attr=string(kernel.layer_name), - bias_attr=False) + bias=None) if data_format == "NHWC": self.paddle_graph.add_layer( - kernel="fluid.layers.transpose", + kernel="paddle.transpose", inputs={"x": node.name}, outputs=[node.name], perm=[0, 2, 3, 1]) def AvgPool(self, node): - input = self.graph.get_node(node.layer.input[0]) + input = self.graph.get_input_node(node, 0) k_size = node.get_attr("ksize") strides = node.get_attr("strides") @@ -868,13 +812,15 @@ class TFOpMapper(OpMapper): if data_format == "NHWC": transpose_name = gen_name("avg_pool", "transpose") self.paddle_graph.add_layer( - kernel="fluid.layers.transpose", + kernel="paddle.transpose", inputs={"x": input.name}, outputs=[transpose_name], perm=[0, 3, 1, 2]) strides = [strides[i] for i in [0, 3, 1, 2]] k_size = [k_size[i] for i in [0, 3, 1, 2]] input_name = transpose_name + + # TODO(syf): The op has diff. self.paddle_graph.add_layer( kernel="fluid.layers.pool2d", @@ -887,29 +833,31 @@ class TFOpMapper(OpMapper): if data_format == "NHWC": self.paddle_graph.add_layer( - kernel="fluid.layers.transpose", + kernel="paddle.transpose", inputs={"x": node.name}, outputs=[node.name], perm=[0, 2, 3, 1]) def Pack(self, node): - inputs = [self.graph.get_node(name) for name in node.layer.input] - input_names = [i.name for i in inputs] + inputs_list = list() + for i in range(len(node.inputs)): + inputs_list.append(self.graph.get_input_node(node, i)) + input_names = [i.name for i in inputs_list] axis = node.get_attr("axis") self.paddle_graph.add_layer( - kernel="fluid.layers.stack", + kernel="paddle.stack", inputs={"x": input_names}, outputs=[node.name], axis=axis) if len(node.out_shapes[0]) == 1: self.paddle_graph.add_layer( - kernel="fluid.layers.reshape", + kernel="paddle.reshape", inputs={"x": node.name}, outputs=[node.name], shape=[-1]) def Unpack(self, node): - input = self.graph.get_node(node.layer.input[0]) + input = self.graph.get_input_node(node, 0) axis = node.get_attr("axis") num = node.get_attr("num") shape = input.out_shapes[0] @@ -917,10 +865,10 @@ class TFOpMapper(OpMapper): if len(shape) == 1: if shape[0] > 0 and num == shape[0]: self.paddle_graph.add_layer( - kernel="fluid.layers.unsqueeze", - inputs={"input": input.name}, + kernel="paddle.unsqueeze", + inputs={"x": input.name}, outputs=[node.name], - axes=[0]) + axis=[0]) input_name = node.name axis = 1 else: @@ -929,41 +877,45 @@ class TFOpMapper(OpMapper): if len(layer_outputs) == 1: layer_outputs[0] = "[{}]".format(node.layer_name) self.paddle_graph.add_layer( - kernel="fluid.layers.unstack", + kernel="paddle.unstack", inputs={"x": input_name}, outputs=layer_outputs, axis=axis, num=num) def ConcatV2(self, node): - inputs = [self.graph.get_node(name) for name in node.layer.input[:-1]] - axis = self.graph.get_node(node.layer.input[-1]) + inputs_list = list() + for i in range(len(node.inputs) - 1): + inputs_list.append(self.graph.get_input_node(node, i)) + axis = self.graph.get_input_node(node, -1) assert axis.layer_type == "Const", "axis for ConcatV2 must be type Const" axis = axis.value if axis < 0: - axis += len(inputs[0].out_shapes[0]) + axis += len(inputs_list[0].out_shapes[0]) - input_names = [i.name for i in inputs] - for i, ipt in enumerate(inputs): - if ipt.dtype == 'bool': - cast_name = gen_name('concat', 'cast') - self.paddle_graph.add_layer( - kernel="fluid.layers.cast", - inputs={"x": ipt.name}, - outputs=[cast_name], - dtype="'int32'") - input_names[i] = cast_name + input_names = [i.name for i in inputs_list] self.paddle_graph.add_layer( - kernel="fluid.layers.concat", - inputs={"input": input_names}, + kernel="paddle.concat", + inputs={"x": input_names}, + outputs=[node.name], + axis=axis) + + def Concat(self, node): + inputs_list = list() + for i in range(1, len(node.inputs)): + inputs_list.append(self.graph.get_input_node(node, i)) + axis = self.graph.get_input_node(node, 0) + assert axis.layer_type == "Const", "axis for ConcatV2 must be type Const" + axis = axis.value + if axis < 0: + axis += len(inputs_list[0].out_shapes[0]) + + input_names = [i.name for i in inputs_list] + self.paddle_graph.add_layer( + kernel="paddle.concat", + inputs={"x": input_names}, outputs=[node.name], axis=axis) - if node.dtype == 'bool': - self.paddle_graph.add_layer( - kernel="fluid.layers.cast", - inputs={"x": node.name}, - outputs=[node.name], - dtype="'bool'") def AddN(self, node): inputs_list = list() @@ -977,10 +929,10 @@ class TFOpMapper(OpMapper): outputs=[node.name]) def StridedSlice(self, node): - input = self.graph.get_node(node.layer.input[0]) - begin = self.graph.get_node(node.layer.input[1]) - end = self.graph.get_node(node.layer.input[2]) - strides = self.graph.get_node(node.layer.input[3]) + input = self.graph.get_input_node(node, 0) + begin = self.graph.get_input_node(node, 1) + end = self.graph.get_input_node(node, 2) + strides = self.graph.get_input_node(node, 3) if strides.layer_type == "Const": strides = strides.value.tolist() @@ -1042,29 +994,44 @@ class TFOpMapper(OpMapper): new_end.append(999999) else: new_end.append(end[i]) + + if input.dtype == "bool": + self.paddle_graph.add_layer( + "paddle.cast", + inputs={"x": input.name}, + outputs=[input.name], + dtype=string("int32")) self.paddle_graph.add_layer( - kernel="fluid.layers.slice", + kernel="paddle.slice", inputs={"input": input.name}, outputs=[node.name], axes=[i for i in range(len(new_begin))], starts=new_begin, ends=new_end) + + if input.dtype == "bool": + self.paddle_graph.add_layer( + "paddle.cast", + inputs={"x": node.name}, + outputs=[node.name], + dtype=string("bool")) + if len(new_axes) > 0: self.paddle_graph.add_layer( - kernel="fluid.layers.unsqueeze", - inputs={"input": node.name}, + kernel="paddle.unsqueeze", + inputs={"x": node.name}, outputs=[node.name], - axes=new_axes) + axis=new_axes) if len(shrink_axes) > 0: if len(input.out_shapes[0]) + len(new_axes) <= 1: pass else: self.paddle_graph.add_layer( - kernel="fluid.layers.squeeze", - inputs={"input": node.name}, + kernel="paddle.squeeze", + inputs={"x": node.name}, outputs=[node.name], - axes=shrink_axes) + axis=shrink_axes) def Prod(self, node): input = self.graph.get_input_node(node, 0) @@ -1081,25 +1048,25 @@ class TFOpMapper(OpMapper): axis=axis) def Split(self, node): - dim = self.graph.get_node(node.layer.input[0]) - input = self.graph.get_node(node.layer.input[1]) + dim = self.graph.get_input_node(node, 0) + input = self.graph.get_input_node(node, 1) assert dim.layer_type == "Const" num_split = node.get_attr('num_split') dim = dim.value self.paddle_graph.add_layer( - kernel="fluid.layers.split", - inputs={"input": input.name}, + kernel="paddle.split", + inputs={"x": input.name}, outputs=[ "{}_p{}".format(node.layer_name, i) for i in range(num_split) ], num_or_sections=num_split, - dim=dim) + axis=dim) def Slice(self, node): - input = self.graph.get_node(node.layer.input[0]) - begin = self.graph.get_node(node.layer.input[1]) - size = self.graph.get_node(node.layer.input[2]) + input = self.graph.get_input_node(node, 0) + begin = self.graph.get_input_node(node, 1) + size = self.graph.get_input_node(node, 2) inputs = {"x": input.name} attrs = {} @@ -1124,143 +1091,147 @@ class TFOpMapper(OpMapper): shape = size.out_shapes[0] reshape_name = gen_name("slice", "reshape") self.paddle_graph.add_layer( - kernel="fluid.layers.reshape", + kernel="paddle.reshape", inputs={"x": size.name}, outputs=[reshape_name], shape=shape) inputs['shape'] = reshape_name self.paddle_graph.add_layer( - kernel="fluid.layers.crop_tensor", + kernel="paddle.crop", inputs=inputs, outputs=[node.name], **attrs) def ResizeNearestNeighbor(self, node): - input = self.graph.get_node(node.layer.input[0]) - resize_shape = self.graph.get_node(node.layer.input[1]) + input = self.graph.get_input_node(node, 0) + resize_shape = self.graph.get_input_node(node, 1) data_format = "NHWC" - inputs = {"input": input.name} - attrs = {"align_corners": node.get_attr("align_corners")} + inputs = {"x": input.name} + attrs = {"align_corners": node.get_attr("align_corners"), + "mode": string("nearest"), + "align_mode": 1} if resize_shape.layer_type == "Const": resize_shape = resize_shape.value.tolist() - attrs["out_shape"] = resize_shape + attrs["size"] = resize_shape else: shape = resize_shape.out_shapes[0] reshape_name = gen_name("resize_nearest", "reshape") self.paddle_graph.add_layer( - kernel="fluid.layers.reshape", + kernel="paddle.reshape", inputs={"x": resize_shape.name}, outputs=[reshape_name], shape=shape) - inputs["out_shape"] = reshape_name + inputs["size"] = reshape_name if data_format == "NHWC": transpose_name = gen_name("resize_nearest", "reshape") self.paddle_graph.add_layer( - kernel="fluid.layers.transpose", + kernel="paddle.transpose", inputs={"x": input.name}, outputs=[transpose_name], perm=[0, 3, 1, 2]) - inputs["input"] = transpose_name + inputs["x"] = transpose_name self.paddle_graph.add_layer( - kernel="fluid.layers.resize_nearest", + kernel="paddle.nn.functional.interpolate", inputs=inputs, outputs=[node.name], **attrs) if data_format == "NHWC": self.paddle_graph.add_layer( - kernel="fluid.layers.transpose", + kernel="paddle.transpose", inputs={"x": node.name}, outputs=[node.name], perm=[0, 2, 3, 1]) def ResizeBilinear(self, node): - input = self.graph.get_node(node.layer.input[0]) - resize_shape = self.graph.get_node(node.layer.input[1]) + input = self.graph.get_input_node(node, 0) + resize_shape = self.graph.get_input_node(node, 1) data_format = "NHWC" - inputs = {"input": input.name} - attrs = {"align_corners": node.get_attr("align_corners")} + inputs = {"x": input.name} + attrs = {"align_corners": node.get_attr("align_corners"), + "mode": string("bilinear"), + "align_mode": 1} if resize_shape.layer_type == "Const": resize_shape = resize_shape.value.tolist() - attrs["out_shape"] = resize_shape + attrs["size"] = resize_shape else: shape = resize_shape.out_shapes[0] reshape_name = gen_name("resize_bilinear", "reshape") self.paddle_graph.add_layer( - kernel="fluid.layers.reshape", + kernel="paddle.reshape", inputs={"x": resize_shape.name}, outputs=[reshape_name], shape=shape) - inputs["out_shape"] = reshape_name + inputs["size"] = reshape_name if data_format == "NHWC": transpose_name = gen_name("resize_bilinear", "reshape") self.paddle_graph.add_layer( - kernel="fluid.layers.transpose", + kernel="paddle.transpose", inputs={"x": input.name}, outputs=[transpose_name], perm=[0, 3, 1, 2]) - inputs["input"] = transpose_name + inputs["x"] = transpose_name self.paddle_graph.add_layer( - kernel="fluid.layers.resize_bilinear", + kernel="paddle.nn.functional.interpolate", inputs=inputs, outputs=[node.name], **attrs) if data_format == "NHWC": self.paddle_graph.add_layer( - kernel="fluid.layers.transpose", + kernel="paddle.transpose", inputs={"x": node.name}, outputs=[node.name], perm=[0, 2, 3, 1]) def Cast(self, node): - input = self.graph.get_node(node.layer.input[0]) + input = self.graph.get_input_node(node, 0) dtype = node.dtype self.paddle_graph.add_layer( - kernel="fluid.layers.cast", + kernel="paddle.cast", inputs={"x": input.name}, outputs=[node.name], dtype=string(dtype)) def Sum(self, node): - input = self.graph.get_node(node.layer.input[0]) - reduce_idx = self.graph.get_node(node.layer.input[1]) + input = self.graph.get_input_node(node, 0) + reduce_idx = self.graph.get_input_node(node, 1) assert reduce_idx.layer_type == "Const", "Only support Const parameter[reduce_idx]" keep_dims = node.get_attr("keep_dims") dim = reduce_idx.value.tolist() self.paddle_graph.add_layer( - kernel="fluid.layers.reduce_sum", - inputs={"input": input.name}, + kernel="paddle.sum", + inputs={"x": input.name}, outputs=[node.name], - dim=dim, - keep_dim=keep_dims) + axis=dim, + keepdim=keep_dims) def Max(self, node): - input = self.graph.get_node(node.layer.input[0]) - reduce_idx = self.graph.get_node(node.layer.input[1]) + input = self.graph.get_input_node(node, 0) + reduce_idx = self.graph.get_input_node(node, 1) assert reduce_idx.layer_type == "Const", "Only support Const parameter[reduce_idx]" keep_dims = node.get_attr("keep_dims") dim = reduce_idx.value.tolist() self.paddle_graph.add_layer( - kernel="fluid.layers.reduce_max", - inputs={"input": input.name}, + kernel="paddle.max", + inputs={"x": input.name}, outputs=[node.name], - dim=dim, - keep_dim=keep_dims) + axis=dim, + keepdim=keep_dims) def RandomUniform(self, node): - shape = self.graph.get_node(node.layer.input[0]) + shape = self.graph.get_input_node(node, 0) if shape.layer_type == "Const": shape = shape.value.tolist() self.paddle_graph.add_layer( - kernel="fluid.layers.uniform_random", + kernel="paddle.uniform", inputs={}, outputs=[node.name], shape=shape, @@ -1268,16 +1239,16 @@ class TFOpMapper(OpMapper): max=0.9999) else: self.paddle_graph.add_layer( - kernel="fluid.layers.uniform_random", + kernel="paddle.uniform", inputs={'shape': shape.name}, outputs=[node.name], min=0.0, max=0.9999) def Conv2DBackpropInput(self, node): - out_shape = self.graph.get_node(node.layer.input[0]) - kernel = self.graph.get_node(node.layer.input[1]) - input = self.graph.get_node(node.layer.input[2]) + out_shape = self.graph.get_input_node(node, 0) + kernel = self.graph.get_input_node(node, 1) + input = self.graph.get_input_node(node, 2) assert kernel.layer_type == "Const", "Kernel of Conv2DBackpropInput should be Const" @@ -1292,15 +1263,15 @@ class TFOpMapper(OpMapper): in_shape = self.decoder.infer_tensor(input, use_diff_inputs=False).shape k_size = kernel.out_shapes[0] if k_size.count(-1) > 2: - k_size = self.decoder.infer_tensor(input, use_diff_inputs=False).shape + k_size = self.decoder.infer_tensor(kernel, use_diff_inputs=False).shape pad_mode = node.get_attr("padding").decode() strides = node.get_attr("strides") dilations = node.get_attr("dilations") data_format = node.get_attr("data_format").decode() - self.params[kernel.layer_name.replace( - '/', '_')] = numpy.transpose(kernel.value, (3, 2, 0, 1)) + kernel_name = node.name + ".weight" + self.params[kernel_name] = numpy.transpose(kernel.value, (3, 2, 0, 1)) input_name = input.name if data_format == "NHWC": @@ -1309,20 +1280,26 @@ class TFOpMapper(OpMapper): dilations = [dilations[i] for i in [0, 3, 1, 2]] transpose_name = gen_name("conv2dbackpropinput", "transpose") self.paddle_graph.add_layer( - kernel="fluid.layers.transpose", + kernel="paddle.transpose", inputs={"x": input.name}, outputs=[transpose_name], perm=[0, 3, 1, 2]) input_name = transpose_name self.paddle_graph.add_layer( - kernel="fluid.layers.conv2d_transpose", - inputs={"input": input_name}, + kernel="paddle.static.create_parameter", + inputs={}, + outputs=["{}_{}".format(node.name, kernel_name).replace(".", "_")], + dtype=string(str(self.params[kernel_name].dtype)), + shape=self.params[kernel_name].shape, + name=string(kernel_name)) + + self.paddle_graph.add_layer( + kernel="paddle.nn.functional.conv2d_transpose", + inputs={"x": input_name, + "weight": "{}_{}".format(node.name, kernel_name).replace(".", "_")}, outputs=[node.name], - bias_attr=False, - param_attr=string(kernel.layer_name), - num_filters=k_size[2], - filter_size=k_size[0:2], + bias=None, stride=strides[2:4], dilation=dilations[2:4], padding=string(pad_mode), @@ -1330,7 +1307,7 @@ class TFOpMapper(OpMapper): if data_format == "NHWC": self.paddle_graph.add_layer( - kernel="fluid.layers.transpose", + kernel="paddle.transpose", inputs={"x": node.name}, outputs=[node.name], perm=[0, 2, 3, 1]) @@ -1403,11 +1380,12 @@ class TFOpMapper(OpMapper): shape=node.out_shapes[0]) def SquaredDifference(self, node): - x = self.graph.get_node(node.layer.input[0]) - y = self.graph.get_node(node.layer.input[1]) + x = self.graph.get_input_node(node, 0) + y = self.graph.get_input_node(node, 1) inputs = {"x": x.name, "y": y.name} x_shape = x.out_shapes[0] y_shape = y.out_shapes[0] + # TODO(syf) layer_id = self.paddle_graph.add_layer( "fluid.layers.elementwise_sub", inputs=inputs, outputs=[node.name]) self.paddle_graph.layers[layer_id].input_shapes = {"x": x_shape, "y": y_shape} @@ -1416,14 +1394,14 @@ class TFOpMapper(OpMapper): x_shape = node.out_shapes[0] y_shape = node.out_shapes[0] layer_id = self.paddle_graph.add_layer( - "fluid.layers.elementwise_mul", inputs=inputs, outputs=[node.name]) + "paddle.multiply", inputs=inputs, outputs=[node.name]) self.paddle_graph.layers[layer_id].input_shapes = {"x": x_shape, "y": y_shape} def OneHot(self, node): - input = self.graph.get_node(node.layer.input[0]) - depth = self.graph.get_node(node.layer.input[1]) - on_value = self.graph.get_node(node.layer.input[2]) - off_value = self.graph.get_node(node.layer.input[3]) + input = self.graph.get_input_node(node, 0) + depth = self.graph.get_input_node(node, 1) + on_value = self.graph.get_input_node(node, 2) + off_value = self.graph.get_input_node(node, 3) assert depth.layer_type == 'Const', 'Parameter depth should be Const in OneHot' assert on_value.layer_type == 'Const', 'Parameter on_value should be Const in OneHot' assert off_value.layer_type == 'Const', 'Parameter off_value should be Const in OneHot' @@ -1437,73 +1415,72 @@ class TFOpMapper(OpMapper): 0.0) < 1e-06, "off_value should be 0 in OneHot" self.paddle_graph.add_layer( - "fluid.one_hot", - inputs={"input": input.name}, + "paddle.nn.functional.one_hot", + inputs={"x": input.name}, outputs=[node.name], - depth=depth.value) + num_classes=depth.value) def Pow(self, node): - x = self.graph.get_node(node.layer.input[0]) - factor = self.graph.get_node(node.layer.input[1]) + x = self.graph.get_input_node(node, 0) + factor = self.graph.get_input_node(node, 1) inputs = {"x": x.name} attr = dict() if factor.layer_type == 'Const': - attr["factor"] = factor.value.tolist() + attr["y"] = factor.value.tolist() else: - inputs["factor"] = factor.name + inputs["y"] = factor.name self.paddle_graph.add_layer( - "fluid.layers.pow", inputs=inputs, outputs=[node.name], **attr) + "paddle.pow", inputs=inputs, outputs=[node.name], **attr) def All(self, node): - input = self.graph.get_node(node.layer.input[0]) - reduce_idx = self.graph.get_node(node.layer.input[1]) + input = self.graph.get_input_node(node, 0) + reduce_idx = self.graph.get_input_node(node, 1) assert reduce_idx.layer_type == "Const", "Only support Const parameter[reduce_idx]" attr = dict() - attr["dim"] = reduce_idx.value.tolist() - attr["keep_dim"] = node.get_attr("keep_dims") + attr["axis"] = reduce_idx.value.tolist() + attr["keepdim"] = node.get_attr("keep_dims") input_name = input.name if input.dtype != "bool": input_name = gen_name("all", "cast") self.paddle_graph.add_layer( - "fluid.layers.cast", + "paddle.cast", inputs={"x": input.name}, outputs=[input_name], dtype=string("bool")) self.paddle_graph.add_layer( - "fluid.layers.reduce_all", - inputs={"input": input_name}, + "paddle.all", + inputs={"x": input_name}, outputs=[node.name], **attr) node.layer.attr['dtype'].type = 10 def GatherV2(self, node): - embeddings = self.graph.get_node(node.layer.input[0]) - index = self.graph.get_node(node.layer.input[1]) - axis = self.graph.get_node(node.layer.input[2]) + embeddings = self.graph.get_input_node(node, 0) + index = self.graph.get_input_node(node, 1) + axis = self.graph.get_input_node(node, 2) assert axis.layer_type == 'Const', "Only support Const parameter[axis]" axis = axis.value - assert axis == 0, "Only support axis=0 in GatherV2 OP" index_name = index.name if len(index.out_shapes[0]) != 1: reshape_name = gen_name("gather", "reshape") index_name = reshape_name self.paddle_graph.add_layer( - "fluid.layers.reshape", + "paddle.reshape", inputs={"x": index.name}, outputs=[reshape_name], shape=[-1]) - inputs = {'input': embeddings.name, 'index': index_name} + inputs = {'x': embeddings.name, 'index': index_name} self.paddle_graph.add_layer( - "fluid.layers.gather", + "paddle.gather", inputs=inputs, outputs=[node.name], - overwrite=False) + axis=axis) if len(index.out_shapes[0]) != 1: out_shape = node.out_shapes[0] self.paddle_graph.add_layer( - kernel="fluid.layers.reshape", + kernel="paddle.reshape", inputs={"x": node.name}, outputs=[node.name], shape=out_shape) @@ -1518,19 +1495,19 @@ class TFOpMapper(OpMapper): outputs=[node.name]) def ExpandDims(self, node): - x = self.graph.get_node(node.layer.input[0], copy=True) - y = self.graph.get_node(node.layer.input[1], copy=True) - inputs = {"input": x.name} + x = self.graph.get_input_node(node, 0, copy=True) + y = self.graph.get_input_node(node, 1, copy=True) + inputs = {"x": x.name} attr = dict() if y.layer_type == 'Const': dim = y.value.tolist() if not isinstance(dim, list): dim = [dim] - attr['axes'] = dim + attr['axis'] = dim else: - inputs['axes'] = y.name + inputs['axis'] = y.name self.paddle_graph.add_layer( - "fluid.layers.unsqueeze", + "paddle.unsqueeze", inputs=inputs, outputs=[node.name], **attr) diff --git a/x2paddle/optimizer/__pycache__/__init__.cpython-37.pyc b/x2paddle/optimizer/__pycache__/__init__.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..ae36ed5ccada2562a2e52dc61d5b1ce0d6468d61 GIT binary patch literal 158 zcmZ?b<>g`kf(Kp~;z9Id5CH>>K!yVl7qb9~6oz01O-8?!3`HPe1o6vVKR2&LzqmB7 zGBGbLF)!V~P(Q*bATcE+CpA7fKP45;)2}cp01NBq7nEe?W>%#Z>Bq-s=4F<|$LkeT U-r}&y%}*)KNwou6@EM3100t5#wg3PC literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/__pycache__/optimizer.cpython-37.pyc b/x2paddle/optimizer/__pycache__/optimizer.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..943baaeec7d01fb960e1d5cc3a8f0b6339d8e4c8 GIT binary patch literal 1897 zcmahK%Wm8@kko6n@;XkO#CF~0;X?v!QS3TC^wOdT+5#y8)HP58DFg%rSESZ*w3JM7 zQ?D!cVxM~GH#o=sP0u_P=&8TZQ-{)OEhjcgfy3d*eJD!T9UtpXApig#I?0 zW;lR6hglfF5W^{&;1FYjTdB2Qul9U(=BvApULa;O=NvJIduMjoVeXHp@BIT>P#^21 zJz5msQ$fdE!WMrT8AdZ~K%T=a9s=Mf#0-TN^DBiow3*Aia}+wP11h`v)(c7VgCBC0 zOp+5Wb%)KyhN4|1X~ZKFf*EG*!2AYg@jHMLGyKMq+ojDcST~tH#w8-?=r`sV8>|C1 zyDS)Y3{`|6zXWporNw%%zGTY_`CBz6J_nf<;GNH96uZstuvK<6(Za(TxNtBAQ-N0pY47A|JA}rQ-0D#yBMAEx1KlJNAbSSH)zJ zotA&=4L_GWowgVoPy#GJ-@&I^lXLhqxhaYL;{DA1128{y=gC();gs!--JCVk%}-#8iy8 zpACQ7*{e@P8RLuH@OYiUv zdf?1g`(*uU;|>~~k!AG3FkN8oMb)tqkmIyOhv-}MdUR?jd|{Q=JA8&t?b5oi8GeW5 zptLWX5-^O!;P^A;miW*+!>7(AeL-_er*xrLxA)M|_A8`1%)0O^jd1$Ne+BOYbfvu; zuDn6FG{?WI}vW}0iC3s(G`D0qNqo~JWyr(6J+&bTbJn@i|F z?ZwcZRP{XxFD}w<^Q; zfN&4K0QVs6LG}6A!>jm%q&Fvb`&QM1)zi~=O@)mggy?{fNyesdtBCFqax|rB&5?J& zj@&fBT(1@|PD%6L-iKKj1g^%Din81H!$-~a8Z_4BpbB?V{}HNuhFjN2!9$p2VleO4 z@bS$ebti>SuaRgPxvlBR|H#$qw3m$zd8DdHTJcul*OmCP&ed0jRG7=rs~W>Jz<)d% BHq`(C literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/__pycache__/pass_.cpython-37.pyc b/x2paddle/optimizer/__pycache__/pass_.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..db9980479f15f039c9ff4fbef3fedbebca9eea85 GIT binary patch literal 746 zcmZuv&1&2*5SC=`M%iYekPzt2J(YzLw>g&*+LY29OrSj!QxJ}&-F2{KSCXN-^wiy# z>D%P2Qz4JgTTh+QW;dloFh80Z&wTUAZ^q*hf%EI)Mtq3~xy8%&#NeEP*dtKyi6oM) zK@?d_83WQ0}eCispNgL!W&g zUl!K-xC33%yN+m4-hx;JBMH>f>?bH6TUAdDjtuJ0dD>`~+FmaTJuUR?#mn^b>182h ztwh;MwY#O?PP-6J+s;)@wNxg>vErm#crHYxD<{Mu&SannU5*jk$u?7lC?wFL1b7GF zI3Q<5f56%RlV!_ipXMqlx~^U%(_WXZYIV}|)+JXeshX}Dq2tQ2W5Q@K(>kWxWp)MK)HY#iTe|tVtPGJ7Y zZUVa_Vw=e(9Qnc`A^b>)rj^Hv(9?Vr)% sIdMIM@BSFXa!{Pcj7K~Q(QLx}u)Vrgr3?3mKKK092Fz>h@f(Ev1z0Pj*#H0l literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/__pycache__/pass_manager.cpython-37.pyc b/x2paddle/optimizer/__pycache__/pass_manager.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..f563960ed75fe7f3b1f4acd207c54015e8d22cfd GIT binary patch literal 1106 zcmZuwOKTKC5bo}I?CfL~j06)!FyO_#WHsj?Asz(9CE!K&(u~ucb&`46(~D+VLEOc& zS8v&yUOf9VJkLo8lK^1uBz_(>Z_WgPRA!8zn}f$KW#!T@vsy)lS5eL zUD%!yK?I$_l4NQPs9?gHkpUC-OVV{Bh**gQ+99m+2sTL|7USr^lJ=AgoW)I7wuCL5 z8JW_72f8a-MsI`eiMG*w;lt}2)y8pHmM_9A97;911+VTFV7U4*zxwj|&)094A3yoK zAQ1%@3Vv}tfmLD|g1|ep{~flxf=s1nXQuPoAPitg*#d+m%9%r%Ud>d0^5<?AIA4avn6vTPv0yEP0aN@S?+!MUb{ zjW?UwjrYIitrkv4gi15Z~$sY4w{snln!v&d>4^-0&Hi2oFGYT_e1{zx64W%|!m{$xA z7z!&&%4^B#V4Jv-Lui~-4-gJgNvKBys_@_H!=rBg`cy)-IBdpvT|t$1P;>U!RE4;2 GZ2tq(oa~nX literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/__pycache__/pattern_matcher.cpython-37.pyc b/x2paddle/optimizer/__pycache__/pattern_matcher.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..28555fc6843a32595578a01042e59eb62d7833af GIT binary patch literal 7474 zcmb7JU2GiJb)Ns3T`s95MTwN^x`YELS|?fAidws_;3_JiI8MT-umKm_7@H+$DDQH) zySy{2tcck}7HxAGTN0Dlv17{t8N{|?#ED^~iXwuIB1bDX#8mOL|6k4cB~1$rxz2TpR5+x0|l>h?0tJW44k~qk)q-*DZRV z6}ie2Gt}1p>8KpTj0xnSlAPt0gojDGZ{=O6I+BPc>ORI*B-E-83xu7cU>R7#KRid-Nvig`bX z?0H_uPic|id&OBn?D(M~9K8N(>d<^CIOJDLXL6<4TD3z7Ew6 zSNPb6%9S8LpI`LEq1HZGxq(Pm&O}Z&n=j>qY<85Wq@5Ji(ig{eo+;fP6&XRfQl25k z(U0Er@DY!o5`ukmnivh0H5_*xS5_K9n})UnXj5eyYTCM@f(->7Owgyq0Om%9=brLX zMl|Mnfj1LmTjcuL=`&f{_O=}~CmIvpeEGDO#UqNjO2u>K8HI&)Q@XHedGizrKosZG zXXo-Imsu){7>%&Y6;I>>Z$Cc#1PVoU)C8V5p2fX=JZTN?+&A41?MDpppF!bT%gQXI z2@)GyQEDOpOA0C-eMy~E9{^%#T{lBhJOtJlYp4+mYez8CC|Eq} z0A^TqJFr8$pf0JdF=vQJ!BB@rgw{TVW-L7Vn6hx^V~T5ny>|69h?&@2d9T@cd*kYb z&CBb}8yB0;UE6;8{LPo2YQDAFeC54A-FPlO(erCwyV8978QF(Hd3WR8*EXO1_U6hr zH!pmvdHrJZt(P{Rrv?4$H1jo^cMlhHXFTEm@$9+IQYU6uq_F=!Jj9{hc>LdjN3L2o zNKDq1x*4dUDfv>j0)0ur`E<0Y!~^0ML`ZpB+lN}+S=DQ`(4qE8WmSK?0Di>kapD(d z5Z9Q$p|KBJ)(65kxH3?%=CoB^Tngi%9mb9-XtC>wFoqqRIcrrF-v|?!>&)5WN@$P> z`#%VEaoN=#Q9^qy7RH#XlcoEEIL!%TL<{Pe6DFuWr;C4U&HO!PzTZEyJ*z(0o>L#H zJrfLIjlm_{Bjn+9eOEXb?gG`xaERZ-;jpXE>7Xx(v)L^UjfBZ?WX?wGL%fT%mI#N} z)o=vwbrwo;qE1MEs66Dt*fh=f3CCtzw`e8v2*{0$v z9BgxwXzA>)iHAG;qyA^^ukrb{{_`(P(urIf{L=iFx$dx(5LE+(VZPruy>SDY_+9HX zn*U3jMiam4KaCr!kJe5GBXTb8^Ef1PSA7q38>`WMVh?`9!^GV9d858J+zZnfgJkZ< z?d}fuvca@mtv-5ES=AQ)fV&x8)%~xv_SXHQ*jpFJXz#8>;NBsh^|9K)Zd*7e?+0`w z>*FlLwy=ch{w*9Q!YPj>+wvf?4x5RvU8 z`VbyVOThmoR6~p78A6+^YiNzDBN##cna0rUA>F`C8XL#FggUM!HH$`QvXxdyYW=O6 zKF&MRJPqrNqNH`H#M&CIZ1K7Yyc3`y$$d#sH-z1|PhC8Eat9Jje+Z6AABB|XivEM? z9b6&xm`bT9QZaE5RY533`~?-{LmA3M7J(Z87vhice7OlT>_)iWC(Dy`{p_=TGtY52By|he8|}IBf)P7)gAS z=_S)g=2t^G%Eq;~cL*(-y7-0v-#n0?y}iNfwo0#h0xTQ@N$lyxIMpaBsbFRjTO%EukjCvJv4c$?IiDZl@#*5)p`HboVrk=djo^e1Ogp)ak7C$A{G2Q! zQ~!rm7k`WGdrbNeJ~5mOjJ#SR^mcY#JU|`Dp0w__d&#uKWjxZV8_tnG*Q(QTX2>Ja zkIY@is?Xx%lu}n~Uq)04mFL)H3TlK%)Hs!N04&$SExmnR}09O60pP&Vs|DclLL9zXMRT} ze`f{(cDlI*@t3p^GrP<2U1;0F<5MC<0SXy~Dv<(_a#?t8@ueS_yVWAq6WQ!SHCL29g0g6WvPRafaY6q$N9Qq zJI4g|+~VHYi+>+IxFdLS#26Hn!yR!#XMd={IpC&X%8!T0#gS9PH#rI2HC>A$_H`9e zhElql+Q;BlXw+D1T>r>1g~r1D=wyBo&qFF)j{32(*&1!Rrd+qK23ANpHFJq%#jd^t zah=#)yS91$#}(nt=4%~m+WgjYes#9DW$VJvAU4~V&ThT_q!g-NnfZ)3uCso9}+NkN!X1 zcxB_gXQ5Z`pMCYO?@Qm4zW4pJm(aN5%p8JD9Bk(i(=%n^r7NP0_W5*AN)>#NjNJOg zD_iT&RdT*hr{c6y`&5yd7JDE_ks__(HYG+=c4YdMVqUt+$S9QaC28$HPDAy4DPW+Q zvNF5+4v^W^#~(J785^Fnf@q9gZRG3>IJ#;;_VhpT_!O=wHb)oGR7V|$x!;eNVMHS2 z8R~BTNJEa{@yU7Lk}MwRB#RTAE>?Tf#SqWI$T?BmV`W0jADdhuRs|(SPFua zW`LXqc>>1w%9;&O2OVG60qgF97ohAVTYlJ?7_G~PMp~F0keEb~ibru=EQ}#C?S3>k zUCkHisyXo$aj;lh+2S>iXYd zBIAox{5`EOG@BQG(6KJ!bJWwvYO=_~#A>P_rOu9k5L5!f6;~IN9mcoQ$Rr8$qwA!t zSt`DQO54pMEoq3!wp`RE$)$qb6EMmY7jQbYpd)ND)+~7`gfV#9!v^_fV<51im1V z{P=~s2G^#^*sUG6L1+;ZL1>Ly5q$=woH=c8QjAh)MKP#Z$-9Et?d_o%BH`0@f*#juQY#ht$F?V=EZZHFT8!r^{@&p-Z@*W=H0Y=hV4jt8m1^6 zl(VOEBA=TsA~SY6SFCzS(2<_`|GZ7i+Wv3k^QKY*x6>qxC{#WzC z6H;9WXR$@!k)a?Z5|2>9nnU=VBchgyGv?{nr+>ttgLMCFvh}VeC0U=4{?QiqbmX8{ zp*wi01*>dvj5hoX700RQA}A6vK|A&A&T`1%!o%oD5DC--Z(+m-;t4^mK$y{<+xkg1 n(~(u4F5nL>23a!oLbGnSN#q=rIPXCUa(JhNp8UGvIeY#WFm$o} literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/code_optimizer/hierachical_tree.py b/x2paddle/optimizer/code_optimizer/hierachical_tree.py index 774bad6..c2aa31b 100644 --- a/x2paddle/optimizer/code_optimizer/hierachical_tree.py +++ b/x2paddle/optimizer/code_optimizer/hierachical_tree.py @@ -201,6 +201,7 @@ class HierarchicalTree(Tree): code_str = gen_layer_code(self.pd_graph, sub_layers, module_name, different_attrs=diff_attrs_column) +# print(code_str) self.codes.append(code_str) for sub_layers in sub_layers_list: inputs, outputs = get_inputs_outputs(self.pd_graph, sub_layers) @@ -358,7 +359,7 @@ class HierarchicalTree(Tree): run_func_list.append(" # {}: 形状为{}，类型为{}。".format(k, v[0], v[1])) run_func_list.extend( [" paddle.disable_static()", - " params = paddle.load('{}/model.pdparams')".format(osp.abspath(save_dir)), + " params, _ = fluid.load_dygraph('{}/model')".format(save_dir), " model = {}()".format(self.pd_graph.name), " model.set_dict(params)", " model.eval()", @@ -370,12 +371,7 @@ class HierarchicalTree(Tree): self.update_parameters() import_list = ["import paddle", "import paddle.fluid as fluid", - "from paddle.fluid.initializer import Constant", - "from paddle.fluid.param_attr import ParamAttr", - "import math", - "from x2paddle.op_mapper.dygraph.pytorch2paddle " + \ - "import pytorch_custom_layer as x2paddle_nn" - "\n",] + "",] import_str = "\n".join(import_list) if not osp.exists(save_dir): os.makedirs(save_dir) diff --git a/x2paddle/optimizer/code_optimizer/layer_code_generator.py b/x2paddle/optimizer/code_optimizer/layer_code_generator.py index 3861eb4..37fdc9b 100644 --- a/x2paddle/optimizer/code_optimizer/layer_code_generator.py +++ b/x2paddle/optimizer/code_optimizer/layer_code_generator.py @@ -29,9 +29,9 @@ NN_KERNEL_NAME = {"paddle.nn.BatchNorm": "bn", "paddle.nn.Tanh": "tanh", "paddle.nn.AvgPool2D": "pool", "paddle.nn.MaxPool2D": "pool", - "paddle.nn.Pad1D": "pad", - "paddle.nn.Pad2D": "pad", - "paddle.nn.Pad3D": "pad", + "paddle.nn.Pad1d": "pad", + "paddle.nn.Pad2d": "pad", + "paddle.nn.Pad3d": "pad", "paddle.nn.Dropout": "dropout", "paddle.nn.GELU": "gelu", "paddle.nn.Hardtanh": "tanh", @@ -175,11 +175,9 @@ def gen_layer_code(graph, sub_layers, sub_layers_name, different_attrs=list()): if layer.kernel.startswith("paddle.nn") and index == 0: continue if not output_name.startswith("x") or output_name in outputs \ - or layer.kernel == "prim.assert": + or layer.kernel == "prim.assert" or \ + layer.kernel == "prim.if" or layer.kernel == "prim.loop": continue - elif layer.kernel == "prim.if" or layer.kernel == "prim.loop": - if index != 0: - outputs.append(output_name) elif output_name not in outputs: outputs.append(output_name) continue @@ -189,22 +187,15 @@ def gen_layer_code(graph, sub_layers, sub_layers_name, different_attrs=list()): if layer.kernel.startswith("paddle.nn") and index == 0 and "functional" not in layer.kernel: continue if not output_name.startswith("x") or output_name in outputs \ - or layer.kernel == "prim.assert": + or layer.kernel == "prim.assert" or \ + layer.kernel == "prim.if" or layer.kernel == "prim.loop": continue - elif layer.kernel == "prim.if" or layer.kernel == "prim.loop": - if index != 0: - outputs.append(output_name) else: outputs.append(output_name) no_output_count = 0 for i, (layer_id, layer) in enumerate(sub_layers.items()): - if ("paddle.nn" in layer.kernel and "functional" not in layer.kernel) or \ - layer.kernel.startswith("custom_layer"): - line = "self.{}".format(layer.outputs[0]) - if layer.kernel.startswith("custom_layer"): - line += "= x2paddle_nn.{}(".format(layer.kernel.split(":")[-1]) - else: - line += " = {}(".format(layer.kernel) + if ("paddle.nn" in layer.kernel and "functional" not in layer.kernel): + line = "self.{} = {}(".format(layer.outputs[0], layer.kernel) for k, v in layer.attrs.items(): key_name = "{}_{}".format(layer.outputs[0], k) if key_name in different_attrs: @@ -298,10 +289,7 @@ def gen_layer_code(graph, sub_layers, sub_layers_name, different_attrs=list()): else: if v not in cur_outputs and v not in inputs: inputs.append(v) - if k == "args": - line += v - else: - line += "{}={}, ".format(k, v) + line += "{}={}, ".format(k, v) for k, v in layer.attrs.items(): key_name = "{}_{}".format(layer.outputs[0], k) if key_name in different_attrs: diff --git a/x2paddle/optimizer/code_optimizer/subgraphs_union.py b/x2paddle/optimizer/code_optimizer/subgraphs_union.py index 5f16e66..a1bde6e 100644 --- a/x2paddle/optimizer/code_optimizer/subgraphs_union.py +++ b/x2paddle/optimizer/code_optimizer/subgraphs_union.py @@ -50,25 +50,21 @@ def get_inputs_outputs(pd_graph, layers): for layer_id, layer in layers.items(): # 获取输出节点名字 if layer_id not in pd_graph.edges_out: - for index, output_name in enumerate(layer.outputs): + for output_name in layer.outputs: if not output_name.startswith("x") or output_name in outputs \ - or layer.kernel == "prim.assert": + or layer.kernel == "prim.assert" or \ + layer.kernel == "prim.if" or layer.kernel == "prim.loop": continue - elif layer.kernel == "prim.if" or layer.kernel == "prim.loop": - if index != 0: - outputs.append(output_name) elif output_name not in outputs: outputs.append(output_name) else: for out_layer_id in pd_graph.edges_out[layer_id]: if out_layer_id not in layer_ids: - for index, output_name in enumerate(layer.outputs): + for output_name in layer.outputs: if not output_name.startswith("x") or output_name in outputs \ - or layer.kernel == "prim.assert": + or layer.kernel == "prim.assert" or \ + layer.kernel == "prim.if" or layer.kernel == "prim.loop": continue - elif layer.kernel == "prim.if" or layer.kernel == "prim.loop": - if index != 0: - outputs.append(output_name) else: outputs.append(output_name) # 获取输入节点名字 diff --git a/x2paddle/optimizer/elimination/__pycache__/__init__.cpython-37.pyc b/x2paddle/optimizer/elimination/__pycache__/__init__.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..3cb420c5c56adbd9f844046dee3b74ccbff7d69d GIT binary patch literal 170 zcmZ?b<>g`kf(Kp~;z9Id5CH>>K!yVl7qb9~6oz01O-8?!3`HPe1o10aKR2&LzqmB7 zGBGbLF)!V~P(Q*bATcE+CpA7fKP45;)2}cp01NBq7nEe?W>%#Z>8IuZ8F`5%nfZD8 f@$s2?nI-Y@dIgoYIBatBQ%ZAE?Ld}&24V&PDZea= literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/elimination/dygraph/__pycache__/__init__.cpython-37.pyc b/x2paddle/optimizer/elimination/dygraph/__pycache__/__init__.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..877c1690d02becbbe2201dda9a4cc91b12b63c35 GIT binary patch literal 329 zcmZ?b<>g`kf(Kp~;-4`wFgylvU;xNxZ~)?B86c6ukiwY5kjogw$OvLH>N;Da7NxM|07bO;CgcK#_6&K_er@H23=4R$4mSpDV0i|VeNv8%R78mW<0a#c+zn}!@)T-1XeI$$ZQy~7)kB`sH%PfhH*DI*J W#bJ}1pHiBWYR3<>uNdSs9!3BQE@L(T literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/elimination/dygraph/__pycache__/transpose_eliminate_pass.cpython-37.pyc b/x2paddle/optimizer/elimination/dygraph/__pycache__/transpose_eliminate_pass.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..9bf3ecd3155befdd756e84c502f17040b839880c GIT binary patch literal 962 zcmaJ9E4=ro5Mzz9jGazL)dKCyAsL zscAte#z7U-VG;T`tfHC~j1tA(5*bN$O(auv8*XV4t9VO_k&NGybo2-AAt`myg|HUp z=ST`6jYKTkhBP8 zAj4}?L_Lgx^@ZJ|+bSu9;-=kzfU6w@#j*OF!u>dK4#8f5+56x+@{L}R?_LeyMrUbY z5S2?fFLl{*jzq1h7C(r&2Z_I*^ICVgUFo&Z3!#_Kp64G=FNBnp;`2tT{VKmaZT;`O zX*&>Pqf9=)0%da9tCrseI&0Tv1Qn-X7BLaJIk{7SY`{GC+u;Oq-Sd42i;ir`*WfAu z*7SeYxXq24RvFQ$G&BjWnQyFPi87b6|U<4`StjBZN~{F5X1y$Ojc0>(GsmfviV^#tE>|*qikBeQ*FCvdU|H6 zdv?c@9+2V)J4hgl6UW&Dk+^WkAqT_>Ar2f6ha93lph%onLJ6Y@8#q<;!97oqr9fH1N5OF zuF_WPD#bq8)!KSpZyR+()D<_kO(37JNrQF6AI?rXuR43}j>EEU zd(a&glWy+!crcB0^_JuBPm9{Ny#3+a6k2CD=(>G>I6rB-Y+J1Zp;L;x>vlSPB9ZS7 zOUZD**SENxb-|K`YrAgWVKo;m>(^}8Zrfe}df2AlzmX}|?z@50bq_Km=G=6cJuD=^ zZQJrDu&$f&ZEKM6mUocxe!rbf?(z<+YLV*qyEf+((UCOxdPtjOOmCBy$o18?rI9>RLIqBQ z)RK55l$iV+8A-j$4hbdjZDot&F&b79lV$4P!RD$QX~}k~PGGnFNNd^Lv)!u7%g`UG z;xI&Nqto|-$Pi;qhefL3cm>A9WEMxy?%9EvNie;B`y0^giwFnHU!Zwfq{F8kvXCM& z9~rwj5u%KdbgYk+V|ts8Qzl`zr7U#ZU^BL&|UV+#ND_ERX}~SUrS3ut9mG z9@DWNs<$B?^?m~Vp6G}1qW>N2Z_-d*Ax&ePgJ?x=OCv4R7-{ABhl7nE50bQG^x8;I zqZ!03E!0=Y3iO8h1=9NjYzn%iSWJbs4?*4~Ax}yWo;@W`19`?2s<9$=F4hQ9PH1ll zaG1c66P)c+IQjpDlj&Aqg_d+e`8%wbwwVQx^3D#)b~+b8q#l<-In3P#-SKM${#xF; zEBvFeuc8k<4kRS#_zcX%SDA#hmJXbJl`sVgPFJ^vxx~{wFUs%sy#{_#ESJv#fH(?O zi7x^l<^90ofq%;h_N$byVD~-loI_`HJZ2A`hh85;6roF0k%rGc$b~x)&#b_%U2r=M z%k^KVWiZ*&P=O+pc_{v0b@b&ieA${3cpyE|2Y(9bSjH26qF?_VoN+kip>(XZG``VW zKGw!c2qP&8)Id81&p9?8f!MGBH-STGqy)O~GkKg3mD{jcl!S7#z?7EEo4sY=iwt}b z{6WI6<{q}c*IN!shh(G*stKwKY6zNRxl#Uv4t@?gmd7)pvWQqguolFqPag{F!vrf`X>mLQHw`!sC)EG_y>deRe%iAloCt|; zJcK-nFP*|PCYW(w@8ca3EP~BSC(7U-0mPF-%CwepqVrmq7gQTBF%3Mb9G1Xi%DCTP z!@t80p$*HK4L`lthM?ykvRyr8Md+eC@8)&SWFz!)qj2YT$cZ%gBSPaI9W=snaxdiRf8K@ikI#Z zMkkg3`xFw!srw_L&ujeJeUUW&;@L=Ee2Q^3sh4Y0E1cFy<3CMsQhg^ePV>|FH(7V4 zt?+B1xA#`2A!K1-L7gvUEqBZM*V#IU=*cl&>dCOPaP!HZNdCnu;L*K{qNe-bg0C4y zp!i!@3Km0gCmX4Q=R*~4X6Hu>$9U6`!iCslQ$F@ssG%)%z8!0GqW88FF@gGI1{VY-W@%0`m!D=aHnXC>&qRy5{cW)_N7Q~*>;yX?JPlSK>BP2 z5gkPIVEEjdo;_&T-N0t+z7zCCcD=rDb-T95)^F`QuDzZlJC3)vo?YClk48$P(>;g` z2Jpynyr}R6Adw8YQ>1dp2UsM#win6%0i+Fnz#xl@)LR_#h)CP*@U|7i7cJEa0`5nO z>-a&Gv)P{Qo1K0T$#5@6fD-FvtZ`fv~w=jAH*w+^n-+?nYGhVU1pb4VpOM|gHw_O6O1E~3=ue@4*xz*Sxis$8puQ7L(SYuK)9 znv;|oa_kxC7mE_FednMia%zJX=M7!7RSU7vo{FJ=Q9;72tG;1KlNTzoU}IX#-Ydq6*MTp@X)=u`+ZF2rko0X6qA O&>OI^pf8+SL;f$?vnlfc literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/elimination/static/__init__.py b/x2paddle/optimizer/elimination/static/__init__.py new file mode 100644 index 0000000..caf34bc --- /dev/null +++ b/x2paddle/optimizer/elimination/static/__init__.py @@ -0,0 +1,16 @@ +# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved. +# +# Licensed under the Apache License, Version 2.0 (the "License" +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +from .transpose_elimination import StaticTransposeElimination +from .transpose_eliminate_pass import StaticTransposeEliminatePass \ No newline at end of file diff --git a/x2paddle/optimizer/elimination/static/__pycache__/__init__.cpython-37.pyc b/x2paddle/optimizer/elimination/static/__pycache__/__init__.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..7769b4260730cb68618f886de3a7a33c20d932e0 GIT binary patch literal 326 zcmZ?b<>g`kf(Kp~;-4@vFgylvU;xNxZ~)?B86c6ukiwY5kjogw$OvLH>N;Da7Nd=cAmSiS}6eZ>r7vvYGy5?l&X66Bf^7DYQGB{*Y0}_ji{WMu_ ziI%`L#-}257J(dcO9E3aHNF6-v4|Ncy^^7b4M>5BUs3wGc_sSArFoT!d1;Ax=>~@S z5k>)tDJeOr@yYoqsc@ctg;4=mSUfjN5Vhkt8>(HnaHs&O6;cnmY@la^sv?BMVO6OdP%f4gdzV!C!p=(FRXJ^c z2psu0oP9###9!dVjMJ=IX1*eFj zC8=meDaJt=R4ijYW@T9Mj8mfd2ckm7uZiNCZrFxqk&ZSb8>#4=B%{CZACgcPo=IzA zeskVYWxz8*m7H14IkNkW;9F17`;I4$M9V*OBk^iu~}B06{uS z0)yCGEJRTkoe;>@>T>>*m`9NKcac_gm)d2$lJ#8Hi&wAH&&Ov{sZxu)QF^;dFOOTl zoi=R;vaGd92Uws_+G9@d!#r(QW&{QI!7M@|bba)Y3ey2`?oY!7WV&Vh6c!!XkZ-|N z0G#Pv&Unm?npPXx>4cdW*L-1Q`^EA3(rV+`_YNPNxd$S~V z`-A$^t@_jcIr{b6xjn z*&R=MK#C*mL;_iyIL=-oap90d4u}&%95^5jIYfOx5u8>+5eI}&F7RIUXJ*&-X0cUo zURAw%?^V4|*Q*yQl_G(^Prm;b`)}VQ3Ebj4xJcVFR7XFKa$cJ^Ejbo9w4F3JAX6bV-VaS5ulp&XD+$|zIbu|PGIY44B? zEfKz<9}taf7$`bkyeb^wHLr0e2s`c2eaZJ)UI0Vwph;7{_*;R7a>CaE2qI`i_LUt4 zCH$Pi20;RHXC|{)ffZSam05*V+0<4gBDTtE>GwR=@X;^AhU>d6HxQtkYlpp?gQ;|3tLF=^ z&FdeIKm-U)%N&<%0Lf7Af--2y7eJ2cU``6^Nt<_c_R>NMf=LDNh zu=Ta4RUBE7O^2)2oSpt%}Uw?TN`oXOCkNm`6XK z13>7sVo-x>@LyInT7;(2!52@DnMSttrh@C?Pltwq^Brj61ptf=mHw9_x zSbRenszWW(Ae@w4h1Vj5sn3$3(ygtNNP)o6);PxCpq3gm*ZwXJ*VWia4^#Jq+X`c2 z*X4oh*L7Zn`B;~s6zk1)FA%XMX>E_iy2N-5XcBM+sn2Y=!p;4$K2q|{48Gs}dK?~KS?APX=XnHNd-BQPoGmhoabjJ*%?UY7Edl*2Q}EFW~fj-dsRu^aAGgpM%*D^An+SR8t0De}D(KA(uJA zZCv!*O~((PZ{%F^ouT|Pv>9mOUrlu75quPP6mUR#WcL3Q(UFSrePrJF9Yi?9c%&Q} zy9QtB&L0{hEdokPgf5IjaGXQy5y%X)aEI7ehMF*?o2jE>q}_qDq9jsxN=(~T`A&Bp z@G=8e1b0yIgS&;}Z+GWKWuFXn2@MHN2`vc=tS~Gd(f%($$LeS*QntW_K|6Y5Ff5_o zQ3;$IWhnf!yr0Rz&$=Sze9DT$;uR9fCG>egO%KWsP-#kQ5 zkTUn_u$;05hw$t)x3N9%S4o5Wn{rz$x^CazA z83ID#D-R)0;d94$T4SEcT=&CuBF=)%Do0xXp90d!K4r$Pc4YEKRFu#d%`pQ!sv1?m zW2$)GV8egF384+ExeY(Q--e*)AM#T@W<{9&Fx4-&oIe!bH%{jpe3I*e+M+M0{L_cx`a#O|xG$*TpHG)K z_KJXwWlOKn_~nO?tnrlxaL@fCFcK;}xpi`+^#2Ib&P~``x$(iUCN-#H%%)MN^PRol z=A-khvOtDsCi*}-oKCf@Ef9#i#F~IZ+$GjjBILM$Yr)qw{x9&#GvJlhE0+oWeR* zQ!w(And{`oBv7uCQ}!Tp7p%wAQX>$!$dz<~u#K9`T&E9>8<>l~pcC;T!`hLeShPw_|zFI@qT?p~5T zGyDd8%`gKc-@=MG8_6r#Q13q%>2NW-Fq}Qai;fb_CLWvNiN`_@ZDI0ji8e=O_gc#N zm)-e3@CF{W;cP@kk3_T5#%A|daDmZBV0Y{vYJiy%QvkChrUF(#j1{gD@V_G5N@{C& z8GP#R;=O*bTzNIEE|&AkdHHfug*6&)4WBrERgX2eQ?%sm=EO5QQ@-yVioG9Sm#g;uvqooAXdXYC>lb+pn{9_+Z?Kh*w}3ImLrmT zmhK3_!&vjZP{ak7ZMmV{?ul51dr@p8+HkxL@e&F!VXo5?A;(t(;0{D;r`?IIWTh9x zWiJ2%JA|FMAX@gqxbTwPE0p(ffw_KC6UK(y7qHv<+1RjUC6jCgt2BA#i%mG$CKSGL zE#H4uan_(b?KoR-mrF+yl`6?0XJe>&Tx&XkebcqGx)II>w^(ZTP*Zt@jH|aiu!74n z_Qht8bC8VHR5{snN~wAs5u5#Fhq1YrBGiqkUoY@SQF$ye<8!$@C}h#{8Ska>#0h2d z#Ikz|=y-9w?AU2kL)c@GtDjVdQ=%=L^sKSknl=re#K~~m`NnC-QD$_uNvNULk zE=Y~qB)h`iY>wlelkxlY$fz?d4A9e{3Y=kC%Eb9>qz0qOzd;Az`hQB6`)n6$PTvb< z{^5^<1^E{cU`2KJzwnp8?L7Ed31`=esK7gw z4PTk&Hhg*T%erwkCv1WnH6VRM%5QP`#l&a$c=}#~p74Ard93I}0<$hAdww1@4>8SK MaIT;)#H^+M4>y=cL;wH) literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/elimination/static/transpose_eliminate_pass.py b/x2paddle/optimizer/elimination/static/transpose_eliminate_pass.py new file mode 100644 index 0000000..3e43a77 --- /dev/null +++ b/x2paddle/optimizer/elimination/static/transpose_eliminate_pass.py @@ -0,0 +1,33 @@ +# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved. +# +# Licensed under the Apache License, Version 2.0 (the "License" +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +from x2paddle.optimizer.pass_ import Pass +from x2paddle.optimizer.elimination.static import StaticTransposeElimination +from x2paddle.optimizer.pass_manager import pass_register + + +@pass_register +class StaticTransposeEliminatePass(Pass): + name = "static_transpose_eliminate_pass" + + def __init__(self): + Pass.__init__(self) + + def apply(self, graph): + fuser = StaticTransposeElimination() + fuser.operate(graph) + + +# 用于注册 +static_transpose_eliminate_pass = StaticTransposeEliminatePass() \ No newline at end of file diff --git a/x2paddle/optimizer/elimination/static/transpose_elimination.py b/x2paddle/optimizer/elimination/static/transpose_elimination.py new file mode 100644 index 0000000..66819f7 --- /dev/null +++ b/x2paddle/optimizer/elimination/static/transpose_elimination.py @@ -0,0 +1,291 @@ +# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved. +# +# Licensed under the Apache License, Version 2.0 (the "License" +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import copy +import sys +import numpy as np +from x2paddle.optimizer.pattern_matcher import FuseBase +from x2paddle.core.program import PaddleGraph, PaddleLayer +from x2paddle.core.util import * + + +class StaticTransposeElimination(FuseBase): + def __init__(self): + super(StaticTransposeElimination, self).__init__(graph_type="static") + self.direct_layers = [ + 'paddle.nn.functional.relu', 'paddle.nn.functional.relu6', 'paddle.abs', + 'paddle.nn.functional.sigmoid', 'paddle.exp', 'paddle.rsqrt', + 'paddle.nn.functional.swish', 'paddle.tanh', + 'paddle.nn.functional.softplus', 'paddle.nn.functional.leaky_relu', + 'paddle.floor', 'paddle.erf', 'paddle.square' + ] + self.elementwise_layers = [ + 'paddle.add', 'fluid.layers.elementwise_sub', + 'paddle.multiply', 'paddle.divide' + ] + self.reduce_layers = [ + 'paddle.mean', 'paddle.all', + 'paddle.max', 'paddle.any', + 'paddle.sum', 'paddle.prod' + ] + + def get_transpose_num(self, graph): + count = 0 + for layer_id, layer in graph.layers.items(): + if layer.kernel == "paddle.transpose": + count += 1 + return count + + def operate(self, graph): + total_layer_num = len(graph.layers) + scanned_layers = set() + optimized_transpose_layers = list() + optimized_reduce_layers = list() + optimized_concat_layers = list() + optimized_elementwise_layers = list() + + def get_index(layer): + if layer.kernel.startswith("paddle.nn") and "functional" not in layer.kernel: + return 1 + else: + return 0 + + def strip_transpose(_graph): + layers = copy.deepcopy(_graph.layers) + for layer_id, layer in layers.items(): + if layer_id in scanned_layers: + continue + scanned_layers.add(layer_id) + percent = round(len(scanned_layers) / total_layer_num * 100, 2) + sys.stderr.write("\rOptimize Transpose Layers...{}%".format( + percent)) + + if layer.kernel != "paddle.transpose": + continue + if layer.attrs["perm"] != [0, 2, 3, 1]: + continue + transpose_layers = list() + propagate_layers = list() + reduce_layers = list() + concat_layers = list() + # 此elementwise_layers专用于存储shape(4) + shape(1)的形式layer + elementwise_layers = list() + can_be_optimized = True + for out in _graph.edges_out.get(layer_id, []): + if _graph.layers[out].kernel == "paddle.transpose": + if _graph.layers[out].attrs["perm"] != [0, 3, 1, 2]: + can_be_optimized = False + break + transpose_layers.append(out) + elif _graph.layers[out].kernel in self.elementwise_layers: + propagate_layers.append(out) + elif _graph.layers[out].kernel in self.direct_layers: + ouput_index = get_index(_graph.layers[out]) + if _graph.layers[out].outputs[ouput_index] in _graph.outputs: + can_be_optimized = False + break + propagate_layers.append(out) + elif _graph.layers[out].kernel in self.reduce_layers: + ouput_index = get_index(_graph.layers[out]) + if _graph.layers[out].outputs[ouput_index] in _graph.outputs: + can_be_optimized = False + break + if not _graph.layers[out].attrs.get('keepdim', False): + can_be_optimized = False + break + propagate_layers.append(out) + reduce_layers.append(out) + elif _graph.layers[out].kernel == "paddle.concat": + ouput_index = get_index(_graph.layers[out]) + if _graph.layers[out].outputs[ouput_index] in _graph.outputs: + can_be_optimized = False + break + propagate_layers.append(out) + concat_layers.append(out) + else: + can_be_optimized = False + break + + visited_layers = set() + while len(propagate_layers) > 0 and can_be_optimized: + current_id = propagate_layers.pop(0) + visited_layers.add(current_id) + for out in _graph.edges_out.get(current_id, []): + if _graph.layers[ + out].kernel == "paddle.transpose": + if _graph.layers[out].attrs["perm"] != [0, 3, 1, 2]: + can_be_optimized = False + break + transpose_layers.append(out) + elif _graph.layers[ + out].kernel in self.elementwise_layers: + output_index = get_index(_graph.layers[out]) + if _graph.layers[out].outputs[output_index] in _graph.outputs: + can_be_optimized = False + break + if out not in visited_layers: + propagate_layers.append(out) + elif _graph.layers[out].kernel in self.direct_layers: + output_index = get_index(_graph.layers[out]) + if _graph.layers[out].outputs[output_index] in _graph.outputs: + can_be_optimized = False + break + if out not in visited_layers: + propagate_layers.append(out) + elif _graph.layers[out].kernel in self.reduce_layers: + output_index = get_index(_graph.layers[out]) + if _graph.layers[out].outputs[output_index] in _graph.outputs: + can_be_optimized = False + break + if not _graph.layers[out].attrs.get('keepdim', + False): + can_be_optimized = False + break + if out not in visited_layers: + propagate_layers.append(out) + reduce_layers.append(out) + elif _graph.layers[out].kernel == "paddle.concat": + output_index = get_index(_graph.layers[out]) + if _graph.layers[out].outputs[output_index] in _graph.outputs: + can_be_optimized = False + break + if out not in visited_layers: + propagate_layers.append(out) + concat_layers.append(out) + else: + can_be_optimized = False + break + for ipt in _graph.edges_in.get(current_id, []): + if _graph.layers[ + current_id].kernel in self.elementwise_layers: + try: + x_shape = _graph.layers[ + current_id].input_shapes['x'] + y_shape = _graph.layers[ + current_id].input_shapes['y'] + output_index = get_index(_graph.layers[ipt]) + if _graph.layers[ipt].outputs[ + output_index] == _graph.layers[current_id].inputs[ + 'x']: + if len(x_shape) <= 1: + elementwise_layers.append(current_id) + continue + elif _graph.layers[ipt].outputs[ + output_index] == _graph.layers[current_id].inputs[ + 'y']: + if len(y_shape) <= 1: + elementwise_layers.append(current_id) + continue + else: + raise Exception( + "Unexcepted situation happend while optimizing transpose" + ) + except Exception as e: + can_be_optimized = False + break + output_index = get_index(_graph.layers[ipt]) + if _graph.layers[ + ipt].kernel == "paddle.transpose": + if _graph.layers[ipt].attrs["perm"] != [0, 2, 3, 1]: + can_be_optimized = False + break + if ipt not in visited_layers: + transpose_layers.append(ipt) + elif _graph.layers[ + ipt].kernel in self.elementwise_layers: + if _graph.layers[ipt].outputs[output_index] in _graph.outputs: + can_be_optimized = False + break + if ipt not in visited_layers: + propagate_layers.append(ipt) + elif _graph.layers[ipt].kernel in self.direct_layers: + if _graph.layers[ipt].outputs[output_index] in _graph.outputs: + can_be_optimized = False + break + if ipt not in visited_layers: + propagate_layers.append(ipt) + elif _graph.layers[ipt].kernel in self.reduce_layers: + if _graph.layers[ipt].outputs[output_index] in _graph.outputs: + can_be_optimized = False + break + if not _graph.layers[ipt].attrs.get('keepdim', + False): + can_be_optimized = False + break + if ipt not in visited_layers: + propagate_layers.append(ipt) + reduce_layers.append(ipt) + elif _graph.layers[ipt].kernel == "paddle.concat": + if _graph.layers[ipt].outputs[output_index] in _graph.outputs: + can_be_optimized = False + break + if ipt not in visited_layers: + propagate_layers.append(ipt) + concat_layers.append(ipt) + else: + can_be_optimized = False + break + if not can_be_optimized: + break + if not can_be_optimized: + continue + + transpose_layers.append(layer_id) + transpose_layers = list(set(transpose_layers)) + for l in transpose_layers: + output_index = get_index(graph.layers[l]) + if graph.layers[l].outputs[output_index] in graph.outputs: + can_be_optimized = False + break + if not can_be_optimized: + continue + + for l in transpose_layers: + _graph.del_layer(l) + + optimized_transpose_layers.extend(transpose_layers) + optimized_reduce_layers.extend(reduce_layers) + optimized_concat_layers.extend(concat_layers) + optimized_elementwise_layers.extend(elementwise_layers) + return True + return False + + before_transpose_num = self.get_transpose_num(graph) + opt_graph = copy.deepcopy(graph) + total_layer_num = len(opt_graph.layers) + + while strip_transpose(opt_graph): + pass + + for layer_id in list(set(optimized_transpose_layers)): + graph.del_layer(layer_id) + for layer_id in list(set(optimized_reduce_layers)): + dim = graph.layers[layer_id].attrs.get('axis', None) + if dim is not None: + for i in range(len(dim)): + dim[i] = [0, 2, 3, 1][dim[i]] + graph.layers[layer_id].attrs['axis'] = dim + for layer_id in list(set(optimized_concat_layers)): + axis = graph.layers[layer_id].attrs.get('axis', 0) + graph.layers[layer_id].attrs['axis'] = [0, 2, 3, 1][axis] + for layer_id in list(set(optimized_elementwise_layers)): + axis = graph.layers[layer_id].attrs.get('axis', -1) + graph.layers[layer_id].attrs['axis'] = [0, 2, 3, 1][axis] + if graph.layers[layer_id].kernel == "paddle.add": + graph.layers[layer_id].kernel = "fluid.layers.elementwise_add" + + current_transpose_num = self.get_transpose_num(graph) + print( + "\nTranspose layers optimized, before: transpose_num={}, after: transpose_num={}". + format(before_transpose_num, current_transpose_num)) diff --git a/x2paddle/optimizer/fusion/__pycache__/__init__.cpython-37.pyc b/x2paddle/optimizer/fusion/__pycache__/__init__.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..2d4cee7661d6a69b3023338ce8d015e5ec2cf916 GIT binary patch literal 165 zcmZ?b<>g`kf(Kp~;z9Id5CH>>K!yVl7qb9~6oz01O-8?!3`HPe1o6vHKR2&LzqmB7 zGBGbLF)!V~P(Q*bATcE+CpA7fKP45;)2}cp01NBq7nEe?W>%#Z>8F(zXXfYW$H!;p bWtPOp>lIYq;;_lhPbtkwwFBAn8HgDGJ)0_k literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/fusion/dygraph/__pycache__/__init__.cpython-37.pyc b/x2paddle/optimizer/fusion/dygraph/__pycache__/__init__.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..49182bcc70bfee1ff8cadad8634a324db2e43c14 GIT binary patch literal 1715 zcmb7E$xhoq7`6!^35k<9iL+65C`~Dt=3G^UBBWLoihx$tTr6WdVCf&*vK>@-qrOeg zoGSIySLmrT;~8fGD%DY(Z@y*LKXargixPQ$e*9tF;{tmf?9=z&Vs-I171PKm~>uU=bHlk>MPau!0nZ^H9bWRAIOP zRjeYF;YFz7I;t~Vga+1-#&8LmxP@8_E6~Or)M2;`UA%;r7_PuGyo{C^uEGjlMXL;} zu!h&sI>R;Cz?*23;W}*LZM4mB1D@j@w8O9lFK`d_7;eH#yo+`jZowuL zPH^X3>UDpUosxR;efRkGDl~)ZL)#2u_r@9cKJ@LAQRIX~U4Bv>m{CLwEp9k6W9$0V z5AirFdY+MZur7k9=avbM7_j+}=79!R-uqq@n_esuVg{2uk7(n_P5=1NPM6hwOqiM2 zap(vBC>=&jY=Y*2GVkof5>ryA7;anTm$bw@R$@FiPkC*YPcm7ml`lObqUFT{3zV!)k%^mJ zD8y~?(&Fwsb5D_pyFi615pMdnVcItL)B6cEX6PhQq>^GMCB^L}%BhoF;#DZ0A6nd2 zrp*#>CXNcFy6(7;j7wfgJX58Hl_w2JFDf+^gV2EykJTts88eBmPFW>P3Wz%!lv5|U z#H&%h77vZ|PAKs5ysu4>iMvCD?^3u#e%R~J7ae=C9*w+P z(;J%J)xm51Ykxq_g`B(P+fLfjzx4w~>pr>buzTl(I;lX{_jEhC1$D!4JvTOt{owXK V{|5W^2%NWczoUKfD#`!z`3s7{5}W`4 literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/fusion/dygraph/__pycache__/adaptive_pool2d_fuse_pass.cpython-37.pyc b/x2paddle/optimizer/fusion/dygraph/__pycache__/adaptive_pool2d_fuse_pass.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..5a95db551382e8b001eb8f1683a3a1ef1dcd34dc GIT binary patch literal 1007 zcmaJ-#;3z!pTPL__NO=x3HgnS-H0GK0kfyz6j8Jw zCFPW2%nDWpJn(T)gk{7dN_6yws8B_hL`9lzf(_-djyHt&RQ!&ly+4qLq|}8|X)Vlm zU#%8KHfJxCY})*@p4N3SR`1l~tSBWwnykTvm%GPqX*qsZ^@aVpc1CwaU)N zjsKn1AWxaEwaEZcURT)`aCQ*zs9BjFR67E*h=DRI#rN`wy&xi3{7hsex+0lN|ss(*B7F!)e;mHZXm>`r7X5PA;y2E0Q4j3w_?F0`Vn$>U&<=@60SIk+Ka_*xA{0 zn={{>Gjm24$H#LDe*gK&znxznSCp^lqyIC|`2n8jM`)PBR9~s8;;mIRyft5M7*#_h zp6+KFX4Pz1RZDyuzWsCMlEN~~ysa?PH8%BXmZ@(kMXLvE%H^V(*gtGV?!{W<7PVyT zN{un!y~Jzbwf@IXYwIqDoO-fOD*d0HcA^DzT;WnxdtX^qnaZ@=c2#FOGj1zYLk?ce zFpJsvHs4narE0;2Y;x?4^)(vm0;`3wchkMnZ2I$z{NZ&qwNGyvi=OBuG_ew^U1d|d zuF*(as;b2LrqWd$gDJ48UrsD0&BE!Nu z*kLV>T^>ZxG9uSs#XIXbb-xxxj`KI=oBB*6h-ac!uwDyRYr)#=+{`cLC0ex&zov;&%89QLRycDi;3WLKmQ(ARJol-lmJQ<@> ze+mPzvQV<{ypAWj3CP4^C>v^w*kHz%x}kNo>nbe?vplt{GJ8{9Q(1OPR~5m@F;h|n zcWi{4XXD_4D!3DZdw@-DA^N)7w#uFp+Mo;m6!CYYWi8fWU)wfzlueaAe^fcDZ0NDl zr7(G6OWQEI2K+kcDX=f}*dcaU*wwc)U1Nuqc~?5qC2EGK!s{bK@92oP$A;!|e1t1! z3aaq+#S!ii_APhxazR=&4q4C0;u5A#I}4x7HKp;z1^Be~{vlbu|KWuRpdTI{vhVyDQn4O<|$ z#4Krvm4uBMX#;)xkJ<+kUz?qzXkoLIY+dDgc01SAuH0I_- zRDmkE2ZZkar~+N^&k+A9NDfEU>}86kV%@?z9(#t$3wfo-K~Zor9e32Gh_OVO3_qYM0Ly zmRAb%Lt5Typ$+b^FXRued68TA8MSY=>n=9MW-xtX&^<9a!WoQk_3a_s?aG1}cJb=z z`K7}A+(Knx@l1K4vQ(HG6+GK_y|OTBxJ-uUPr=QzqgoZP8XhKGDGzI{HhIBmSI!l@ z022w;T!>VP`zP66Bw((VPZi2gQ@gpA9vHg|cS?kj6XBFuf&KK8&byTpkRQQz2-7r7KK}$K{W3{eQ;>@D~z}3(9EM@I4)6PsE4VZ&@Lxx8?}HQ#Z(1p9^(ffBb!O|5_`=J zoS@cl&yfKQjiPpzo^ouz)AD@gr1eH=*>BcsessRnH^b=!FRI+etCQQ!)+8g+wUejm zoEH7lvpw=2m76KPqr_^qV&rtxv4tXXxJTtq$4Wo>bNK3!=^lAn)H#(K{0N#vT~Ecl z)gDz6z5n3{+xf7@-5_>+ssyE56T+O^&*mgM$mYEk`9WCeWaZE(vU*hB^rz%|b+<=w z_a@L(1a9g)^07w<_boJ=VNbX}MZ~-?yXMAT>^2A+=C*s!6K70tWTq80J_Z_} zM$>zNxK@f>k|i8PX_)9pm2ak>HTz(bCYBnY)a%hVmP}wtE0O}a7Wb%b?NK$?IlgQE zLR$Y_xVd(PD!@0A3^_B`%a9#y;A6itheDm7Cf*n)gGm+zrd*H26w4BcX(q-EcRiBa zcMw-Mghz>q*!6<7BqJh^6X=PV;*;g$_~ZmlVx=)zwY^noy*iPS9AR#DzRn4Q(lo=mN0o-_$=IB}$@Vtn>BWhkyHPyyn4qrL- zkf7uSavEq-XA)&|UqjN!9`z=KF4-sPJZw#>dHqQvl7y6%qe{OfVJok8rk^#x>!Om! z+$asH|2m#%2~AhMuAm-Qw$-lI)pt+>$9Qk+qC!?tAEWMtjM2@Yf*aJk;*z6X<|6OS zB!}F@)`Oc%l#GiV#Qjy8s3Pvm7>J)mljvbHESh4qlZ?0)@fD(5japp4<|1l>zG^i; z=sNrNyGbvgjn^5-`~hYnae)=sOaj}^Yfr&vRKHFkO+T_h86Y6~7#-}xxKW>=`MSj)7`Z5%zH>rtd3}CCFO=P zCO2=T#^Y|4M3;`A^$Ph)tCF1`qJQ1 zU+QmnGJ!a!1GRc^AXsm5w-oXwa;+hDZMim6!vd|?^TnXMUdo=VEtci1&lBt-$7nCg JW8bp%{{tIWpB?}J literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/fusion/dygraph/__pycache__/batchnorm2d_fuse_pass.cpython-37.pyc b/x2paddle/optimizer/fusion/dygraph/__pycache__/batchnorm2d_fuse_pass.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..803dd93e7aef4767aa32c44f2e3d34cc26649001 GIT binary patch literal 983 zcmaJfo17>X9L;#Vp3J=W#=dM@?~zYNlV5*556(wD&%k%6K+Z%5Nmg*%cCWcY@}gI`dG#MJud!Wh`^ zzF04{Xf9@=%`WD(E~oP4$|wz2;|5Z=R?FP9N@w`T!(tGefteF4Z5cGUVj(FU?0d10e^=ODhGit00YPV8HEnGaQDETgPEJ)TJnW%$ycWZFr)LB zX~bnC&hsj7IY*jC6^m~~-++tXSyEPQVpi2!R0~lppFB<8PS1suh2mK))%7X)IBneL zq;A^0%r{CWfGDr4WDhp!1v_chdH|h{!3<&{bo2P%)F%g^nahQABnVHSBZ*a7vLm0^ zmc6H2Dyd|j8IBw1dUNZybyEY!VTr==4|yd)0xV$%b)&RsRqW{q&w-n<4Nz9?#ONG8 zJR69nDc1KPG}xQakZw-?gW9Ff;}C_eM5#EpL(a=uu0Tg_N1R`-M6ti|ad)EV+fe(j z-Z&DC05m@GBQO(!iA=`BQ$Tt*ar-iSt{KLP#$kpKVy literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/fusion/dygraph/__pycache__/batchnorm2d_fuser.cpython-37.pyc b/x2paddle/optimizer/fusion/dygraph/__pycache__/batchnorm2d_fuser.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..6367d242227dee5cff6e930cfac3a0091ddb3794 GIT binary patch literal 4166 zcmb7H&2QYs73Vj(>y@ndBaT!zS*K0yb?W_~Y|C=(G;k6JC<>uyawxVah}8}yQ{i%% zAs5vy)<6;o$e~btD0*pvKu&|8y#z^*O_2NxJ@Zr`u@wFby|nMWAy-RD4q7faoSFCi zoA+jC?##@TC3ybzi+?%`^OE!p4)RBX&WrFy*f5F7zSNfCtN2P!ZL2cQslL|J+j`Gv z8#LB@^L6P}iRsMPmYCtHTWZ^4@@rDfOhJpZQj-(w?jUmSbRxH=Bqx5=Va#`5<(=?e zKKNy4=yKqbFLzNSf7GlKU4@QIuC$eRr8Sw!Oxd>DDpQ%ZEwwdaysa~nSui%o$zz!rTk^Wh%pJ&y zjAk)S%s+-+V&y6N36ig{nH^1N*p=BVX#fgoIEfq&bNx!Jk{xANdnj$mY;Infm+mXE zK33Qh?9`5OUmdI9;b~6-D`!aiS@tApS9XoD`ViJ}id!4wtcJ6h|Fa}nA<1*>sobxP z<1Oo;OD|YHz-FFw2yD=&1@;5*31%s-^F%e+56J@31#&t4=|j@b5NlBP#7a7vWZ}dVVj|CAp6IP94~|E}0yL}w8Y0VtWS(VK?!y#!X^gWa@=S+b zo`relXk(nEEM$;R5RDDdh+12m)Y?_T>>Bu~uu^QP8+M||XkGeUy2}N7o zd+^}9uRlvac%EJK+3QYgG>H(RahZTvQu>(RYV=pIY= z-QW`O){c;EAr5LT+^$|Nki9iBU{{Mq(J=3Lkz4&Wwy$h&vW!#~0bG)#`6Ft_A$coa}p5_uE;*3~G+Q;Xs;J_9_WnT1|rt-FB}bb79h3bOy< zrN{jbuMRw)IbkP`T^`hL`TZ{B=?nE-$EX}~uO|vC(Y+uX#G~?RaD{qTnjD>lgOx$! zC^7qk7+9i_MPiVSBa?c8%TEGpiu@@a^=c|V10Q}CnnWIEY~EUm4yXTthnvWzB$cqk z-5_>+Jg9LCCP%ZR)6MHH#q(2AniSu(ZhjhQcWW|zQyrx%B)#s&UhMV~&G(`>#goUh z7{yUr~zkD zcR7O?p7$^etUd`4o&)?w zmmY`5VbLz8Ek0QM*e6KNLkG51WmiTR-~m{7uXDJMJ5Ew|oL-*|eC$^o2e%$U zp$ivZ0^R)Q*r2^+CcE-=UEg<{gOY%X$8Cq>20|A`+#1zabBasrAof;}(v&R`bGiO6T~~ZZHT=!OTN&ieyxhj7ge#l1cB9WPT5y z1wEj>yN`g__MYsK`Ma7M#O5;q=LnorNR0z$59~RZxel%)U+J2Ba{>Szn#N3HK{n#N zsEdws8fIzCKzva2w6_uwirgNFC)FhfT+KQ`1JqMH6?%uy%m$)u z%jJCt4Pp}-($$0iRP3&%afs^HqEejOA?H;i7a$(DBhEiAM7cfjadu+pn~?V(T0{OG`O-=4{~J-rym2kyBvm8iuW^gz}Q c)ZW_f=b@@?^t7k^>;{&OL`D=oZ$zWfUtd1=i~s-t literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/fusion/dygraph/__pycache__/bn_scale_fuser.cpython-37.pyc b/x2paddle/optimizer/fusion/dygraph/__pycache__/bn_scale_fuser.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d6bfb59f13738c62e1e044adb4ed8e41bffd683f GIT binary patch literal 2948 zcmai0UvC>l5Wl_uKHF(Y8>*s8q@{woYV9QOfKXGRP*H(Ugj61og|xbOFUf`T`Pe(t z*qo)diPQ(w_JIc^T2&4p1TUz(0&2ek>f=5jq_#=(6?kH1@9e~BO7FCLv%531vorIX z+qrVtQQ-OK!@uatn4&zw!Qjz>co$lX9a9(ylsbX08mLXJt`VHm0=;R}jiy;Q*sG|b__51a;9cZyv^*Gmc+!-*dR}x(-NP2X|zoE|tB67PD~~z(Wwg z!|jI<>&eYqx1K(@w|(cz)1R(x-hQxo%U#w{CVTapSPk;vfK)v{Caom=Do*%{IJ!(y7lAlkMDo|&QJ=4AwrE-_~rEE z>?9^bb*O3Tl$@`OaA(1k3$zir+-(j8_gQ+bXtN!_ep&KVVUYMy&^fg_J22+VWM_6C zeP<@!RX?7cevwTxgTp+lW9E4_fyiRe_E|MRoX1r!@S0wjT=rv+LXv}VyOn{>``bA4 ztc}AY9w7M=CUHq7aRLdqMvfqU)pU#>>Eybx;HWxl3M1L$~R@h6-~0 zytyBK7TSKmXyhh|$HVHmpamF=&sPWX-HF`wVOX8R%$mo{x*!ZcjM_=pSqRTc>g*Ip zo1q|@avTe*)lPsV?oP;LX>iX?JlF^l!GpcFk)0nvV@aobU{8Rz(d2D}a`)K(`Fa~m zeAmiNs1SY_#`93dto%fx01=@wP{NRBt!TUmOne+VQOt~`%x@%neOMlG%~4>F--b?c zY|SJpG07M-72YwTlM=jx(cn8e+dck5_c!n~_+a%#TGlZ9Tck=~f!d&Ls68D@y4F`W zAQ>1rs`QLhPsxgw8c>n6zMdKzBro&od=?W!lw~-*%ef=;z>gE5f9`c+VP;7zw5SzT zOz9e7$UlN`hQWepHZk~KEDGUJO=(EdXEX4iA3_;V=#Xa)juG3?Dd4Ju#39|2`#EH< zSOb#)AIxj?i}cCgKmZ`rK0!b9H1rL61N@<<`ihEPS&=h66DYlBrB>fyD*8rJnxs%l z&!x;If-*;xo-MT*Lv5L!LQ48ZTG${xCw1_j&^J@K7#vkL2zVYfS$iakQlbbFrL+VR z_HK#9$}DDuMKv`*lXFK+t#8$vq*qSM0Pu`Af?Q1%z@@Sn%CglKd5L0mt8C zxA0Z3nlW+A(yj$#>oG+)_&8U+t zOz&C%8W;E}G+?ym@kyBb09rf+9VQtf(vi)PqY{i9G7fb1MGWf_$u`*^og=WR*mxnr z!%!+5N}Da#hU*FyB}%_&yFs=hZ!P>)uz;V&ZW20C&M!?3F9;~zF&qjO7I%)hEw5Y{ z6A2QVlr+0yTXGieeTF=@P{T-^7~G|*!@E>9uXZ#;Dd$)|>k~u9MvHsZh_?XOO^$$) h(G=sO4DG}Zoe|IhP&yJl5=ijR0g@`wum599;ka literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/fusion/dygraph/__pycache__/constant_fuse_pass.cpython-37.pyc b/x2paddle/optimizer/fusion/dygraph/__pycache__/constant_fuse_pass.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..a699d7b9774b5433134b3fa1894ea0c44ce5ca00 GIT binary patch literal 962 zcmZ`%O>fgc5ZwyeA}1`~^8`{8->=I!xlG$1g3z4FO;n`fOQK>;*L+R$p5|+k_f_v5N&9~w4@s$yPNj1& z@4Z~itZdFtYU5gI+E;U@E$ojPI2TsW3fF2o#lNJnT)<#P4qv7+)zc?jB;0_{omrPN4BdpspK_cianUDrw{l8iKR~ csK0kz(Op&hrW=)iMV+=$xL1uCh1zCe6ixH6}sIbyRi%xg#rgwU}d#iim z%+L#)>_QZY3HW9t?5c>qNb+hV`2#-Wxt@e1yV<|si=I=`Knj^_P-(x{P5Egz-ES=LW@LziaSO}G#ia7&x9 zAPO?Z;FOW_Z~^{h#=Ou?Q^tNJf1hu4qO6s6qrMw0xY6RV9m6i6^H_W5|$>YYdw zK&?2*f=)2tqP5UXgE(pp6=``xlNhXX(MP39RB7ru!zKV|DKUQ+@0D^(Xgl?|i@dhtKdvYJ>|z`}{_WQGH@6?%{c(Ha z^7c=^Kf3qz+hZ2w?47K?Q$K!eVr?N7b=G_9SUrg90)XX5dM6u`Y{=SdHpgs(S|Wl@ zbEJgSddGJVIPj?zL`gSGWhw4viZCy^VVBP}3|UItEaM`QWpED*F=C~8p&Nw0m_}wQ z_P|fn#=~BqWKEr5L*9e1l-e+ml0~Ozg$@oq)l(Bp9%dyr^s)(E1??&jyQ%OM1aPiz z&`q)iZaYX;3^fO*4d-TVEzIZqOTaa zk<+#&s%l5e^-T)VrNKE~IlAV8`HOgWi(9#St-btmN#ECQ^W<)BjHUth2rF9{Uy*A8M_@vV2 z{vuDIbVz-XXAN7{)CN?9DB@n`R61_vU1X8l;m)3dR`DolThdDqekjhVTqwek1-?@n z(olbDf+wNtai&;`z{HpohhcaPdRm8p(EZe=Rft7PQTVUGSB2R$-Ae~EPmRzZg3^<* zh<{Ps-JAhc+%P@WoV*p$5f~F-oDl|TCXF`l0|H8^RHcr15%y)dzycCa)RE~GWU8p7 zlBO!WvXZz0pnxod3{`M&z_5TU@TI9JXJ4^UWbhEXVoBZUR7VU9?|K#Q!lZ=+Mqq?W z+o}KBhg&J;Iic0*DV4hwz(?iq;s1pYyh-p%peq@Z6~;QT-wlzjGPc}x!(vC3a`8MW z(t<(OhL4EH!;rDv>H^wU@sM&T=G_74DH7mJ-L9%KR~F#Sg=6 HirV_Wc&1)4 literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/fusion/dygraph/__pycache__/conv2d_add_fuse_pass.cpython-37.pyc b/x2paddle/optimizer/fusion/dygraph/__pycache__/conv2d_add_fuse_pass.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..5dab9a1c02c14edf2e7ac40027602538ecbc1bbe GIT binary patch literal 955 zcmZ`%J&)5s5Zw>kF;M~(AVmTJSGtKIHA2VT2_%Z>BpT#qwPZIrgYEcs9Ymr~8g&t&phr3slUklM{t+qnVxcPUgKgvnRu0pTPR{=BJPWA;0l(SRn+*VDA6{zrvs8G`#-_b16(T-$26}=@%?+?@=33b7_v=;XJ zXX}NL^~LG3`ZPIvsnn~L)dntxH6#e57rAY;nc?3Ki$QPBgf zx}}OL_Lbq7K`WoSK&yoYf&Ef8vkUPtuN3IV60WzbwULcZxQX!``q{b=dun_qT89tL zg|e=T^#cekViOwC&7&JR`{Zd7VUH_WY9U--h;pe`&_dw`LR_w7vA+p$_YE<3q4Iyo zI1PjF2A}plFdKo3S-@iMzfF_ajSf>7Ur%A&zDGaiqmJu#NaOvG+IBqd$n#}NS;+-- d1a*5*gCx3Wy7+mk@x>kFip&QTetbaV!9R>_^H~4@ literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/fusion/dygraph/__pycache__/conv2d_add_fuser.cpython-37.pyc b/x2paddle/optimizer/fusion/dygraph/__pycache__/conv2d_add_fuser.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..30f446007f770333e6f43e7f8452a00d0a1281d5 GIT binary patch literal 3710 zcmc&1O>Y}TbY{Qp*l8LFRRJN3B2<^uX#xm@+KQG45GqlnqEgJq;Ci;nw(H&P?7GC( zS}HVB4-hKB4G>BIA-Ev8p;DE;Ld+4Uxl9I0@CwRZN+n>X*hd7tCs z<#Lh0^ZP5mu|LNNxrLMdqXFgku(6__io+S2M8MV#vDEu(I<%(^M( znrppAULsuQ#uDL%tzA*;1x`L#dA zNG4`G^C;>C-@7z3dyI2bMKo!~8oww8-|!I_A`;P(f}`o-;cWR33O{r&Z{e0tgNDo@8u+* z<-kccHAKm9x&g_9r`S<4e*l7NPrC$OsjOYQcK6pGH@>)h_uKc^KL2%X`K$HMmQr6n z0lvKR`OWoTK3)5A>DJ{Bik0lyvi#ErYu~Njxp6J%L<_!m=Z7!XKmPIdFJG^(yubd< z_1iZ;d@kRV(%t^~@*g)o%aO18XC2q}?CA6~JK@heQOIP@a5mj!(~HyWtOE%gR*qCK zd~5eep0m$lG7=3h4E)e$(@z1bG?HsMcK_m$A&ie*nm_6LE}L$!44kY$rnQc|d1$io zhn+gLk01ZQ2wf zX1Fzn%kB&_{2*x>0Gme1!sz-(|6v&YAHMw`Oq&Ozk<94}}`VPg922!Za{_OuYYkxL1a?w}+- zcxVGRZ(z6CKs~EkVqAcZX}e+qrs4q@5)IQn=jM4e8KJX@E_qZ{@gS1s^xa}67m}f> zNSO)kbj?g690p0}s=7e2Ndd^1i+7Pkm&Ooi@x*8Zf$eeNUPkWsBhJjswB6q20V6&H zq~U28$SzAQQjIF`8)n0#;K-(I_@Dy14`JEI zX90-G3W?R2t}4(0>53AotF)&r?vJ$>FGTqHqL9DUNA)^R#)^`$Buci z@;Y1wWe8T49im~t`bJWgfn>jzSy*fOK_{`eZ3hz4oycy5(t(K@_(9c_=1z2Z$4bDH zS&%_kwHi@#!4B)C41gOF@{+D&4o`FmZTHQX)?j>MX#A$$gmg_xL z&J38H07&H~P?!>D0d6$bde+6!*U1i?k>-;aIV!wx7`z&O660=_7h%?yvk+Tw-_l}j zFS%fdg^a>HOfD{D-W-{GaR+B68eCvSo_cx5er>ClW&Yt{$_({a4tg*r+BWbmf<*sgGX@OYS z76f7+Yk~O)(-(Fv5I*EuOV&bxF`bML60{@7l|nbQ ZOrf;Z98OPt6kCu?D+?-G3F?@o{tfKN4Y>dS literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/fusion/dygraph/__pycache__/dropout_fuse_pass.cpython-37.pyc b/x2paddle/optimizer/fusion/dygraph/__pycache__/dropout_fuse_pass.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..44a7e4cc535a2c1dbb80ac5f01a487c5aa576532 GIT binary patch literal 955 zcmZ`%ON$dh5bnoJGKRzlimWKC9_NzLJ&P>sqO6Co;6>)rH0e%ubY9y%117{%_UeDo zqyNTpp9D|)7d%_LnELAbs!v9v0fF)B-A{2A5b_%r+Yv%=3}#QjDWYgj z3Yt=iG0R!OQ|@D)2Su2Ml<4p+QGp7thzd1b^EFL-ny*RPSH1Tn?q5S55>pqPOKV}i zf4Z0(Sznx*s;=7RO>4D*?NJTK!svNs8*Qfe({?Zjj=}6xaGIo4k(4Q#a>W$CB5BaU zr(p-^9_}I_wtJ*HWT84}F#~LYxKj$5ap26sUW3`&;2QFcuE=-KAK1`I%na7!A|bLe zYlJ{*R_C)H#N2{|KO!m0Cb4a~kmXF4^XD&;4@c)xsa%U`rS#1z`FvFS?WC%ktjLzy zBv4USmC2^iWT((^y)b=1IRLX*2cgTu{dC9M0;m2STt#Bg?v5O0Xvms;VJr5LuBf7l zeP!5MpvfmLXrMj7ZBWQ&dLce#r2^`h!u6|K8`(O=yoUpZup{e{~Y~P~1`$S_n4~qNr30+6gxl;F1LEqnJdyc#Aj5pos#5mk@&!;J5Dd(UEvi6|v b-cCFB744#z9phJbuyW)vr10?}jfQ^#{YLoL literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/fusion/dygraph/__pycache__/dropout_fuser.cpython-37.pyc b/x2paddle/optimizer/fusion/dygraph/__pycache__/dropout_fuser.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d956e2e57d15297b31fce3609288548f25398b71 GIT binary patch literal 2168 zcma)7-H#MS5bvJvU3U36!~{jN8jUQG-2+Ta^dtm{5MwSxO}MNdongE0nA_Qzxt>|v z-OMGxU4q6S5Bgxl7*J#M#lWi=_!oTcCllZf{skY@s-C&MoZ-QqOm%g4Rdw~RtE!LJ z>s12n&$s_ze^&^(hl7n~0doxg6bUDshNMHGH$tOtc1((MW@z>8j@@@UjviZ~`#w2I zxXqna!X06*m>rMP_ej%y1XjpwlPd46K`M^>sc0H%+i9QkP@I&0a&BYrjz1C->61V*lF^*37>fp)zC!V2V8)zPE-&nWcU5hWWurtiO=YKpL`dm?w(=MY zq99`oq^wkg^Uzlq>xO=sGWHAk>w|Vb%G&858u`(@A1zEDYR?@mfMDI2i^-@xJe+8J z+HsNv{a`F)dw!4xanvpyY4hSl45o!_B_oOViHArHAcT(hJ?7}G4G*_qV1xXds5t{9 zCyQiJL(9#8asYL~;La7eb-iM6SI>JI9leV4%;Xh7syes@CK_v3uRgqc^U=2#9{zN3 z?b_Y7E8nkwv)YZLH1nhE!L>W< z`WH9v-~M6!`o;Ahf4hI@^VcRSs5v~+c%yM>ddu8=EE{b2>U1NB8WKPj`-exGTh&mt z)oM-D#=8o%TT#?1BSqOklnk=*juPjCdU$Xeg1P{HS{cJb5Xg>RNR& z?gB7oo0f7DKg)!SR0S@Tg&5|_K063PE~h{QB)35)>l3maI91mYtkm5LQ)vwdaUE(< zht}v69Y6QDOWvZ_#calg@qst3qx}Lfnp?pZ0WHv8QZzN2uTJj z4sdYmz_~eHw68n4#myTAH+!bchx>6Z2e}jEqok)1Dz~xq)lBIYUm9rP>(`Oi0+C zksi~W0Wx0=-v!)Vl8>}545JMuN~qi6a7>jfwSQ6##Vs`mY?dyYkZUv#z=ypAhHdy1Zi=s|O;dbWrr7I5X3 zJQSU}6iMiJ1v~GD15v~khLn5}$bR^BwIlQi>d`8MQWd@$jHc*zI)43eAr(YWWil3t zPl^p&GqArOrbk;_x6&{mkvo9Ev8*siGiCL}2+pqENp;K@AXN$!Aa&$3sPr@{RdqHf zTi01tO?(L;K>9$AX&?lkSAYelwu@}qRZJ8)SfWd91h62OAp2fW|+aU805~=fT z=YJ3R5-g>dpoG@yJLyi1)0FPVFZ>^r-~oa+0q>A8Rb{Ln^FfGwjj^SH9~LXRa!V}T z@-PrpFQ1IA2t&p;s{%NS`LCVT56&JKrIBp+=KH=GpHC)Z0^`@)pFH&m`Hh?12w-srX3oJ$B56fx znp28vR%!g&VvEV z-v2K-BHCj);PF|(XMo5Nt)P$(2hJYYD=>2et|#BE`VL(D&9l1cGqY%xqM3vE-Z28_zK z$+jG`eU53j)FS{m0W(N}(AC3R`qCY(!i7K-%7Ly9lJMuAY{(b3W*_L9N-Eh`h8+ZP z-qiYi+qIx*T#LTA;vdUKf?!NxM{TFH=vCtB2={?&n+>pt+DRWAz9k!ouB(=hU}+GW z(2%Yk+$QUyrb&p^Y(%X%w`0!hRxY4e+{T=LT8L_U`*(=1 Ujb035U*1L1C?Tfsc`=RRe~gppzyJUM literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/fusion/dygraph/__pycache__/fc_fuser.cpython-37.pyc b/x2paddle/optimizer/fusion/dygraph/__pycache__/fc_fuser.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..559fd9887663804acff209b0333ec2298996cdd5 GIT binary patch literal 4197 zcmcH++in}jb!Tr}UL>nZWJRuHGilnIwoF~D`qElK;06g$1V(`rMZuz2tT;nymF03h zyF?^<3B-|r6ett}?L&da2&lFwnwKJdZHnX<`aDks5=-GP^rby#W=V>a(xNCzV&`(^ z+~&-j%iNoqDroTg`w#x+{(N53KF7)6XTsnW_+o6RMoFO62=scOH;tM>5M~5s)2dlb zyJpL|8RUMXy{=J<+WQ){nXzZo97^8QD!D#z(e77>a9(T2?4B31iY|&j^e7G3>)eaB z29r0uUB&^AT&*L?;Adomcntls_#2BgBsM_*J`E`UbAS9Ixx52(@d?F z0~PXO`qkY{6!lsib>MZ9@x&ha;}>6qhLTiswS**ks_p692Gyx?KyZt|$Gp#xK;)?6 z2D%m`TQ}K_Sj7~jzz>Db122wU_ap5e zKVE8v$x_@7cfGLTg`3MOOFvmvMsn*d%7(M0oz+O%ZmAU|e$($TzSL;Pek)wc)LLrP zU4SrNjdpnvq@RZ`##7aZO*-@Awx|v?8ydv|#2kFzhcDjH(I{YOtqZ2v*0u?nn_50e z3AOjgCZV|lxD(l$Lrn$$iuF7#oLDbPdYhIGOvRx`=#=CDE0V)BVjK?m>pcUg0d@#S zA|;xMrTU(IsO=GYZb4hnx_XjJbvi@8cA$5S)Bv4leGMp|IiW8?pQUpODrKFQtMl}^ zO`R@CjYgCgCMdrSbO56k^gu^1K0(c%&p6D}O9vKQ8Y|Ve9eVk$($(o1)RLMb-?H2% zP>lAOSyJ!iQv>l%?7~dVGk1Z||FjFld10vOi%)dW^c$o7WLmxiT8?CVbF?m@-%>Qt z!U|P9zMbv%inJBdsyzYx%srXIQ!CKRc!Ov~y4F>2t!r%tK3G0_{P@eyKJGtx=gU7l zI{M(Vql4cb|7O2YfBM0v$DjP_=)?Wb-+8xC&VB&?^SeiXJbe1s$I))G)e4{f>BHmq zKmOvA-yQcJ9sm9>UwrzD?+%fXW@qKbjq*BltGBE1mKU+JRs$AZLBz`B;+iBy4(sLB zu|>autk=rp(omvwZ@=L)?%s6YX8z_@0$kAjt7j~YG0had3HZw0s|okQIBLc0%Fg;3 z)Qe>ZHqG^)t}K?9e>SfDDOEOnFFsQdTsTXOTO&0f_B5MU{4i=K<7y!7VtHpB?Ymgs zh5p9P#qtJAymipY3!*I;oK_6bz60E$yEP`61u^@7SoHR2EB~8CLE}3kUQI7)wgWht z6WhOidbeW-_un4F+Ojg#o$|HvO8M&e_4=x#Q^s9e9vXKT=LgB}O^ov;Xedb_Qtgp> zOm_NxL3SWZ8F7M1S@$WQ1{~F#gOPAHS?GpdlPw_w$PPB9YFdkO# z2Ccdm#4lF|q@6;<{ibq9VaZ$7DKzTW<>>lyAGt!gV9R+da;VE>=+jl|e+i8?=NLB8p|qKX?*e=}DG&$# zm&h!+NJ_*6hz_gRfUmsq*Gnm z1EIzcDlxjY#5E_NPzwUCH34-Jb5HB$QRTU$i1Jm#YY@WyF5;UUdhi(HxRGY(@l^`q*XB|hq zLLPP@Bp|K~N}c?}ZdH{aog54|AX^W^Sr}H|kS`aY7>3SDiI1+(5GqXS*AsZ#U;!oU zMvKG6;TUg3ewa^%9!4S;KWH-s&l9}kz)XgKTd2=%iLGJ?+$>?qp(w~8Co_sDsz9Gj zl-_ERN6ppHf~>-0QOsUln2Puv6fmBsIQrHj_~KP)@OGBqcAorf2o$pbnT7yy9s>S{;9*!DJ`Aga0IQsvb4;hT vxglb`#aT7tEeHoq6~OVV#OydiJMjZ4aHOFcsVd;E+VRUc8z>3>?->6Anu8)n literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/fusion/dygraph/__pycache__/interpolate_bilinear_fuse_pass.cpython-37.pyc b/x2paddle/optimizer/fusion/dygraph/__pycache__/interpolate_bilinear_fuse_pass.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..b29b28ff68806c96bd115a647d75716d90baef52 GIT binary patch literal 1042 zcma)5J&zMH5cS9IW{DD@pg0MM0}8|nDGOI4bP5OrMRXDk+Qu@mFSoE?*c*vNp~|(i z)KKzos9hmZ@fWC=@g@gF6mT?Sdu`9W_hvF44*LYg_ZQ#9r+|>n7RO+A3{DY6 z3sTaIQe3lwl|18q&5NK6vyc)UJ|ikn;W<&Erfa^YS)`*i$$Bb!Ns``Ah(i+Uf)i;i z%!f}FGb5X`=T)msQx~$;DSl3}po5ZyU2=DOe>TMA?x+rt$p1hRuQ!;)DS#3PXi6a^4xBmIV=#LeTuVOF z75U;t0?KrfFoQr{EFfmy3W5BsE>6D^a|sT96=_+ushw8~S)IyicK<>8>gYr&RcJA- zmA=@e?~fY)Ijx&EFY~1~DIm)0D&3$?_oE&)3)6#?2VfRK5xTthZ&5~DYss2?U@P{9uBf7lePq}ez@JZC(AG^2r2D08r)T1AUMbK7L%3euXahx0xQTHe z`fhL`>TJ9^ox>J&p=_FBaUGTxa}yfT<=y{4@1ISQ2rE#@QVZewLX@?dgTTTKgm^cX z#pWcy*^6YZLXN*MV66-SG5EK*24*90F$-AC{kKUHyMt|wMi(_2bsg-R;MR`gu7sma t2|Hm9cii(~N?FMnXoCv)r0&N4n(t!Z;)fmmM^}&rk{VF>@d1qozX9%%97+HH literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/fusion/dygraph/__pycache__/interpolate_bilinear_fuser.cpython-37.pyc b/x2paddle/optimizer/fusion/dygraph/__pycache__/interpolate_bilinear_fuser.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..4bd054eb82861ffb7830011ce990f94872ed6b9b GIT binary patch literal 6426 zcmbVQ+jAS$8Q;5BvYmunNDOHZloAOX*^(UFp)sT(giEQ1mJD=6nVpT!%2~5sDQ8!$ zty#ALXNCvrW~O~B8OClqw0-H&r_KOF{)Im8Q)eI!`WO1re&2U?b+MIZs60M<&Ue1+ z`F`hmabhAT!S7$M{?q!$lq7wHjr6BMv`w{^W%3(RG1CQY(o#{5&CBh;xo8JYQHjT1w;A)C zm$==!mNs6q*BuUg^3!z`Nq=fm31*?BO;Vw;#22H3{1S9*37&>l^`b9K*%@OY&j4Q+U8= zmC|5(!!`zzS@?ee{)1CGzz1_9ZO9=&g=t&zh7u_&GG>TnTse}NxhXHpEW4%367}Sm zA*dsJ$JqEl?*z#nUf&!QQ%Q}c9l>Cd1YJMkv3&^ zY*Lz(Hq=mx2vb|ih8Ahy^~3DA(C*615<9^r$u8N8)D@jQavmhMjY!)8c#uDj0tQdA z@7z{E?i6`-0^}as(v$u?8)uJ`S5HvycL!p5lDq~w^?z?DrcwUt;I9s&{8Z21(=-|d zk}R^P$yWJHJJ5d?P&ntxY#tD}z|PV5pd5I{ z=7Ih!(7#W}%&eeCP!1u)1dmL@wX%_ovLJar(Ky$`v+C-4yHB>lpW?os2FgE7b@85H9s5PFFuQ-tOQMr;)Ki=(*-5dzr)ODVE@;}RJw z`gR6Kn8p}Z3Gg0DcKNnWGnVECjW1}Rm50g@x|SbzP$JO?{+`F`PGg0bNS>qYDK^M_xE@IvUB%~Tdno* zTGPM#w~z1r=F_h~|Kq*w_wN1av#-DW@Y$X?YFVq4&z;RL;&-833a;5LXJo(U_yyuE z?(3ct;-KbCzOt`(9(vz?XNcTr^1Ox0&gES{&tY40NcUWE*tgWb`I6mgIX){aHGOBV z{mRS?SU~xO@>G6pVJeSYq8#Qel}bZytW_#==pB9`hQ)k!sM~E&&wPHrQ`~j~C;t;H zUtFs@u;DiS!s%Z1^srqKMrCdRY@u&MRva&I_8S0~<}s}MWhGHbJR&o|Kho-aSwx7e zi}~3h+o(lE30(V~NR7-2N6yW^Ju{p~WWXPKn_vi-%J*feRxLMh{UEgcx&uo44M<<5 zTAj*^BTT8`wOv+v)8?*y)!UPCVKQ}mSmo##5*MpgF%s2k$#1W=)(gc_*u+yWSw#dhKesQVqY;(L3BZq@>&SyYbQk$jWX<1y zfDqFO`(pm=K3`B~}>%>@aX!p%iR0*9liHWHfYwz@%UK-xO z)9*MG49>l9X^@ZO6+u~w_0tE_5?mYjX zL-^IU>oJS&5!^4G_nH8_;JH$&i)TPhQRXImI=L2iSH&V9>#pBwhn*t>NIpZPGc#R0 z!^7>&pmq>vn(YvHgN{k^0E?s+w6F3*z}v;kLl-?Rs{CR2aJ)&x@_HvLBw&_x@p6#1 zo6Xj8m41Z_%MJP!E_D9`zs?v*hy|ac%+ip~jN>`0jvsb1iL*&iVY}ryqoSEDDm4+v zOcGGJi^(Z!eNPjHO7g9CGUSAFeU!pV;`x&ox_CWIc-=-9(+Q(C-m*D3XnA;#72|$D zs7tqsE~Pu$+b;$KK`YC%TDNq&T97<&^e;7i(}u# zTIV2f^jFwXkyJaUMr40rjqV&9y&w-JS*?3P=lDoHjWPfakp`3NE;v2d=3w zA{Zy!iD5~N#|IbFE-LJvz+w^$I6z6=N3eJli<3|k^*%Zhn3PjWC9Ob`1%l2NjkMb9 zoJ}cnX=Nd;oa>lEt~@hpOv~TS_$-VNCapNe5lnFw?(hORu|WU|Xb!~MyUuz*oPy;l z%=33+qupZgh7s$u2*lzwf%`#hi1`Cr5$YV#6K9e&BF=d*{SqS+3DFV?vu;Cll;d8t zQadulBv@-lQq}>y)Fwx8Cj32p)xGQ9#LyCkSquoxr%g2{YqBC6@&WqhdS6pEfpR#Z z9+eM4dpv0=P(L7(qyn;v2^vgA=m9yZp{BkJN}#@@ASJYnw5XzvVaZ$~iFBa@z12vY z2{Z$I#6Dp?2Q_TrA5g$MXes^J{-<@)fz}6g3eSaBw2`(&ECx|%{CZKNPe>7cIijPS zB9AMbGv8{nsnbglqzhMRoQCx5^CeUwX&W9uB6SC!PULNQN7~RsWmDQPn7ol$t3~>@ zx+7hZ-dXyYvSGsI*$g#!I??I*MVXY|Rrr-C6I_Y1krA2a;qeHsvb>PChxj$(FT5^Y zFT5qevw^XZi@@qiCd$EsjInJ-Il7do+u0pSmex)JO;|saSkFc(TDktoTkxg{TeZG) znE-f&(`u57CulcyevwR?hhX8gmc8ax+Y^)^)|gulVNZqSSj0wN>*ukl zwwkS2uY2&`m`u%Br?$zbz6(z&JLhJ zUH|}aAg|8!NeNy8;0vkOn1L+|-bLW$25$ZY18OlycSLW(OI>`J<@mcOjYmv)fzDYL+&0>9OR?f{%fhpU z=%J@zjv3;3lqFB%iR4wf?s%SM^__+NYXX_@NJDSjxSIud9)uNHGk+yrH;)m$nB#8( zHP-3!6)S!#ejbA-{348zBTIVD=Q8B8nEI zq#30cvx1d8<38p^P=;AZi4M<*3RHMWRH*5OZ)g_jXhX7|ir$i>_XqNjgt}lNt%dpE z#d2aU%z{~u-d@puz^Ei^gOq%HdFj%Hy8vbVD=$6O){!T#uUxCVv1jqEaJ$;|RC$W8q%xQL>l$(=C#za<;;g{|2~x~7UM z_LX5nK_VZ!K&!b1UHwwF({u4DuN26}6s}h{+Q?QX+{CyJeHUGbr8K?(ox`W)LfJIM z@*ad1u?dan>h4~?{^B@^u&k9VwGgf^L|Ll^R86>n5T6&a*q#J9`$Cx8ko7-QY=A*% zgAe!+%tqj17OSw|qg~-gSA`#SFVA=O(VpY3FQaW=x>7vYbI+$KWhLiO4b<#G a4U*VL)5T9ajn8f&S7bh*@bLkS2mb)3v+5ZD literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/fusion/dygraph/__pycache__/prelu_fuser.cpython-37.pyc b/x2paddle/optimizer/fusion/dygraph/__pycache__/prelu_fuser.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..80ac4579c5528fb00c0b0a38357175007e08262c GIT binary patch literal 4020 zcmai1O>7&-6`t8YE|))&|Ky(ps35e78Q7F1H>gwCZkh-QPz0*sIB?4*SkRoIwDNLE z&+f(~vjmbzKn{iKLvKkCNNrIxKv1ODrbzBR^}MG7iS6j#Q!ni|vrE#J6oCD(74TG7?!rf#AX5(&sKn&6-z@uV%WM!}>Oq5i`x!=aU=(;Vx8HTR=Df#cO7dt_;7{C{@8fsB{quVdKKkmx_8%U8xTRjn{g1wU z_{HxZe6scRyYD;m`7c}f+xrjxvU~p@_j-e9y&K&B>n9I?_xU$p{PE$=dk_Eg_iw)Z z?N2BCQmoYt-d=W_es8@Q&R?0A@l=15b_jDP_T43S0PZY$>h0L~&oUXeSDtPA-DY%g z`Pp;x;d&E<-z>d#ZocjLzWa95kNK4g_2+Ac=)8_tADfqPzjQWp;$`XH%3%^eza{<6 zJ!swv&-Sl;pGX+rf=JtsJy!Q6Muc_l^9~Q9yI#m$EbmitZhnv}g|)++z)?({z(eHq z{J}&?cB)e)FmP_Z+4sULOW&c)U3!8#=%4D!U&530A+@L4@G$X99`-&>Q-3B|JI1ru zW!B2XXEZ`)ZqkzyzqHkq_JKTl7)hwrJSJwqlMS<&rS=*R+@RUvHK{;i0HMRBgK%*x z_I&2{no-0>P`~VVTTMT_S|3XfPh>GEcr)T)6VM1wg`U8F%L{sO6sCo49LdSB^hi+d z$vvz|%P3+$s~O@1CIXd=08wRlrp`1pht*s%2uQ+N#nso1(jHTB#D@B-W0;OIFBRm#|hC z+5LKCFuF-WSr`?QLSiJv(`3WEt%*yCzGID_cPS-zF5M(|p1Db)!X_CxYAmU-%(PLZ z|7Jor>|Fb5rah^M4--2nCxy%8&K9&wz5nblOl;18=Q4!DY=I_*Kb$f%iuj6e;30&l==Orx~@@d6+uE*i_i4pm&HeyROI5 za?XQpLAs!HOS+jw8}%A8E81Q^6A2qm$IC+zy27~q=#m$@0p|?z`(tv}3(!_XSkCwS z2DTDDhk?+e2)3gLw<?WqHfVF0mBHcA;ljay+pl&Ka}t({xKzwt0yY58 z=nmZlrj3lm*w7-KQCWlg*FZN^ysaQ2%#oE?8z$2b1y?b%n0}g!3W{ccb{(?r#{nQ2&0y7SbL6lYtdI={v0qBi zd~P_b{1c)P*Fk>Ruv1RT8zp72GSNH>%}3bLu~u|OC3Y;gpe2>4oY?3O|&5 z=FON7GDX>gv5NF3_htssTuo-y^lS%h4d_27yK?OT*b5 z6C*@-*z1OTIQP!Tojkeiu5c9)3z;tBq#Vv*sF_(NShrC-xc>;g8|ari!%ByXHB=4R zQoJ*QspSd$0_9Z`pPC9|hc*lB1&~Hh{Rc`=y7dAVt#62TSzq_KeB{ut5XZr)C<;Xvl)^v%7wZd z&SQ{iilzmb&&4O+Q;vA0^_4@JBo^2kB#4$v=o%9G#~--3AJ&tUN+O zf%=@1p%mH}Y+5$#7wl7tS}TiZKub-$bU>gM^wLtR>-)SF;p-^OKzCvM`Ba~LKGpN$ z>}CimglrEQCnk)ouHf~a=mM8JLYBfT5RN|15J#S`+?e{E&9c7ga&`_^WrBZS}#%$|c&MA3qj zG@}$_R7TS8=6Hr+K{ZLqPHaJ{f0awp)RG)TsHc}!fFHC!v>Cp(X-sP+D!1L-Cz)$f!T-PG|8wU8B;XliYb0cvY>;{ z!Vb{g-A6!dcTaW5fa0OW6tD&2PAFu?finkt24=5=Yspu-Cf_`NU_-|VGgyy{g~+SC z6#}VQT};0da}5suh_tNQ)Gn%}tfsP>Jw8d_onA<#3N0qJ(pRhW(`njTfPF_{dx+o2FRa zhtMK6p%Gm@_)o?D>Ntr|+)9>O2-g>)tknXv6K){H$Av7mCjri04099m{)3BcFbHn& zG2a2R5xAHIEarZ1oW$;EC;#Y5{!#Y^eSaSvIPSVL+IFWCNSM$MqOXZU8rs#%R(EvMI}pZ|nhAi@&4 zH6n7Zv1-(8K`)YWeg~wG`7%xH_haRrca&RZN&iQV5WagsI^j}x@uAalCD73mb!6#& zMuw=BL!dxRYGrhcKiM+63ox4R$Qpo=>%OBD=f9A@FI5{sR8?`%a)JdXSR5O#em0Rhgx8zG-Cb3e zCqnI~YBP+yhSzpwbs<(>GpJ^YRAp)rhiz9@!j{Ciq7jM#AhiAJKBiZ?8t(c4iBb9Y zV5qMEXy=j@@WT~yg`$gv<tCJL}h<{d8mN?xU^S-*11nmYM$a z?!)Z|-)`Modwl(7p_F~yv)^uR{k-|~k2_&2T51MQf4#T;&HX12e%Ri)vHj!kPab}K zc9#Pgm&d24O0y*|h>in#y4-s>4TqPO$EV&p38YhRBb=VP+)H%^s22R#6BQpbT2(S- zjxWyw*O`+gor8c+Pw(BZ8;Ncxgkt>7BjNhy~zB*Uw^0)0!dW{M=BeA?7jHC8Y zX2>bMJT)d!S19J3UaKVEj3b~^?LN)vd+G(A@m?VlP&_IBST>Jm^TZB{Z;$Nf%3djx zWhRHvdQ1Cu2hA>%xh3mBvABmWzmI<{o|aJFFRGCd1Ohis6J zu}PsAjSU(xfW{^yi=d6T$VSgcD`eE9BeLs_Xnd&-ufyah{ zmTRjQ0D*#_8x-wafo9TyHa$v4$%@f2pk*__Ou!7>U9t4Jk)AIfM~V$|lgyFJuYAr{ zax3{xzGGcwa;}3stdpA|*XBM=8@@TOGh7~oD^p)t@+jiR5MbKKLja&WBDbN+cA6w_ z;8tH+`rH$VrP10T6cW4(P#?NFlm4_=(uDzzP6~UcJr0#6kohhQH4K2zBK!sTi*$tA zv|ZZIoBxh!*~Ui+n*!%af%8UF#6IFh&acIepPuNaL%s$V$Tty`03-w1!&P^EpYuJv zpl+}^N$ffL*+Wk#jFe4_hJDPo^e();RuDJBR>FcXIof^CD!cDlB`Z3frjnEx^xC0a v!g^D>l~6Vz`5Mx|%P~6C_7qI9v?b1@2A{(g(bbU-Q``6lrZZ=7 literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/fusion/dygraph/__pycache__/tf_batchnorm_fuse_pass.cpython-37.pyc b/x2paddle/optimizer/fusion/dygraph/__pycache__/tf_batchnorm_fuse_pass.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..8eb795fac50ae5fc8748f314279c0e3d7fb364a7 GIT binary patch literal 967 zcmaJAB8rxz zqB*4)vyxRj=RW3TP=$F&i4I>76{zr%s8G`_-_ktN(U#-`6}=+K;1|>(33b7_v=;W; z&sPg0+l$v{Q`yZfvc^Y@_@d0Ji0zY%i{E`F=Ds=L%K>y@nMvR*uToW40Zmr9jd%o?SyPU**! z)_+c$wkxV)txXDuil$C?Xw$=JC+*4%pw$tWMKFY}AO4&CWRG;_b0HlW!XxNeVwH|; z$tSj9@92gqs@P|S0|&l5bAeV14J3yZ>L%V7wE_*Wgc~%iHnP(RH!+?=KV=u9uEwj; zJA8UBlxH=z+-AO8onPtTGFm9Ax_g>XY5szxnANa038d|1kIcN5_5MKL#_ z_FuhmCI$f*eCD^nYy>W50gJi+mL;(}+RtKgHH%6AGX1b09JuaBHQ9}-7v%ndJYS}g hwOoKE$lNzI2;vY-7eDDaKD~ih5&4M1kB?|P`U96I`NIGJ literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/fusion/dygraph/__pycache__/tf_batchnorm_fuser.cpython-37.pyc b/x2paddle/optimizer/fusion/dygraph/__pycache__/tf_batchnorm_fuser.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..60a2b4a7e2c8ae4e530c146166b1dd02c2961620 GIT binary patch literal 4811 zcmdT|UyK_^8Q14^UPh=S+9Mw z>-5fDi?(-CAJCCVytP$zM-eIrq4G*m`_2>Mai5UdB)xZ@c!A$HyY}S_M-eesIKYBkpr)i(iVDy;iJdZcJfre|G1zIEfvWAYa9vB_7VKRal zfz`1aw(6Tfq2n~1PO(u`V=HiPXfJBq=7n927lgfIG)kPkq}7~5NY++sEG>Ocav=r3 z?zdtLofrC%xYmqBO;1bTZ*m@p7iF_|do=iNb6ZHzu?sDdFnWx<6MX|6p*0xS8v2{s zI^&G%yH3O41~+%LhRH2%<7@E(cks2})J&~WfJMc0@%r`#sk-sPwPxJ9z1o!>GDfyo zE}Xukh~DTt8qN~U(_$9uiMFG+4X$%zk1;JacQjxuu0gQ5Dv2>I@Ju|6caAM2`LLi?sx$DvFD{`;74)l^qO%jWEg?pj6|@GZ_)EwK{JXx@0Z$N zf4I^K{`Mb*=RIovl%uxd%~jCN9K=Lk5-Kszl}FKi;>zIZOCFc z5jXePP*3zW3=X_8OB#dE_@Y(MPv<0vh3 z`!RI}r7Y&v*ZV|cC#l2Zcc>oSkvV!Vc7N|Sv@G{C>d?aSfLIn zq8(DM98BlPPCtnKUa);=5sISSAqDe8vU_l5J?Q(q9#960>LL&w5yr3hk?^Aa+M$}0 zyB0pB41en4A*JWT5-}EX5N1vmuhuL{TGJxT^pv>NQY$B?2y8cdJrQ!uZyH0nK(Io- zrjy%KIkw_YfHe9O8qIbLmzhk*Zz}&fEQ3^Yh&#q)6eIYEC5yCzVdJRT#FW#9$Lw9w)zVOiMFpLMuO~eVBFG&=Fp1SzMdEdY-n#?Pwai;K)bHJ zdTkrk>}&Uy*+oKfvUKJmhhnwwVCDHN~d9%r_?z3tTDLiAM6@?+77QcI%r z5ON+|snP59BpsgA&YV8A`#p~1TANa?nOdq2N^$D)(~I-kp}IpvsT+<|MQH^^PMqoW ze4dtb8k99;f|5;eD?1C!Pb)ns*8Q!F1+qGReo!DQ>aRbJ@*^)4f`fnZY*hUag-;B0 zKIi+GN^}8@hRVmqJB8m8>LB`=Y!0uhJAf^qR|bX4246kN?IwkfJ#QKED|l6%bPLOe zc+2+L0b=gZOw6_(8=O(*zR$L<0ArHh-DfXffNxksJF(jq*WnLgVh;<6(Y6z78T-Dh z4V{D~4tXIr72aqUXsmEW;NYP{SMifS{pti>JI7_b1@IuxB_({zir2ZL-)GVWH~cb} zUI9e~6jNg;@m16aaAm=WX8ND$9afcW(tUnHMke`GatiZ0j;A?S63D$%0GHi%IL&Zo z^E}QP&Ifm)xwNS9GA|}GsD~>1ZZdN~b*I6na6y>fFnM*)8qV@lz-Vopli^%~3q!la z3(sh}=9Lf+%Iz}Q56Sb%e7mCNvM`oE1NpOjZX_G^=NzBU_2|h$QfN={g=H z8F$3?u>`5I=;PV&c>4Q?d?(3i|8;`PQs&?Aq`G%$P|Y)DjCc|Cx{(;nzB=^ABX7Mc zy|qwM^pR)KAm^!=V&B0XNCc0iZWC9g&cN*mxgo&v>JaiT4`y@uVOSsE=2A-~(t!m{ zYhmiH`$6E{Z3caDqgGPl<#p0`R2G!Oe#dLwZiY~a*d%(9A9TT~0+xphnM{joelyCt z)lT=WA8vRZ(adU|bS{V7ZA!lxwuGvL(gI>mEO4jOy{BtcNryc(u>}z>c3l~#h9AbV zPL_~4X;JVE5qT(NQYY*BVP+#$8L5q>LO_&MFQ{Tho+hwG)Tq={Y@YyIs^Emi6%9zX1s5l-qrZ$?LPV)#=dTgAQ z{O}0lu8BA&Y+&E8Sjr6fBxx?9hRVjNtF|j!Gk5G77t;bR%v4gQw%Ug%Wvx`F4IR^5 znuz7{YY_cD-Y7<+Avl*2q^a_t5I)7Iwi)%PuAm^EwHapAPN5!h@!DunyC^Ce(@QdS zaMG84&KPcj_T>8GSX!+)bn&Oj;o|RlUZ>0Z0rktCx7lw78Ajd2C1oP{JT)}Cw37cB z&=S}=?`Q!~HCBm;%5nAA0=aiY6)9}BoXA;H&{|o39kA4b%Ok7mVJ|JUx7A@5wGweMgn8QhDvnae{vA2g-~iyR%+) WiMX1r!*f*7s#}PYFLaeT#@_*I*~py$ literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/fusion/dygraph/__pycache__/trace_fc_fuse_pass.cpython-37.pyc b/x2paddle/optimizer/fusion/dygraph/__pycache__/trace_fc_fuse_pass.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..b290ff5f3d50240ecaafe8a1105afc07d906da47 GIT binary patch literal 907 zcmZuvJ8u&~5Z=e#`LJw)B9JUm0Od74NR6T>NI{~oB0+(!Sys1e=ipxKo+WIV%4zc- zpyc0BZ-qp~U!Y=UFOH(%Jzx_qOhGdIY#dWgT)D$eF{#Kj4F~bMKi9L;+G@~dfY7Rk=((p z)QIK|JH!(6T+9JYAW}{t7Y>{`*efu5A6!Si(GB_Tr2=s}OPE0hE*7Gwi%tksVRbqG zLCih4`1>KP>Mpg*dL`?*tQXH;r0Hkcp?aB-QW&&oB2%+oA9fi{!wcH0l3`&BY_OkHTj%>+SwqYOXhAOJq6~q34 zKt6Lp*R&0&8dkE)FT|&!Rv;QvxIxouBRidN6XQPgG+l^=G+uu1@FlrWwr#n33`>jH zghq7z=q^nuR3C!b z2wcnp7IS|#OJXy>3Wfg(rLBe40wuash=vxgOLYiG4I( U{Ib`0e1Ke$`G~^DM>HP&1IP#AW&i*H literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/fusion/dygraph/__pycache__/trace_fc_fuser.cpython-37.pyc b/x2paddle/optimizer/fusion/dygraph/__pycache__/trace_fc_fuser.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..977dea8e6ae2bceafce14b8fd421ae4d5a5f6c87 GIT binary patch literal 3668 zcmb7H&2Qtz6`$dkC`$5roq9Lkpsf(JNtAe{)yD2_QadP8AVG`3D3BZir3Jy|P&OTk zVulnVDHO1Y4dl=|d+4D+b`jWZQM8vLxdlm({0lwxz*B(i+VQFP9@@S)LrIp3EnEs5 zzIpTJ&6_uGe(&kUO2s1ZeEsvk*>@iy(otI~MV=aH z9lfP@jFut#T2Oe6yiAzRjD5lkPu){nCZn&CTHzG5kgXcc%$Itx_o5SfH6<(k!eK1% zUgl1;Gw!_N3_K2U^l2BTj2|^8;x!;VVMyrr`OGof$1 zPc+ghfCnT`i!DLe$sqD-Mph2|&~v!$hs^6|dfbaV&JCPm+Q9S^+s?GWkCV)d zoFwsh7=sQi_JTHii?;0sP8{3zU&udy+318xBkqL*Cu}?6_O)i?_3P4gw%cXiq}S+Q zkAz1X-6-)p{?Ox%b}#n3VIw!Qk)R9gwrd05^J*C#h~$kyAhSgtM#kZ>-$}1u~555UpVxbz|?T{^q&WtybZxbvEB< zu2!%8sy3(mxvBDwlXQ9kXzt&5_WEjd0N=HnGxFzH@|N$!|Hp<=eAc4%YcL&Fey;jV zwOM_-dd@Y^D#=gmv){~}Sikw~*fb!|&geJSrhYil;DWb%OYX z`gkHobgNNr!e_WFqS@u1lXx(NgJOVd0soN046)HTD|CAa>^L4?Y6m@^)x|Z5rxDDT zK-Vd5t)_CsTYeduOv6|mR`Ph4Avv_#?p2}o>b068CT4VSiW@ma=p7a#?sw|oXZsY> z^o&AX_;pwq&!=Ls;o|&_8kTd8IIfzS>0*XhjbA~n+SE+`Jz!)S9^F<==V(M$#P!84 zGhMioSCDNuQRIaTY+8UWUq)Ubcl7*0Rc4OiF%XS+pdpK< z9y2J8nskwtm50xgdJNEr(u$BDUU_&-xae4XFe@T0avc^QDLI5Zma0d}NK0s{9x6u= znvAZHk)CQP-BnY4g8*Mkb<8WGTu@}WP}8%Dh%xVN={k<#ooN7`ac0Pn$<(MD)ePaj zOcympW=+Bl0u6%1-SOhA7>>1M zsANiMVOvQJwg`FM`knFt9hK8Ez}3aAQypB!F}hFx4UfEG2n479!tSF$WhFnf+W;^< z4hLHXD4FzzrYt1i837catN@;2@tsj!rrTiwNX#pfJ06DhS48p2zYp>==wQWeH}V8H z))qwY@h5S8HSj`t-Pdsm2~Es*6R2EwoXE?Jc9%oo@ghnWCn+OS!YC`mZ}mJ6icJAV z6?yRyd-FQ1~2|hI)@H3*S2P26^|A` z6>r&gr^|W)(q-FztLFrAM3mP2hhPM6V1tHbmAtZdy&$mdsqK($E}#%*F!4VDkyG4K zfIi*)nP~_v{EgAWUI*&-AEUO%#}$8lQt{Vwu(c&laE^JSwLB4YyWFcsybH+E5lAkM gEnAuu=p}w27MxlrDVbq6)4d{*7x^Ca(Mi zSKb+*KI#xIXU;j7d+*Go(`h1DZ&$C_tRwU;Hr1*DIDD1P!s5oxngFYm56PQ(CsEhdP?iRS%ZB#yuvt3v2`qI-f}LP|f} z2+9Y4=psWCQY(smnw92;V61PZ@acl>r)#_tt#fF-v>QSg?g&Me3E~Ti`|y$Bb7?c4 z+DXPYE?>L+{^;1g8_%Irn9U1`zunnS<4DqW7^MLVp8eQ{;e(LdiM&%A{f$`(jckwzPn6BZS}_%uc{5qG(Af zno)`|D_O-e-o?BOsxS*F(cw#?0u^2n6>7TWTbe~W+LCObq8UjBzaS4us1GirwJ_hE zIq8aAOs2D3mioogY6IIxEgTD@7ln1&?t@+GO9?%6wSC|ieHf|=;5=l z2lV&%5fIxyus!nIT3Cb@bHEmen^VY)17{BQ9L(MX=g1ejAzwTGz=lo}X0RR~3sKaC z69TDOUCzG|a{~^3i?pg;YM1p&)^k}e9zRLnoLxwz05ETqzFwst&)RM~ZCY1U#af#b zDk_>f?F&u!3mvyBGXRuhFpG5%x;{Bfcf2bw@9x1>BnIv7$zcXZw&WAruy=Gr6;#KJ|fX+6K4{E9vq}@xG`PP{$O0(6rh}rxR{sT!#xI+e;s!oTd|@clgMBDBHGN zJ%rFAHlYz+Kl)F_-PLIlp}4iIv=DwMMAfJzXeazgh!0Cy{yhnB?!+*+A@4u9*am~( z1|Rc1FdKo3S-@i6?M;)|AMfNJU&}x4-(c6@(*wtScgB5pdM!>4Tw14(z7 c{?2|o4;Ag>r#<6mx3F^LF{1GC5sgQG0B$w*G5`Po literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/fusion/static/__pycache__/bn_scale_fuser.cpython-37.pyc b/x2paddle/optimizer/fusion/static/__pycache__/bn_scale_fuser.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..6ec77e3a235cc351336ac5c5116aed88bf9c3be3 GIT binary patch literal 3692 zcmb_f&2JmW72nw}E-A`(6+4d66kyX77E_s$KqCN!C=K1%Bw7Q znb}>(WR^e@tH_~neCVM-69iIQ6z!!*uT7Es3q8-NKw>Eb`WN)lyf?F4k+SR*NS4^y znKy6Vd-I!lAH24>STo@H_b>iMUsy7X&v7t)Z0P(H-UJ(FFcKOa0$(#U`&P#yNV7t_ z?{u8L+i}&{4lA!1FB;5Y?vBA+Zr!#z9wXO{W@Q3WjEyFd-V0g6ulWgYnsVV~pRtg? zDEz^V>EI=Q$OZ6`r@FXg`dCFLc^W!gnbI-eGkSy(X6|?$i&@OxF*>$Z-f>ukc`$b0 zGi;+%0TZh7k?X0S23>k><9gQ*IT|9mq!5f>RDy5vC^V^&lH9m$Zkwp$4#6FQ7;8gT z*l-I?1}kf-L(}1aH!bNT*?iN;u+Qrpht!qGNH)gyXt=?K6D4bT~K^pXf5f`hyED7RhHBlQ{-Hd30JEB15 zqBR%_JXhRAVnSn((c{N#y*0J4!~pm1!23MB$r2!Yc#{S*AUw-!S;& zx6e-%D%$y#M!|8@7l9vj`BAPZ9XRMOHR^k;+H$WiqxnAr>z8L2&u~6KN>}asVkxJyoR{fOL$r; zc^KrH^`V%rmxBIp%(c{S1p4~73i{D4%mNPKxCC5-3`cN0?(4_kel+J?$9P+ZhLIFv z-)(XhGmW_+$t}nlR+3^uGy^6Ufk!tdpeMa8uKo3gD4+=SmyQedn^_Ptxbi97_o#I> zj3Hel&$XuOM`s@pRivZB%BZ#(T~NIXYXYr@P(Pv^u0OB zM}1G)=zJ42U7UnQlHus_xfQdp1IQE1Q4=gqCYY%n!VG6L!Av}=6oxm6I1S?sNYs@A z30Va$s_v+C6h=);EC7>$a>$SidW)0=-SjB-i*t~k1NP+S&=_^ka*0V?asplxe)?O3 zuTARko25Q@cJ%$PJ;pAgfDi5x`zmc?eUux!2Bb4<&m7xO#8|uL9$YF$&KYATw{x;> z8Kh=*2;hi zccPD(kSDSlm162;m)?JqPX!TFE-5W}UV?2T9yA6(c8b)<=*i;%Il8Qi*8?9s#Ilt6 z4-CMu%w2*wj4i|l(EvQ`+}<`3m2E{CyU?@8mDFM8uA94i1j>X8v#uIIv1cfVA4+jc zDc0Cil*%DRA@1g6w}O(!)x4T3Y3B^6PTgOd=Zr#%_qA)B*)^%B?Pt}VnLD}1Y9E?; z<#*=$WL(Q@5IPq0S&qE7#=0^rPDkbZQ5C2<+{$WM2~7S6kHDHyd>5JrMueuRyhxk@ zre;+3N98#r&J`HyT;ucgD((OTc;-}&~a^XE~E@&TXBH<}$WK~})xERvW-dI;3 z$d(vdWKEAS{mTX2*E$yUpa?;o1s#l|tPeHtk5K%4#1uaO+ArZv)}XOM~?T&IcI5lSoouzl*$^V{g`D?ms44%^o`69^^7Xd#S>tGxPT2> zB^Qf72VEY9lpY!mVGC1(z=*H@S|IYuH9S(cyi1;|sPO;BQTvu=G>}hB|1-48{|v1n z=27jNLSWGCoGBT*vEZ$Nh{2ot0^LZRm0X%<$kHHG`VI}$J)MxQ>J2}Kv3zPHxUXk@ E1qn$W0RR91 literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/fusion/static/__pycache__/conv2d_add_fuse_pass.cpython-37.pyc b/x2paddle/optimizer/fusion/static/__pycache__/conv2d_add_fuse_pass.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..be02a72df40c558b60425d3f113a96d0c28857ca GIT binary patch literal 948 zcmZ`%O^?$s5Vf6;hN=Y)E3wiF(Q7UV^o-E5i$LN~t+W^9i)F=bH;|;kPFJ)lr|y;i z07w1}XP=Nb@fSET=`KQiyV8J~^ELjvRH+aH35g#5zAenb$Qg4r{0iYQu< znsQ1pW)-Ug9{4z@!aCv+B|3UTRH&jGq9RRq!H)7+$2-CYD!wG?;5XzUDRtq6v=-*0 z%T9J>alUM>r>|Zr^?GHsf&Fm{=fdbkX*+ES{A)iL1gBtj0!|Z772!-#9w?@Q8^XgL zpGQ5SdvpK-(cOvl*rGsiC8Sbf4)6k@3ku0`;2eORgW3DwI`Wln$v0mEP()`bGf2rL zLX=I}34!vguIAr~xd#V-WL`I2Zdc7(Hgnl5o<7UpO)mhh(txe>?JEB?ZT)t>Y`d~9 zH`?UDs9ZLAPc=V8HEY*q0652B7MT#bncSf{+jA`ZMYxOFpzFOpOwo}Y`OLQLJ>618 z75l=l!=N;nxlpTx26@9;cEy$WP&NwGV+uD|w%W)}r-4au9W9J(uUv!`HNFnL!?)%l z*|ydC1VW40gvNC9_%6!+?ktV5z>Tc65N;?$y;LixpKv1~KCWc-?K_)7*7sd^I2-y$OY5` cReMZ>91gK`$@5;~iw8&*iH|6J!H6cKzso%EdH?_b literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/fusion/static/__pycache__/conv2d_add_fuser.cpython-37.pyc b/x2paddle/optimizer/fusion/static/__pycache__/conv2d_add_fuser.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d4f68697174b557131550c35a47a7535c0cbe9b7 GIT binary patch literal 4323 zcmd@X-;W!`d1iOLUi<9JB@Kj#mRbmUw_xn^k*6v~B)LL`;DnT@M5gI>8OBf&(WV(?{2~`@;SWQl~hqy+h=?{s&B3YQh#+n zh@7r7;G(|X3mxC9r*hTXcv5D;A`2$Cs0Dq2mOp_OD(mL*kk>mUJPjG zpoJGfTT=66NF#6rrft)q5-V3I%Q+;bOutDxl;yWoN+i}`1y*z@%*@Fdot4-eD|a+D zzYWeL*BnV_*#bLyTLHK%R~|b|BgINO>8#GmNNI}BkJ44pgB0wzq*a=trLZD!dE~aj z9!0yxdn}VA6Z66|BoNt{q!&UJ8$oQ_{rVRfB*CGrgYuik8k|zv-ifx*Zj+l%RL_bs$sqA zcbq7+WZ*EX(Y6|!4ePQ4K^#_^mDCAqsn6Bgg4+?d0$bQ!4vxOWFesXf>#lD{&#x?2 zE8&I>h&LO{%Z*FbiU-S@i{FzQ>}Gu3?zxc+D#v!w>zkPbo>#+A%dj;N7i99HrR%oa z z=9=$uFhnFAhL!-4!4k}IAwBxu7=dbayiRlR{F&E&vG_k6=pp)?mQ;>b4{3BIRqB7E z(X(*KGmK4C3MZLu%h|m94_9dF{u8B)Ju8c*PyX-&il!f)i~k=yY<1+XIuqcwvkAMOfmSv`Uzd1wgBs&Ey6nCy^Pjp?oewG`b^Vs)1ewauki+fjGP zD!KIIUn|4hP>=O18lzY9q7th+xj2Wl*R{$U@T`zNY|<&k>q%Mq9Dgg*sMhv_equ1r z0}17vh<8J2yF?BAz|2eaCOPSz2^cY}N5JX2cGTYBVN%KfD65f{G?nGBBqyQ!ZL#2Z zP(nwRH)c zcA&0St}0>$Acd(X34?de@061S^oVmh%)Q9+LsgCp7+nBJWd&G~m$)L}=3{MWT$_J` zY_c(ji-$PinmaL4)Y#l<0$*h($L7#7u^QfqkrL39Gbzs?HlT9XW9=lls*7N<;#!bd zvb6Tn!;(UxVJ;Gn!SN5if!}wN!G*)VAalqALp;&&at04r0+OjqW^#%{iqGfe`)^iaqS=P0l?WQyGu_vAa z#$p*e6ecNWkIgoBUCWxt3+Zbf7vU8lpRSn9Ld@z0EosI%BQLRLQQ%RMoY(6@)F@t% zJUM!Z*2WLfT9(kPG{l6!jNN)^KjBh z_;l}p*$7H%nr7ys!G?s_K*e(fPKX960WqGU=Pr7vkZ8Yd%XQYq^A4pkj|{ Zki#LCE`HKWe0mG1BJmN0kB?|P`U@--=Q;oY literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/fusion/static/__pycache__/prelu_fuser.cpython-37.pyc b/x2paddle/optimizer/fusion/static/__pycache__/prelu_fuser.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..6cf4ac5263870fdf5ef8b44bc31a614048d10dac GIT binary patch literal 4088 zcma)9O>7&-72erDE|;XF%C>AdMxd%cHUpdbaSAwvQ<{*FUqyy+qach6l&7KmO=~6+|=qNN^U7tYaf!8UJ3i9eJ)HBn*C9J_(=4gei4GmAd+#(j=5n zb<3)2RHOQqQrD?LO}q_SpcdZddy1~q3$Un|PTY#z$ZOnujlKGFVFhoHT&j2r4dMym zEG1GRk|>+%nkwXP6R}&!(3Vp(ltN&ky&~*(qHc#(wbTgX4&%H4=G1Z=FYqGA0hJyy zzlnFzaT>lGhK}BD=j4Qv~3UQi0NU&0g4z!lP^ zN*@kbQ`QL8JS8Di-y|!98rvFCa=0ns1qm+zXVK!eF~r)mM9UrlXXOaz*bp~<1UEs; z-~pU|>q8Z@$s-geGHWtiPIdBFDnLO63v;SdN2sQNhkY1<%c)KuqEhJ@u@gEi!{98! z;FlD_Z-4!VclJK`d~f^XNAGXR^LY5dXOBMp?cPUQU%vgWJ(K@3 z%wON#`{VAzzdz`7qt$ls@Gl=d`pqX_efqmcJMTRD{olU&?AJdW(hIgmJ9z7P$Mrj_ zZaA|vlf^bbrMu%%o}5Y$8z=g`_q?Erd%=)zmBAG4+T+I+QojIFKMcNU!P z$k=A=`%ktEb~QcUcLqX#aGcE75aJ_&u&WsdjR;BL`ZW>$7dDo%ZC&@=@I^|Glkwu+ z#~kS14ZjIb7~h53rXPE>=8L)%)|k&)EQs!VA#;E}Eal=%H#Y_Pjxz)98SMm%$m{ss zp_J^PM?#?C2|}EOr-*^vk&OI6j+2ARGLnMD&<^47LlQ>(!@@DBnzuyZC}W@`LUGt6 zr$wz)ms7HV!l#9isOPFj`50)jVH~s6T48}BqVj?ekQc1^BL(Yj?D-T?9wEeo+Eu^Z zaQ*OFZNS`n_UT!{g_*c)OfTrfQJ5CmaU>?aQqCs=bYBpy;KA>UA60dJ5)*z30}8uH zqSPLRX>WRzA%`N7Pk^OYk*R)3D!eqmFLu_ma~_1=zPM(3=bjcR*^7%8rs~ z39gANSAZ+@t&LiwQ?f|_ne7*oLZT`>R?cP#aV*(`=*#>Z z_Wttis-0G360rOEk?M_hr<+=ov5uS>Uc_2qYH+j|j8B3q)jI7C7Xg`?Sw&0DxI=M` zRL5i>q=xKjQiN-Ms~}@qZq!(1VQL2hRbg5|GZSTc9gn8voCfWSFhR-|a3c#g;-w-f zn%+jn0$D9Dk3`4{*ORJTeSe2!mJDmoAkbYB;F279fq7p=OP{%a_3MaP~gZ8*+r4->xsUmMj9oeFW9&Z zxGux(CkP0mZzkrNK~)67HOx$^O)LF^gz13&A~E|GRt0JRUX*YX@Yx~EUY|wyS&2>y z(+agtE3`Q{YdxN@3p?#S}X&*I#;Xy*AehN_VzfO!wM z2=gB+@H+bT7Gy8>&K_IWrL;joHt#cUWi{$qP?{lhDVHw^MWb6obuL#+sqTAWlw0B{5tSB*;57B1jEK$mWXzaaka2 zS@z|9TF2Ff?9Hn}mdX;5ma@w)Ti8Rd&8LRsNJ#@T(Nd)&1ur40nT#$HX{+o52Q!DsPC198(761UaViA1G*5OMFU+4#9C!+i_ZL8v9~i zcAWLt^)rn8=HcIgDf|U72=mfP{<+m)XfB=64sa<}1d(V!Q>vxASb+VJD7mS40^Rt`BASAG4|kg(C_tk&Ue q6xkLR_gZ%DlLran$n&KcBfGO%);3+u{BTXIWZz}Mx8fAwYwdp?AYs=4 literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/fusion/static/__pycache__/static_conv2d_add_fuse_pass.cpython-37.pyc b/x2paddle/optimizer/fusion/static/__pycache__/static_conv2d_add_fuse_pass.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..00da1e3b5bffffaf8dcae166a390b184e18f2671 GIT binary patch literal 955 zcmZ`%&2AGh5VpNPS)v3uRH9Oa$aODS$Qhy1R&a|d^@4n{thJLY>~0o&Q;{f~(re!V zj=TyIFW@UDUV#%c-lpP^dNt#}+4;Wt#%IG}pTMYo{t`SS#m^}lhh@vH_ zD5n(HtYlTd1HTT+u!?v@iH_bA6{_fls7TXYu%kTI@s9AGimyo8`vZAMN?mvS%mVBdI^4)6y713$R3{rB5 z5Jg?ILST7Tm$M(lJc5hAi>#{K%&zLStY@;GzkHQ_m|Ox}sR3K*`(5^B()iEk`s+y_D|mI%%7%ES~c*^#eo%RbUA zRaCKChK+`TgQ*L(nro=CU&%IKh)+eWKm<(Tddo%|+3GYf3GSo0kl0X7#3nSRo97Qv_Oa76MxAR}X(3!+h-#@;Afs>tAwI8Ud3O`y?By^|A@P5} z*b#&12H*A(n2o_DEM!UG&!%bOj`y7y-*;l%y+{;{j}BaSjT(22>cn_)Ak1f}WG&~Q e2UhJd4RScd(j~9E65l+52#drA6uw|Ulfgfk3i9#* literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/fusion/static/__pycache__/tf_batchnorm_fuse_pass.cpython-37.pyc b/x2paddle/optimizer/fusion/static/__pycache__/tf_batchnorm_fuse_pass.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..67a565f667bd59f180c17d1887735f0fe8290900 GIT binary patch literal 960 zcmZ`%J#Q015Z#Zv^I_RQflv@aiRiBJLFx!ajua#cD^e)XHLK-z?Ht?}yJrboPUW=x z2PpYB)LS7@@fWC=*~1Zqv)0Vc-j3(Jd9%~;ct~LVeDg!h144e`VYfmEPQmOHoFa;r zq@o$67_*X9JmWs*Wl)7#NQn+#6BVfNhNw`}E#J~C($SV=0~KA8Wbhm6kc7J6LRt&^ z!R!$@O){!T#uUxCVv668 zEa(|zVUOva>;*vv?wR%YqMM8N$e;!WVh$7p!siqKaN*9u&cN(_a2@$dH{_dF0{Wt} zgc+peVj+sU=!C!~tuE)^iMaQ)IbU)ciyD|enIsmiCh0yi$J3F83c;-G8vavsyf8Qe(>ByE` zvkm(|H&ju@zA&6MbjxQh&}yNf#bJf2iH}9CKmjb_22HDt>~z9SjOTD+WP9a8RMmJb zdWWyhg|cnS)l ViOwC_2J#5efBJgQ07`zS_n53qH5FtMdBk0KR%-I=r676_T&Hn literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/fusion/static/__pycache__/tf_batchnorm_fuser.cpython-37.pyc b/x2paddle/optimizer/fusion/static/__pycache__/tf_batchnorm_fuser.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..8cd68941251d5383dae3b59339ec3604ce52d644 GIT binary patch literal 4768 zcmdTIOKcm*b!UHEE|-)n%lbHu?Yc$^x(#JJY2gM&5V(c|6oFd623!yc7AwwBUV6D) z%`R<;9Rf)tAcsP=f!-1$klLa@1Vwsnisar?(c_#7B(}qQPrbD7%`Pd6jhmu96s54U z^Zsw%&b;?VSE^N4gWq4i`zQa*S2gWZ1jauTCa=Jg-hhE>ltfxHf0L#TupSv*vuP5< z8Ijeso3@&pQK{=Roo=~VRbk6aKcF;Xuks+dH3nV_wipLKa;}9M#y_K&q%XpRX-z`4rv8rB zCX`Tp$7veWpyrO&G^s^xcw4kY9eCUCXr|UIfk$Pza6JpMuyy0wl^|=~TJ7;J`odeJ z(2Z}jgHL)AhD^(d(6;rCj#hRF(K2IO6Pj;Q4Ky07(oU61@Yh|(RruLf!s?c^(tg4? zFM$&2_*p`5{NQ39D!=bJ5lA$j24xv`42}12{ zvQB7e*C3k0Itp7>*fLPuUD#uaQpvet1S<6&=rF5JN!RF1jtvRCXQ!yS#UoSLxttS5 zpcHzKPVwZD^HW%q1RRXOUq~LC;>jf!rm%}312`CgzmQy-;+dn%hcq9bqGLP&2O~%p zI-WViqtg>~IVAL1+?PV@7<;DB_i>N@|9*)ti^~7vOZ^vLf}VsdajM^kvvT^;y?dX3 z^5N*go1g#Utw-;D@@V&`kKf&yoXv;tef;>NpFDbh=hHXec25_-1M2VIe)OyThkv-2 zY-P84@x$M~|M?BgH3G?78k2{1Xenbeqh2eB z(w7@!6|zLPRIaQOUIs;jlX(tn4B zYxkpQq?io82)rCq8)V>HLBa-Z!31ftuJbvhkgXA}VBi%bo^J3=kw%!^Mwk=_GX=5R zkFqd{wni3`F!PNtsgE3FI?;~$A#FrhveE{NSeM1w-7sZ-+Fu)~#kp(o^UCq(&yTPk zk1+QRN(Bl+fF7^bEsj=Y8Qk=hI%%me@+zY3AW2wE>jtj@$d4heR0MeXQ&kUz{{*n6 ze}+M`9m6Fi(c!OFyd9EbW($lvq2wH6{8-{?uvf?-eU^Coe`;r}=gaKnb_Qqvo&DER zB@Wh)k((5*zXBk%Jxv$_3eCRpx;`|ARz~)8VeFHky?IsGd&a(YReR$S&`LvRvyqvU zY!kqo8J0y!n4-L_b*$Suza)%3dkkX&-nevA+j#D#mX)@(p{vG<8a=8HtDCP2(s2sy ztGRYjl^sP~VXXC^imz^1Y&Dq~B)`N2&{^}{Miy1C>{O}KkyQ9&=% z%*$tj_>O8>%vJssSo^gbbyrr^iDY++4QIA`$(D2|V~LtsVaB?tv^aDijL(CVG?HGz z@$5)D-}BP$ClpR=y{2}iwA8_taOjG|3&*vK{Rc~>8;?z;vI@OUmfQ71D!qaSx*K#s z=|;Sj9|aQ1YQkAN+{{HlS2rxq3Fu1u?I%fL>c@;xke@ypvtbMkPzK~;&G%p_={Xp% z2Xf)5!QT?G)t^b`;hea-1Goj4dBEY4!PlO$>!x*2HXddpOa-cAiXeKBdC4BxhwM8v zh1t$ zSd^l}lAq zpRA3Ci{U?2KfVkb_+OX1yny~4ep2@@3}%b^nL+wUlU~XOM}IiI^(#}?#HbR?m0o!P+gLb7aY)`3cSbraU90$ zewPKYtc;;M0hjh&7OvmQq`4Lb>A(V))?(?l!zl9a1W_NZZpsaQ75zGBCQ2+x!>AX7 z8!t2s|NI3UF5;lT7wa<|yDDiypi{Wi^>`+YFwS@b^`J6YW^|pUJ~S`V$){nQTfojj z+7J-Pf~t4n#}Q+}Es>|no%&K<_`iwT@B++VKn}>F(0D+i^YLEEzm0^Kkl>AWdeH3#2c*gqu&jjf0mWToS-}W-7$TgL_;YBv4E<6LOIP(cs%w<) zI9W?$Dp7ah38YEy35-~Oxn5Jt4+n4bSTL(Q-c!uDJJ8v!?Cie0BeJkv>?C55ef{2PFi zR^0DG>D1#yR$9F%V(N6Hd9gk-{x?Bm@^6Ah(LMTk<>eeZ-{!FiW2?tmBjG)$=UvsB naIL1E!)VH)L literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/fusion/static/bn_scale_fuser.py b/x2paddle/optimizer/fusion/static/bn_scale_fuser.py index 9ae04f7..05d1780 100644 --- a/x2paddle/optimizer/fusion/static/bn_scale_fuser.py +++ b/x2paddle/optimizer/fusion/static/bn_scale_fuser.py @@ -79,7 +79,6 @@ class Static_BNScaleFuser(FuseBase): graph.layers[new_layer_id] = new_layer matches.pop(new_layer_id) - def gen_new_layer(self, parameters, matches): layers_id = list(matches.keys()) layer = matches[layers_id[0]] diff --git a/x2paddle/optimizer/fusion/static/conv2d_add_fuse_pass.py b/x2paddle/optimizer/fusion/static/conv2d_add_fuse_pass.py new file mode 100644 index 0000000..ee51b4b --- /dev/null +++ b/x2paddle/optimizer/fusion/static/conv2d_add_fuse_pass.py @@ -0,0 +1,33 @@ +# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved. +# +# Licensed under the Apache License, Version 2.0 (the "License" +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +from x2paddle.optimizer.pass_ import Pass +from x2paddle.optimizer.fusion.static import StaticConv2DAddFuser +from x2paddle.optimizer.pass_manager import pass_register + + +@pass_register +class StaticConv2DAddFusePass(Pass): + name = "static_conv2d_add_fuse_pass" + + def __init__(self): + Pass.__init__(self) + + def apply(self, graph): + fuser = StaticConv2DAddFuser() + fuser.operate(graph, match_kind="edge") + + +# 用于注册 +static_conv2d_add_fuse_pass = StaticConv2DAddFusePass() \ No newline at end of file diff --git a/x2paddle/optimizer/fusion/static/conv2d_add_fuser.py b/x2paddle/optimizer/fusion/static/conv2d_add_fuser.py new file mode 100644 index 0000000..cb3912f --- /dev/null +++ b/x2paddle/optimizer/fusion/static/conv2d_add_fuser.py @@ -0,0 +1,121 @@ +# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved. +# +# Licensed under the Apache License, Version 2.0 (the "License" +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import copy +import numpy as np +from x2paddle.optimizer.pattern_matcher import FuseBase +from x2paddle.core.program import PaddleGraph, PaddleLayer +from x2paddle.core.util import * + + +class StaticConv2DAddFuser(FuseBase): + def __init__(self): + super(StaticConv2DAddFuser, self).__init__(graph_type="static") + self.patterns = list() + + def build_pattern(self): + """ 描述需要替换的conv2d+add图结构。 + conv2d+add层模式python实现代码示例: + 模式一： + MobilenetV1_Logits_Conv2d_1c_1x1_biases = paddle.static.create_parameter(dtype='float32', shape=[1001], name='MobilenetV1_Logits_Conv2d_1c_1x1_biases', default_initializer=paddle.nn.initializer.Constant(value=0.0)) + conv2d_transpose_14 = paddle.transpose(x=MobilenetV1_Logits_AvgPool_1a_AvgPool, perm=[0, 3, 1, 2]) + MobilenetV1_Logits_Conv2d_1c_1x1_Conv2D = paddle.nn.functional.conv2d(x=conv2d_transpose_14, weight=MobilenetV1_Logits_Conv2d_1c_1x1_weights, bias=None, stride=[1, 1], dilation=[1, 1], padding='SAME') + MobilenetV1_Logits_Conv2d_1c_1x1_Conv2D = paddle.transpose(x=MobilenetV1_Logits_Conv2d_1c_1x1_Conv2D, perm=[0, 2, 3, 1]) + MobilenetV1_Logits_Conv2d_1c_1x1_BiasAdd = paddle.add(x=MobilenetV1_Logits_Conv2d_1c_1x1_Conv2D, y=MobilenetV1_Logits_Conv2d_1c_1x1_biases) + + 模式二： + MobilenetV1_Logits_Conv2d_1c_1x1_biases = paddle.static.create_parameter(dtype='float32', shape=[1001], name='MobilenetV1_Logits_Conv2d_1c_1x1_biases', default_initializer=paddle.nn.initializer.Constant(value=0.0)) + MobilenetV1_Logits_Conv2d_1c_1x1_Conv2D = paddle.nn.functional.conv2d(x=conv2d_transpose_14, weight=MobilenetV1_Logits_Conv2d_1c_1x1_weights, bias=None, stride=[1, 1], dilation=[1, 1], padding='SAME') + MobilenetV1_Logits_Conv2d_1c_1x1_BiasAdd = paddle.add(x=MobilenetV1_Logits_Conv2d_1c_1x1_Conv2D, y=MobilenetV1_Logits_Conv2d_1c_1x1_biases) + """ + + def gen_name(id): + return "x" + str(id) + + pattern = PaddleGraph(graph_type="dygraph") + pattern.add_layer( + "paddle.static.create_parameter", + inputs={}, + outputs=[gen_name(0)]) + pattern.add_layer( + kernel="paddle.transpose", + inputs={"x": "conv-input-0"}, + outputs=[gen_name(1)], + perm=[0, 3, 1, 2]) + pattern.add_layer( + kernel="paddle.nn.functional.conv2d", + inputs={"input": gen_name(1), + "weight": "conv-input-1"}, + outputs=[gen_name(2)]) + pattern.add_layer( + kernel="paddle.transpose", + inputs={"x": gen_name(2)}, + outputs=[gen_name(2)], + perm=[0, 2, 3, 1]) + pattern.add_layer( + kernel="paddle.add", + inputs={"x": gen_name(2), + "y": gen_name(0)}, + outputs=[gen_name(3)]) + pattern.build(inputs={"input-0": "conv-input-0", + "input-1": "conv-input-1"}) + self.patterns.append(pattern) + + pattern = PaddleGraph(graph_type="dygraph") + pattern.add_layer( + "paddle.static.create_parameter", + inputs={}, + outputs=[gen_name(0)]) + pattern.add_layer( + kernel="paddle.nn.functional.conv2d", + inputs={"input": "conv-input-0", + "weight": "conv-input-1"}, + outputs=[gen_name(1)]) + pattern.add_layer( + kernel="paddle.add", + inputs={"x": gen_name(1), + "y": gen_name(0)}, + outputs=[gen_name(2)]) + pattern.build(inputs={"input-0": "conv-input-0", + "input-1": "conv-input-1"}) + self.patterns.append(pattern) + + def insert_new_layer(self, graph, parameters, matches): + self.gen_new_layer(matches, graph) + matches_copy = copy.deepcopy(matches) + for layer_id, layer in matches_copy.items(): + if layer.kernel not in ["paddle.add"]: + matches.pop(layer_id) + + def gen_new_layer(self, matches, graph): + is_transpose = False + for layer_id, layer in matches.items(): + if layer.kernel == "paddle.static.create_parameter": + bias_name = layer.attrs["name"][1: -1] + if layer.kernel == "paddle.transpose": + is_transpose = True + if layer.kernel == "paddle.add": + output_name = layer.outputs[0] + if layer.kernel == "paddle.nn.functional.conv2d": + conv_id = layer_id + for layer_id, layer in matches.items(): + if layer.kernel == "paddle.nn.functional.conv2d": + layer.inputs["bias"] = bias_name + layer.attrs.pop("bias") + if not is_transpose: + layer.outputs[0] = output_name + if layer.kernel == "paddle.transpose": + if conv_id in graph.edges_in[layer_id]: + layer.outputs[0] = output_name + diff --git a/x2paddle/optimizer/fusion/static/prelu_fuse_pass.py b/x2paddle/optimizer/fusion/static/prelu_fuse_pass.py new file mode 100644 index 0000000..bdd65e8 --- /dev/null +++ b/x2paddle/optimizer/fusion/static/prelu_fuse_pass.py @@ -0,0 +1,33 @@ +# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved. +# +# Licensed under the Apache License, Version 2.0 (the "License" +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +from x2paddle.optimizer.pass_ import Pass +from x2paddle.optimizer.fusion.static import StaticPReLUFuser +from x2paddle.optimizer.pass_manager import pass_register + + +@pass_register +class StaticPReLUFusePass(Pass): + name = "static_prelu_fuse_pass" + + def __init__(self): + Pass.__init__(self) + + def apply(self, graph): + fuser = StaticPReLUFuser() + fuser.operate(graph, match_kind="edge") + + +# 用于注册 +static_prelu_fuse_pass = StaticPReLUFusePass() \ No newline at end of file diff --git a/x2paddle/optimizer/fusion/static/prelu_fuser.py b/x2paddle/optimizer/fusion/static/prelu_fuser.py new file mode 100644 index 0000000..c87abdf --- /dev/null +++ b/x2paddle/optimizer/fusion/static/prelu_fuser.py @@ -0,0 +1,139 @@ +# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved. +# +# Licensed under the Apache License, Version 2.0 (the "License" +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import copy +import numpy as np +from collections import OrderedDict +from x2paddle.optimizer.pattern_matcher import FuseBase +from x2paddle.core.program import PaddleGraph, PaddleLayer +from x2paddle.core.util import * + + +class StaticPReLUFuser(FuseBase): + def __init__(self): + super(StaticPReLUFuser, self).__init__(graph_type="static") + + def build_pattern(self): + """ 描述需要替换的prelu图结构。 + prelu层模式python实现代码示例: + conv4_alphas = paddle.static.create_parameter(dtype='float32', shape=[128], name='conv4_alphas', default_initializer=paddle.nn.initializer.Constant(value=0.0)) + conv4_mul_1_y = paddle.full(dtype='float32', shape=[1], fill_value=0.5) + conv4_Relu = paddle.nn.functional.relu(x=conv4_BiasAdd) + conv4_Abs = paddle.abs(x=conv4_BiasAdd) + conv4_sub = fluid.layers.elementwise_sub(x=conv4_BiasAdd, y=conv4_Abs) + conv4_mul = paddle.multiply(x=conv4_alphas, y=conv4_sub) + conv4_mul_1 = paddle.multiply(x=conv4_mul, y=conv4_mul_1_y) + conv4_add = paddle.add(x=conv4_Relu, y=conv4_mul_1) + """ + + def gen_name(id): + return "x" + str(id) + + self.pattern.add_layer( + "paddle.static.create_parameter", + inputs={}, + outputs=[gen_name(0)]) + self.pattern.add_layer( + "paddle.full", + inputs={}, + outputs=[gen_name(1)], + shape=[1], + fill_value=0.5) + self.pattern.add_layer( + "paddle.nn.functional.relu", + inputs={"x": "prelu-input-0"}, + outputs=[gen_name(2)]) + self.pattern.add_layer( + "paddle.abs", + inputs={"x": "prelu-input-0"}, + outputs=[gen_name(3)]) + self.pattern.add_layer( + "fluid.layers.elementwise_sub", + inputs={"x": "prelu-input-0", + "y": gen_name(3)}, + outputs=[gen_name(4)]) + self.pattern.add_layer( + "paddle.multiply", + inputs={"x": gen_name(0), + "y": gen_name(4)}, + outputs=[gen_name(5)]) + self.pattern.add_layer( + "paddle.multiply", + inputs={"x": gen_name(5), + "y": gen_name(1)}, + outputs=[gen_name(6)]) + self.pattern.add_layer( + "paddle.add", + inputs={"x": gen_name(2), + "y": gen_name(6)}, + outputs=[gen_name(7)]) + self.pattern.build(inputs={"input-0": "prelu-input-0", }) + + + def insert_new_layer(self, graph, parameters, matches): + new_layers, last_layer_id = self.gen_new_layer(matches, parameters, graph) + matches_copy = copy.deepcopy(matches) + for layer_id, layer in matches_copy.items(): + for i in range(4): + if layer_id == new_layers[i].id: + matches.pop(new_layers[i].id) + prefix_layers = OrderedDict() + mid_layers = OrderedDict() + suffix_layers = OrderedDict() + is_need_id = False + for layer_id, layer in graph.layers.items(): + if is_need_id: + suffix_layers[layer_id] = layer + else: + if layer_id == last_layer_id: + for i in range(4): + mid_layers[new_layers[i].id] = new_layers[i] + is_need_id = True + prefix_layers[layer_id] = layer + prefix_layers.update(mid_layers) + prefix_layers.update(suffix_layers) + graph.layers = prefix_layers + + def gen_new_layer(self, matches, parameters, graph): + layer_id_list = list(matches.keys()) + layer_id_list.sort(key = int) + for layer_id, layer in matches.items(): + if layer.kernel == "paddle.nn.functional.relu": + input_name = layer.inputs["x"] + if layer.kernel == "paddle.static.create_parameter": + param_layer = layer + param_name = layer.outputs[0] + if layer.kernel == "paddle.add": + output_name = layer.outputs[0] + transpose0 = PaddleLayer( + id=layer_id_list[-1] + "_1", + kernel="paddle.transpose", + inputs={"x": input_name}, + outputs=["{}_transpose_for_prelu".format(input_name)], + perm=[0, 3, 1, 2]) + param = parameters[param_name] + c = param.shape[0] + prelu = PaddleLayer(id=layer_id_list[-1] + "_2", + kernel="paddle.nn.functional.prelu", + inputs={"x": "{}_transpose_for_prelu".format(input_name), + "weight": param_name}, + outputs=["{}_prelu".format(input_name)]) + transpose1 = PaddleLayer( + id=layer_id_list[-1] + "_3", + kernel="paddle.transpose", + inputs={"x": "{}_prelu".format(input_name)}, + outputs=[output_name], + perm=[0, 2, 3, 1]) + return [param_layer, transpose0, prelu, transpose1], layer_id_list[-1] + diff --git a/x2paddle/optimizer/fusion/static/tf_batchnorm_fuse_pass.py b/x2paddle/optimizer/fusion/static/tf_batchnorm_fuse_pass.py new file mode 100644 index 0000000..36d58ba --- /dev/null +++ b/x2paddle/optimizer/fusion/static/tf_batchnorm_fuse_pass.py @@ -0,0 +1,33 @@ +# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved. +# +# Licensed under the Apache License, Version 2.0 (the "License" +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +from x2paddle.optimizer.pass_ import Pass +from x2paddle.optimizer.fusion.static import StaticTFBatchNormFuser +from x2paddle.optimizer.pass_manager import pass_register + + +@pass_register +class StaticTFBatchNormFusePass(Pass): + name = "static_tf_batchnorm_fuse_pass" + + def __init__(self): + Pass.__init__(self) + + def apply(self, graph): + fuser = StaticTFBatchNormFuser() + fuser.operate(graph, match_kind="edge") + + +# 用于注册 +static_tf_batchnorm_fuse_pass = StaticTFBatchNormFusePass() \ No newline at end of file diff --git a/x2paddle/optimizer/fusion/static/tf_batchnorm_fuser.py b/x2paddle/optimizer/fusion/static/tf_batchnorm_fuser.py new file mode 100644 index 0000000..796556b --- /dev/null +++ b/x2paddle/optimizer/fusion/static/tf_batchnorm_fuser.py @@ -0,0 +1,227 @@ +# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved. +# +# Licensed under the Apache License, Version 2.0 (the "License" +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import copy +import numpy as np +from collections import OrderedDict +from x2paddle.optimizer.pattern_matcher import FuseBase +from x2paddle.core.program import PaddleGraph, PaddleLayer +from x2paddle.core.util import * + + +class StaticTFBatchNormFuser(FuseBase): + def __init__(self): + super(StaticTFBatchNormFuser, self).__init__(graph_type="static") + self.patterns = list() + + def build_pattern(self): + """ 描述需要替换的batchnorm图结构。 + batchnorm层模式python实现代码示例: + + """ + + def gen_name(id): + return "x" + str(id) + + pattern = PaddleGraph(graph_type="dygraph") + pattern.add_layer( + "paddle.static.create_parameter", + inputs={}, + outputs=[gen_name(0)]) + pattern.add_layer( + "paddle.full", + inputs={}, + outputs=[gen_name(1)], + shape=[1]) + pattern.add_layer( + "paddle.add", + inputs={"x": gen_name(0), "y": gen_name(1)}, + outputs=[gen_name(2)]) + pattern.add_layer( + "paddle.rsqrt", + inputs={"x": gen_name(2)}, + outputs=[gen_name(3)]) + pattern.add_layer( + "paddle.static.create_parameter", + inputs={}, + outputs=[gen_name(4)]) + pattern.add_layer( + "paddle.multiply", + inputs={"x": gen_name(3), "y": gen_name(4)}, + outputs=[gen_name(5)]) + pattern.add_layer( + "paddle.static.create_parameter", + inputs={}, + outputs=[gen_name(6)]) + pattern.add_layer( + "paddle.multiply", + inputs={"x": gen_name(6), "y": gen_name(5)}, + outputs=[gen_name(7)]) + pattern.add_layer( + "paddle.static.create_parameter", + inputs={}, + outputs=[gen_name(8)]) + pattern.add_layer( + "fluid.layers.elementwise_sub", + inputs={"x": gen_name(8), "y": gen_name(7)}, + outputs=[gen_name(9)]) + pattern.add_layer( + "paddle.multiply", + inputs={"x": "bn-input-0", "y": gen_name(5)}, + outputs=[gen_name(10)]) + pattern.add_layer( + "paddle.add", + inputs={"x": gen_name(10), "y": gen_name(9)}, + outputs=[gen_name(11)]) + pattern.build(inputs={"input-0": "bn-input-0", }) + self.patterns.append(pattern) + + pattern = PaddleGraph(graph_type="dygraph") + pattern.add_layer( + "paddle.static.create_parameter", + inputs={}, + outputs=[gen_name(0)]) + pattern.add_layer( + "paddle.full", + inputs={}, + outputs=[gen_name(1)], + shape=[1]) + pattern.add_layer( + "paddle.add", + inputs={"x": gen_name(0), "y": gen_name(1)}, + outputs=[gen_name(2)]) + pattern.add_layer( + "paddle.rsqrt", + inputs={"x": gen_name(2)}, + outputs=[gen_name(3)]) + pattern.add_layer( + "paddle.static.create_parameter", + inputs={}, + outputs=[gen_name(4)]) + pattern.add_layer( + "paddle.multiply", + inputs={"x": gen_name(3), "y": gen_name(4)}, + outputs=[gen_name(5)]) + pattern.add_layer( + "paddle.multiply", + inputs={"x": "bn-input-0", "y": gen_name(5)}, + outputs=[gen_name(10)]) + pattern.add_layer( + "paddle.static.create_parameter", + inputs={}, + outputs=[gen_name(6)]) + pattern.add_layer( + "paddle.multiply", + inputs={"x": gen_name(6), "y": gen_name(5)}, + outputs=[gen_name(7)]) + pattern.add_layer( + "paddle.static.create_parameter", + inputs={}, + outputs=[gen_name(8)]) + pattern.add_layer( + "fluid.layers.elementwise_sub", + inputs={"x": gen_name(8), "y": gen_name(7)}, + outputs=[gen_name(9)]) + pattern.add_layer( + "paddle.add", + inputs={"x": gen_name(10), "y": gen_name(9)}, + outputs=[gen_name(11)]) + pattern.build(inputs={"input-0": "bn-input-0", }) + self.patterns.append(pattern) + + def insert_new_layer(self, graph, parameters, matches): + new_layers, last_layer_id = self.gen_new_layer(matches, parameters, graph) + matches_copy = copy.deepcopy(matches) + for layer_id, layer in matches_copy.items(): + for i in range(7): + if layer_id == new_layers[i].id: + matches.pop(new_layers[i].id) + prefix_layers = OrderedDict() + mid_layers = OrderedDict() + suffix_layers = OrderedDict() + is_need_id = False + for layer_id, layer in graph.layers.items(): + if is_need_id: + suffix_layers[layer_id] = layer + else: + if layer_id == last_layer_id: + for i in range(7): + mid_layers[new_layers[i].id] = new_layers[i] + is_need_id = True + prefix_layers[layer_id] = layer + prefix_layers.update(mid_layers) + prefix_layers.update(suffix_layers) + graph.layers = prefix_layers + + def gen_new_layer(self, matches, parameters, graph): + layer_id_list = list(matches.keys()) + layer_id_list.sort(key = int) + for layer_id, layer in matches.items(): + if layer.kernel == "paddle.full": + full_layer = layer + out_layer_id = graph.edges_out[layer_id][0] + if matches[out_layer_id].kernel == "paddle.add": + var_layer_id = graph.edges_in[out_layer_id][0] + var_layer = matches[var_layer_id] + if layer.kernel == "paddle.rsqrt": + out_layer_id = graph.edges_out[layer_id][0] + if matches[out_layer_id].kernel == "paddle.multiply": + gamma_layer_id = graph.edges_in[out_layer_id][1] + gamma_layer = matches[gamma_layer_id] + if layer.kernel == "fluid.layers.elementwise_sub": + in_layer_id = graph.edges_in[layer_id][0] + beta_layer = matches[in_layer_id] + in_layer_id = graph.edges_in[layer_id][1] + in_layer_id = graph.edges_in[in_layer_id][0] + mean_layer = matches[in_layer_id] + out_layer_id = graph.edges_out[layer_id][0] + add_layer = matches[out_layer_id] + if layer.kernel == "paddle.multiply": + in_layer_id = graph.edges_in[layer_id][1] + mul_layer = matches[in_layer_id] + if mul_layer.kernel == "paddle.multiply": + in_layer_id = graph.edges_in[layer_id][0] + if in_layer_id not in matches: + input_name = layer.inputs["x"] + transpose0 = PaddleLayer( + id=layer_id_list[-1] + "_1", + kernel="paddle.transpose", + inputs={"x": input_name}, + outputs=["{}_transpose_for_bn".format(input_name)], + perm=[0, 3, 1, 2]) + params = parameters[gamma_layer.outputs[0]] + c = params.shape[0] + bn = PaddleLayer( + id=layer_id_list[-1] + "_2", + kernel="paddle.nn.functional.batch_norm", + inputs={"x": "{}_transpose_for_bn".format(input_name), + "running_mean": mean_layer.outputs[0], + "running_var": var_layer.outputs[0], + "weight": gamma_layer.outputs[0], + "bias": beta_layer.outputs[0]}, + outputs=["{}_bn".format(input_name)], + epsilon=full_layer.attrs["fill_value"]) + transpose1 = PaddleLayer( + id=layer_id_list[-1] + "_3", + kernel="paddle.transpose", + inputs={"x": "{}_bn".format(input_name)}, + outputs=add_layer.outputs, + perm=[0, 2, 3, 1]) + mean_layer.id = layer_id_list[-1] + "_01" + var_layer.id = layer_id_list[-1] + "_02" + gamma_layer.id = layer_id_list[-1] + "_03" + beta_layer.id = layer_id_list[-1] + "_04" + return [mean_layer, var_layer, gamma_layer, beta_layer, + transpose0, bn, transpose1], layer_id_list[-1] + diff --git a/x2paddle/optimizer/optimizer.py b/x2paddle/optimizer/optimizer.py index 69deedd..30557f6 100644 --- a/x2paddle/optimizer/optimizer.py +++ b/x2paddle/optimizer/optimizer.py @@ -16,13 +16,13 @@ from x2paddle.optimizer.pass_manager import PassManager from x2paddle.optimizer.fusion.dygraph import * from x2paddle.optimizer.fusion.static import * from x2paddle.optimizer.elimination.dygraph import * +from x2paddle.optimizer.elimination.static import * class GraphOptimizer(object): def __init__(self, source_frame, paddle_type="dygraph", jit_type="trace"): if source_frame == "pytorch": if jit_type == "trace": - self.passes = ["dygraph_constant_fuse_pass", - "trace_fc_fuse_pass"] + self.passes = ["trace_fc_fuse_pass"] else: self.passes = [ "dygraph_constant_fuse_pass", @@ -39,12 +39,20 @@ class GraphOptimizer(object): else: self.passes = ["static_bn_scale_fuse_pass"] elif source_frame == "tf": - self.passes = [ - "dygraph_conv2d_add_fuse_pass", - "dygraph_tf_batchnorm_fuse_pass", - "dygraph_prelu_fuse_pass", - "transpose_eliminate_pass" - ] + if paddle_type == "dygraph": + self.passes = [ + "dygraph_conv2d_add_fuse_pass", + "dygraph_tf_batchnorm_fuse_pass", + "dygraph_prelu_fuse_pass", + "transpose_eliminate_pass" + ] + else: + self.passes = [ + "static_conv2d_add_fuse_pass", + "static_tf_batchnorm_fuse_pass", + "static_prelu_fuse_pass", + "static_transpose_eliminate_pass" + ] else: self.passes = [] diff --git a/x2paddle/optimizer/tensorflow/__pycache__/__init__.cpython-37.pyc b/x2paddle/optimizer/tensorflow/__pycache__/__init__.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..ac7f8f6b99bc543360a207ecc049ba4d6cb78cd5 GIT binary patch literal 169 zcmZ?b<>g`kg5Xyt<3aRe5CH>>K!yVl7qb9~6oz01O-8?!3`HPe1o10KKR2&LzqmB7 zGBGbLF)!V~P(Q*bATcE+CpA7fKP45;)2}cp01NBq7nEe?W>%#Z>6fJD73UYF<>Z&^ f$H!;pWtPOp>lIYq;;_lhPbtkwwFBAp8HgDG1&l0D literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/tensorflow/__pycache__/batch_norm.cpython-37.pyc b/x2paddle/optimizer/tensorflow/__pycache__/batch_norm.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..c70b0cd7a274e10c9dfa06ce816cccdcf2fa4bab GIT binary patch literal 3649 zcmaJ^&2QYs73c74x!e`CT1&E5mL2&kfi9A*q()J=L68dVrBDJSs0*S7LG6rIlt}Js z$d%;{_fqB1O9~x|9t-K%YvDhj=%GMwMUm578x&9vMbAAnd2cwo;>Jn{4Ees_@4fjT z`QhB0tHJfjzyB6|x~ggaL}qdsKzslr+kzX_D2X*6pL$FZ-PZ}yMrYr=AK@Doc6S81z(flyXyZbwRo_sXOy96BY!Y7WOz{pIvQ4Org&L80>%#6i* z+ybe*{qNJwB+WOoVfrde_rkRI_{rukw@Pe5w@=xW+dSDCNXVQ0K^`U1i1E#wrCFcv z#r>C?yND!60m{bUm2iR}N~1goYKQ?#(#UA3igsgS(Un>-cpJt%3<>+St!X*5`;-tZ zC+B)W4vY&j*3aydvqC>NFSKp#%zp9fU+H6GY!>E$MfF32KRsFl+CHygqb~HZRan#n z`7hvU;S==az$uK5HnuTlVN<)K0asI8O>$0AgNP1z{02N6pzA>Y4(Qr7(O;|Ruchsq z*}nw64gOyqt>tydbygltgKP9vVy8q7T0L7K=jBroko}k4jMp0MTUB#8y8%3kU z4t_rczs8%0L32=ILaLIvPOmb%hX!b|e(acy8t79~=4z_UbBZfuPT{k#7fON*Jrd>xS(5SI$wkMa|aE?0T`NxHITh0u{J02j&?9No`;UlmpulH z$^17}H7$w1rK(v_Tv<)KSWwmI{I?LfeN8nw|C6c)s0v;gai*HT&uFiz!7l6kKQg`t zbLC0?|G+QIRA=zEs!qvebvKF|#XNWz{IS$JMN3t-C~H|%l`Sc*tZcbhg36lGZy)@Y zRh?;peNBLZW=Q4~AjN33)54_&oDH-2AdB3>3t-W|UeoNN&){nIUDd*5Wvp7Yn zn8;iy+Xk77vNP*w`+4=$UCUdiu*(8C4^F&dO`X2vg$AeZz486xUg64hYw^Dn_hq(w z&oIg#uDE=Gdgtxp-UXrav{l?M`w$Vj&?D5Q7OhdYXVUf;*7yO~-cYuSbcrrR1kUsZ zRA$rVjz({SPOT@1hMYPF6sPuw;#6mWyY)qV{4ifA42b|D;@p?90fCDc6oN;^16jp` z1MituJe2f9Nk5YGBc!pu=i1TQbLh?=x3Mu=-HV42 z-H7q6pKY+1B`nQfMi~p>)hf*JB+4$ohU*f)J2<|?FY)$0e#tQY9&I+y zE@xrRf}%rz{qBln#b@CTjg*E(vqMcp~F&jc(4wz522cmh=+VaTpKT z$b9Fs@jL5vjvo!Yg|`K~wc#deY>-89KNYSlG6?gWi~4R9W+mB5`o~e)3lbKlK?+?E zi%N1F@+eHZOiI@mg*D(&nhUer$L|`-*g#TtIfno#XGtd03k$Pk!sxMFIE?mK246wK zh*>J?B@5M-_)1vV`%GBbehAfL8N$d|&T+;#t^i>}^O7(Z*6uKh>H6CoK8rMA^>{eg z=Xa5IB#r=DYe|aNLF&7*Q6SA8`z|bzKo1u>L`YxcJ@=vb5k~x0H3WE<* z{8PxD$}en(G|4h?iB3GyBD(Iv(}TN?GF{3_4Os!%s+55Rc%jGQcgxV0wB_)w_mUQ% zQ&wfZNoeJj?ZPFTw=<~9mvxJ8tMm)PHdnZ?1QRM zpX!@nw@gN@nG+Ejx&sFT{4L-X3LDHwg80lsZ91X*`)NltH&wq%w>Ay=!;nZ;a z%fYQ7e|H?>7*eusbNq=CFzs>7x_Jc0m)ug!8P# O_C8L3hUu^=P2(FwwW}fk literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/tensorflow/__pycache__/bias.cpython-37.pyc b/x2paddle/optimizer/tensorflow/__pycache__/bias.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..263afc63abbb8483f0870c64d0cff2c5a254197b GIT binary patch literal 1798 zcmZuxy>BB$6rYdX_4+f03sG_kQd$GDAtj1VQ8*o)Iq8a=ETYxonK)~&*Xx<}onx&V zaurp|D;xea$2fTQBwf#&DthGjJ3~ie(!r`-aD(;D+Jc7-+u`{tr7AYZZ?Yp za|)k+3Pcb=W6~pE6Tt+3M+CoNJu2uc(sJ%WPP%=ho5g@Rg-;&>kpwhJj|nQ+JJREV z3kUuk;exeiN@r21ze%$`1zlYDj}E?qPd9*|;eh2NfHc5_cNb4C<5?uyad<72Zuck2 zn`5!~_ahiIy#Ds;C$gMzD3Xj4yv8%7qYrc01!F>*!f>9VVAyGWh(9&gz(xmR%&@vL#Mp(96Fm1;^&IltCxBpdo|mj2AGCj6`vGL> zLvH@CD#{1M=Ec^a=L2dr8~G~%elzSyLtIJxp=Xx&YxNRD)`zd?)>5txbfPup4T_G zID$8!c60Cqr)@R!+6lak>FznXrPII8$y$aPK~%_S`$E8`w!L?6t%nsAdWBE4+Y!wZ zGCevky6=PTVi&sAzWgtbHV?YpUhF+;zir>$V>>G(i^E46GL~bRWN#xa186AYg!4$> z{|)QEovdV-$si3?IF^}I#?@D0Dvgs$HNM9Vy2n{%%Gi5Bm}TmPaT3rqblz$jH&s!R z8K*zNwiHsPmh*}^VB9E^V{N>VREdn0^~7VDn6eZDse{QZGkhR3QvxoXW!ktHLL0x3 zM+vw{gq)kQka00xj29)S+G;2VGjj(jOs~{O$on>K5DDX2u2%!Ug&Bg?DA{g$bz9Na zpz`l_ErN5y1a)txsJ+-iXu~mYIx_T)oo@xf31hgLB{=o)}Rg` zjP3(pp&u4I%Q3nOSPdn#fjn4QJ&W)2JwThZLMx#C6u3j$1P-x9^{kljZHV1%xi}s1 zf`Y(Qf?zxmGnlf0sRhAw7RH4{l^{FSLV_do0l_L1UOS$xM&eW7wMdJpCmZPl3+uoZ Q@1GRWp5cSEW7+5b0O7C72><{9 literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/tensorflow/__pycache__/prelu.cpython-37.pyc b/x2paddle/optimizer/tensorflow/__pycache__/prelu.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..b65008fa6087a2a94da867002140978f4fe55e37 GIT binary patch literal 2861 zcmaJ@&2Jk;6yMn|uQ$%taT4dFMd|^mghRfNDg*^pP{E~Y5nn|X)oSaRxSOoko7r)j zy6X$!zy&En`~k_48;AZAaN>yOzy%2@h`)dX@SE9b|$ZwZ#> z1HP9G9&crwCBx0^iS)uSPGb>skYd_&}b>(3Z(s*5JQkt zvmpD{85x@=_R-D4Jhjh^TgHiffA>psY>n;0-gl^ZVDamRS3$d{HIhR;A z)=IxUW^{-$rTeQwb%90s@Zh1u;%j7n#m!Y*iMmnD0h@ndoe;hYFO3o@ zxSN`L0V^~VXkBeVr!Z;%&oN=$S5w-$%2K689-%$0F#A>h<~6SQiVG^P+`*+H0IoT0 z_bco!foo{)56!)Z(&T=b(-TlrIJEyMlqP>|Mmv*N*nR;2!Ty{=1w@U58f2Xc`Fxep zrpDT=kS|nRiM&;`fOQ^NP5#|fq^7c+!SN%s^QgT5l@p+10-Q{M^9T?>z>WiW3VPl# z8805aTP)r(?q44-=~zqe7=@y|0~b;LQB-ju^R?_Z$b2O`F%N&cH+>(L#ie8HEX0W% zy;>|+_k!GS<6gAKD@D6l*bji-;$IdkI$HaNabg|5`J59?8k{Z^?K46z(PptyW-BIi zp+~5TMpmP(o=q3Ob;he$yHu@RrcJsK6YO|Sv}hYuz5-2^Yxr9I)*Y{jW?^ZsSf%n? zu7}r_@>_XbFIF|O)&1axQLJftP1Ea|Ue`2E@51|jj@uhaG!8l&!pYj;M==$X2Yy>)fcoi;Z|SN4)&Og9tNne$DSus%!0 z<2Yv_x}UV8qc}hR7r%4$uHdTAo~voJFk`2>;Y(cghskJt#$Yc_5;WX27f~ukugvJW zoJE3#S;V706O6yaaXx%7YR(u(5Au=ybR6!C+B0&&qcqP3Ig{k5vpC_xy#Wt(Ge1`y zdc%=sOr?9kc*+v##%VSbxvUL_LKAkzcwhV99kpisYag=KeHf*%9NE{Nj<0RFys5Ib zpr)3Z@B%iZ-N%bY`g(t1Bn01Bkxs_rR7kr!Q12AVSf(krJQwN2g7tIlNIJ?SmsXDn z=`q@4Io?3hN?0oEB@47=q%wxeA=c+ilyKBWxjIhiV)y+>NH-S<-;+-MFv^%T(@fSl zD;ej(Fiz;k6<$*oPLD^~Lw;G&o(2~pUyi1@`p}TR&Kacnc4to2MW~hA(nl3kbf+;{ zv=eU6a627e*M>Hlve}t(yo($p*~6$)$N85or?hX#I!d<`cj~BzYJ~+zIstM{<+haO zk{n3#NcxkC*j8r&*i#~niRU6zx8eE61ozZ_Ej~=u3qAiClR-SvAZ_9klLVwmOw-3R zz-%g+sbypx90s`Z-vF;hK$b9HDMzLKUma;ptd$-XmPuB192H-Uz*NsRD4*CGkB5~F z&_D%^NnFsbj%ZFa4XjZ3uai+@(j~CLJGdD5(n)b4!%+HRh!P$qif)AA;V?={4ObsQ z{B1Q&KDkOtrHVV46rsP;JmOhevQg)25K1Q<_OmDanxb(+Ny54!Mu*_4UOUSd-(#EI p0cV>TAD|NZWf`|-bwd#++VlT`klu|F`@5=*<*HG}Ww literal 0 HcmV?d00001 diff --git a/x2paddle/optimizer/tensorflow/__pycache__/transpose.cpython-37.pyc b/x2paddle/optimizer/tensorflow/__pycache__/transpose.cpython-37.pyc new file mode 100644 index 0000000000000000000000000000000000000000..4f71250e429d685c5c0cc5464a405527ff7195f9 GIT binary patch literal 5628 zcmbstOKcm*b!LBDQY1ymvMkApBe+Qs7KW5MKmw$W@cR7WeY?Gq!c|6JrqSz1QZYyXwX9sK@LTazO=}xhaC3WUK|ufk3}x+d$S*YBpC&| z#J-)G_vXDf@4b05kFOO9IRej}FaKI3S|VTnevcSRHiZwz8ce+0biYEfQuRC z-{!XGw_CpRR$EjlOvDGj1$e&$uYVT+azPF_n`LU5fLMf?EX#6ij^$Z_71=y1u?4os zF0iFq0q~SvWXtRlTd8T#Q`jn7W0%=Fdx~AD88D(U0vbHM-M_Hc=(=pPVIMf$->kO0 z_pY=4rK!PPTU75`UW+&TPfZ~rYD*-G=wF^fbH{i4j^#Nvv_yaXEaa}+aFJj}k9Moo z06i9`rnGGaV+&Jb+-Y=Y$Z+Vzerbx#-u3%SQzO1xYqnfAL+A9`v)ud+7yaewxer|b zzP0zl^%-S^?cMKRoF;AUiFTvw&zQ8~*!2VJ4CVJ4trnkEito)R<)Z>K26Y-v)A0mg z9Lr|x30S|o`vh#W+js(&x$pH?rdG%u)~z~L)3LpoMaCGr(Ks8?w0mbGZ13P~gx_tR zExpZKtgKb?Zqu$gRus%(fw|nNiln>9E3QL>(Sg!? zE};8lL{G?f$-T9Alz}?X0&QPmN?qZl0CLGoWT13PJ0wsrn{9JUIWs|?D*YIdWi>Qn zo9V7_nto{19qu`evd(cKpo%{z zurD(-ta^#&XrA`3Os)OYl+*|~4cVe0{7*q$0cJR)ln8oE_LW2WVMeHjrx(QBlUT4MqZZb zAkUoOYOazAF4-GoNIOd@A6W4@d~*K-pLcdhYRfG6XKs`a)F68TTqzv25Qh2f{wf@U za{IizG3ENr&Gc}*06xyci(mf-uuTO7dH{9=AJJi^cQc?P6$5{mx%WJTJ_LWD9P4$1 zU+b(K>mw}yN{9?y7{~e{JQ20_}t%U7+sIF|Drh{mu%s z^9;NNhp@sj=m!}8ptB+hhh(5jXh@im(3CLCvV+`^_890`9L)zx4ZIB6(R;TCbEtPP z2mVJH3jZqYCvrdvQ>2`KvfLndn*{o(6wJwp`YffMJ2v(g>c)`rwZJ$BzbNr{(>WKI z0r5ljFUavXWA5k8UXoOQI%oFJ;@Qb+@}qgUm_SqvssGzq&OzoZ4Ozaf3>B`NBjF#X zxR_%u5w7#u4iOi@6ZxUm`;CBobV!*|*Mr}H-z5(pdWBpR^43*C9bVlMIh5tPro3y{eKac0n1^IersPcb}{dkHK zdQ#=r&g7)>-#=f**t_13#?qvBRsLlvWm|JW)o&O>oOYDN#Lg$J9$DA;4oXrWNRp;S3oF1LhDdiML zt6X0kt;WO(Cp9v7KIW5HWSoHv|GD#J$Y_;%$I%*D6LXQle|4tL2JbzAA4Y4$FNs#L zES%1{q2e#!hN$f<%bwx?5zZ6L7SU-`5ElbDjSlHR?_CRYxIH{GSd?e+ViJ|mL+cy- zi%9;^=zJl0WqhNv(gSWVt29^)$mnvgD7~8L0m@7@?{ZHG7I)$%dk%s5`t z-hMqa>yFd5nA@DG`JkGt+9IUAkRJ43`l09as!m%tY{Pd&*A{Ne+qiGH+m6RJ9^7{u z&PIIKcD>q0dRZ+$6>8NMUb7iU!4{uG*?46LRk)*vI)@u73)P0>g{pr57g=90hx1T> zz~N>a8hcRa+Ja*R9_qFb+z+*e>x(e!u$tprt*!`FxQvEIqz%_l0Io310IuB?KF4c0 za0eo_-D-zsG}HCMyz2ph<-* z5{g@p;=sUFsQD_Tbp3I$Qz=13)V6E3NU9;JQl$_XIi`}DS4vgevvwUTDP6!eu#8Ia zJr@k`u%vv-RlA&nFuc^OZd-(zUbKuba}YPE6VgW62m< zD*`N~bVcBq&Pm%v&?@3H%{Ag{${8h)qk&9(bF>)0#$;OoO_W8TSb{f^kF@{41_o$o zPz5Z8(s2bzzXW5^&!GMC)XkoEz0DhqR@H9!H#cXz5~`3SWd7hw;2OS+0Fy|l**({P z{2zGw+qsnAMpU3CdBI`VHg+A@&Cbi?{D6V*%KL+B8}^=X_-RDB5bAC2dV*s?8f~~T zw(ws)zKCEA!DR$n2)>5k>i{6Lz<&e|zY6VgPR5|z3H%leRF)!D#xaMQRI(7%@!po( z`rF8dy8JbqaxvL)<9JWPINoIZWuW8VLU04YD+q2Pz?z)Dia^%eD%km1==rw*5O7Q` z31|q(wV#q#rF+0l8U#lXf_4dL79q%&Wcz9aEjAaYOa_WIaXF9!#|x6 XO-z)0s1Wmwimkz(V~m+RD@D literal 0 HcmV?d00001 -- GitLab