[jira] [Updated] (SINGA-476) Autograd operators for ONNX
[ https://issues.apache.org/jira/browse/SINGA-476?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] zhangzhaoqi updated SINGA-476: -- Description: For the demo purpose, we need to implement these three models, and these are their components: h2. [Tiny yolov2|https://arxiv.org/pdf/1612.08242.pdf] MaxPooling2D Conv2D BatchNormalization LeakyReLU Reshape h2. [Arcface|https://arxiv.org/pdf/1801.07698.pdf] Conv2D BatchNormalization relu MaxPooling2D Dropout Flatten Dense Softmax l2_normalize acos cos h2. [BIDAF|https://arxiv.org/pdf/1611.01603.pdf] K.stack Softmax K.expand_dims K.sum Constant Dense Lambda(lambda x: 1.0 - x, output_shape=(dim,)) Multiply Add K.concatenate K.shape K.max K.tile K.squeeze linear TimeDistributed Bidirectional(LSTM In summary, we already implemented 12 ops, and there still are 16 ops needed to be implemented: h2. Already implemented: -LSTM- -Multiply- -Add- -linear- -relu- -acos- -cos- -LeakyReLU- -Softmax- -MaxPooling2D- -Conv2D- -BatchNormalization- h2. To be implemented: Reshape Flatten Dropout max shape concatenate Constant L2Normalization Expand tile squeeze Dense* TimeDistributed* Bidirectional* Stack* Lambda* *means this op doesn't have a corresponding one at ONNX op sets, therefore, it needs a converter function by using basic op sets. was: For the demo purpose, we need to implement these three models, and these are their components: h2. [Tiny yolov2|https://arxiv.org/pdf/1612.08242.pdf] MaxPooling2D Conv2D BatchNormalization LeakyReLU Reshape h2. [Arcface|https://arxiv.org/pdf/1801.07698.pdf] Conv2D BatchNormalization relu MaxPooling2D Dropout Flatten Dense Softmax l2_normalize acos cos h2. [BIDAF|https://arxiv.org/pdf/1611.01603.pdf] K.stack Softmax K.expand_dims K.sum Constant Dense Lambda(lambda x: 1.0 - x, output_shape=(dim,)) Multiply Add K.concatenate K.shape K.max K.tile K.squeeze linear TimeDistributed Bidirectional(LSTM In summary, we already implemented 12 ops, and there still are 16 ops needed to be implemented: h2. Already implemented: -LSTM- -Multiply- -Add- -linear- -relu- -acos- -cos- -LeakyReLU- -Softmax- -MaxPooling2D- -Conv2D- -BatchNormalization- h2. To be implemented: Reshape Flatten Dropout max shape concatenate Constant L2Normalization Expand tile squeeze Dense* TimeDistributed* Bidirectional* Stack* Lambda* *means this op doesn't have a corresponding one at ONNX op sets, therefore, it needs a converter function by using basic op sets. > Autograd operators for ONNX > --- > > Key: SINGA-476 > URL: https://issues.apache.org/jira/browse/SINGA-476 > Project: Singa > Issue Type: New Feature >Reporter: zhangzhaoqi >Priority: Critical > > For the demo purpose, we need to implement these three models, and these are > their components: > h2. [Tiny yolov2|https://arxiv.org/pdf/1612.08242.pdf] > MaxPooling2D > Conv2D > BatchNormalization > LeakyReLU > Reshape > h2. [Arcface|https://arxiv.org/pdf/1801.07698.pdf] > Conv2D > BatchNormalization > relu > MaxPooling2D > Dropout > Flatten > Dense > Softmax > l2_normalize > acos > cos > h2. [BIDAF|https://arxiv.org/pdf/1611.01603.pdf] > K.stack > Softmax > K.expand_dims > K.sum > Constant > Dense > Lambda(lambda x: 1.0 - x, output_shape=(dim,)) > Multiply > Add > K.concatenate > K.shape > K.max > K.tile > K.squeeze > linear > TimeDistributed > Bidirectional(LSTM > > > In summary, we already implemented 12 ops, and there still are 16 ops needed > to be implemented: > h2. Already implemented: > -LSTM- > -Multiply- > -Add- > -linear- > -relu- > -acos- > -cos- > -LeakyReLU- > -Softmax- > -MaxPooling2D- > -Conv2D- > -BatchNormalization- > h2. To be implemented: > Reshape > Flatten > Dropout > max > shape > concatenate > Constant > L2Normalization > Expand > tile > squeeze > Dense* > TimeDistributed* > Bidirectional* > Stack* > Lambda* > *means this op doesn't have a corresponding one at ONNX op sets, therefore, > it needs a converter function by using basic op sets. > -- This message was sent by Atlassian JIRA (v7.6.14#76016)
[jira] [Updated] (SINGA-476) Autograd operators for ONNX
[ https://issues.apache.org/jira/browse/SINGA-476?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] zhangzhaoqi updated SINGA-476: -- Description: For the demo purpose, we need to implement these three models, and these are their components: h2. [Tiny yolov2|https://arxiv.org/pdf/1612.08242.pdf] MaxPooling2D Conv2D BatchNormalization LeakyReLU Reshape h2. [Arcface|https://arxiv.org/pdf/1801.07698.pdf] Conv2D BatchNormalization relu MaxPooling2D Dropout Flatten Dense Softmax l2_normalize acos cos h2. [BIDAF|https://arxiv.org/pdf/1611.01603.pdf] K.stack Softmax K.expand_dims K.sum Constant Dense Lambda(lambda x: 1.0 - x, output_shape=(dim,)) Multiply Add K.concatenate K.shape K.max K.tile K.squeeze linear TimeDistributed Bidirectional(LSTM In summary, we already implemented 12 ops, and there still are 16 ops needed to be implemented: h2. Already implemented: -LSTM- -Multiply- -Add- -linear- -relu- -acos- -cos- -LeakyReLU- -Softmax- -MaxPooling2D- -Conv2D- -BatchNormalization- h2. To be implemented: Reshape Flatten Dropout max shape concatenate Constant L2Normalization Expand tile squeeze Dense* TimeDistributed* Bidirectional* Stack* Lambda* *means this op doesn't have a corresponding one at ONNX op sets, therefore, it needs a converter function by using basic op sets. was: For the demo purpose, we need to implement these three models, and these are their components: h2. [Tiny yolov2|https://arxiv.org/pdf/1612.08242.pdf] MaxPooling2D Conv2D BatchNormalization LeakyReLU Reshape h2. [Arcface]|https://arxiv.org/pdf/1801.07698.pdf] Conv2D BatchNormalization relu MaxPooling2D Dropout Flatten Dense Softmax l2_normalize acos cos h2. [BIDAF]]|https://arxiv.org/pdf/1611.01603.pdf] K.stack Softmax K.expand_dims K.sum Constant Dense Lambda(lambda x: 1.0 - x, output_shape=(dim,)) Multiply Add K.concatenate K.shape K.max K.tile K.squeeze linear TimeDistributed Bidirectional(LSTM In summary, we already implemented 12 ops, and there still are 16 ops needed to be implemented: h2. Already implemented: -LSTM- -Multiply- -Add- -linear- -relu- -acos- -cos- -LeakyReLU- -Softmax- -MaxPooling2D- -Conv2D- -BatchNormalization- h2. To be implemented: Reshape Flatten Dropout max shape concatenate Constant L2Normalization Expand tile squeeze Dense* TimeDistributed* Bidirectional* Stack* Lambda* *means this op doesn't have a corresponding one at ONNX op sets, therefore, it needs a converter function by using basic op sets. > Autograd operators for ONNX > --- > > Key: SINGA-476 > URL: https://issues.apache.org/jira/browse/SINGA-476 > Project: Singa > Issue Type: New Feature >Reporter: zhangzhaoqi >Priority: Critical > > For the demo purpose, we need to implement these three models, and these are > their components: > h2. [Tiny yolov2|https://arxiv.org/pdf/1612.08242.pdf] > MaxPooling2D > Conv2D > BatchNormalization > LeakyReLU > Reshape > h2. [Arcface|https://arxiv.org/pdf/1801.07698.pdf] > Conv2D > BatchNormalization > relu > MaxPooling2D > Dropout > Flatten > Dense > Softmax > l2_normalize > acos > cos > h2. [BIDAF|https://arxiv.org/pdf/1611.01603.pdf] > K.stack > Softmax > K.expand_dims > K.sum > Constant > Dense > Lambda(lambda x: 1.0 - x, output_shape=(dim,)) > Multiply > Add > K.concatenate > K.shape > K.max > K.tile > K.squeeze > linear > TimeDistributed > Bidirectional(LSTM > > > In summary, we already implemented 12 ops, and there still are 16 ops needed > to be implemented: > h2. Already implemented: > -LSTM- > -Multiply- > -Add- > -linear- > -relu- > -acos- > -cos- > -LeakyReLU- > -Softmax- > -MaxPooling2D- > -Conv2D- > -BatchNormalization- > h2. To be implemented: > Reshape > Flatten > Dropout > max > shape > concatenate > Constant > L2Normalization > Expand > tile > squeeze > Dense* > TimeDistributed* > Bidirectional* > Stack* > Lambda* > *means this op doesn't have a corresponding one at ONNX op sets, therefore, > it needs a converter function by using basic op sets. > -- This message was sent by Atlassian JIRA (v7.6.14#76016)
[jira] [Updated] (SINGA-476) Autograd operators for ONNX
[ https://issues.apache.org/jira/browse/SINGA-476?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] zhangzhaoqi updated SINGA-476: -- Description: For the demo purpose, we need to implement these three models, and these are their components: h2. [Tiny yolov2|https://arxiv.org/pdf/1612.08242.pdf] MaxPooling2D Conv2D BatchNormalization LeakyReLU Reshape h2. [Arcface]|https://arxiv.org/pdf/1801.07698.pdf] Conv2D BatchNormalization relu MaxPooling2D Dropout Flatten Dense Softmax l2_normalize acos cos h2. [BIDAF]]|https://arxiv.org/pdf/1611.01603.pdf] K.stack Softmax K.expand_dims K.sum Constant Dense Lambda(lambda x: 1.0 - x, output_shape=(dim,)) Multiply Add K.concatenate K.shape K.max K.tile K.squeeze linear TimeDistributed Bidirectional(LSTM In summary, we already implemented 12 ops, and there still are 16 ops needed to be implemented: h2. Already implemented: -LSTM- -Multiply- -Add- -linear- -relu- -acos- -cos- -LeakyReLU- -Softmax- -MaxPooling2D- -Conv2D- -BatchNormalization- h2. To be implemented: Reshape Flatten Dropout max shape concatenate Constant L2Normalization Expand tile squeeze Dense* TimeDistributed* Bidirectional* Stack* Lambda* *means this op doesn't have a corresponding one at ONNX op sets, therefore, it needs a converter function by using basic op sets. was: For the demo purpose, we need to implement these three models, and these are their components: h2. [[Tiny yolov2|]|https://arxiv.org/pdf/1612.08242.pdf]] MaxPooling2D Conv2D BatchNormalization LeakyReLU Reshape h2. Arcface[link title|https://arxiv.org/abs/1801.07698] Conv2D BatchNormalization relu MaxPooling2D Dropout Flatten Dense Softmax l2_normalize acos cos h2. BIDAF[link title|https://arxiv.org/pdf/1611.01603] K.stack Softmax K.expand_dims K.sum Constant Dense Lambda(lambda x: 1.0 - x, output_shape=(dim,)) Multiply Add K.concatenate K.shape K.max K.tile K.squeeze linear TimeDistributed Bidirectional(LSTM In summary, we already implemented 12 ops, and there still are 16 ops needed to be implemented: h2. Already implemented: -LSTM- -Multiply- -Add- -linear- -relu- -acos- -cos- -LeakyReLU- -Softmax- -MaxPooling2D- -Conv2D- -BatchNormalization- h2. To be implemented: Reshape Flatten Dropout max shape concatenate Constant L2Normalization Expand tile squeeze Dense* TimeDistributed* Bidirectional* Stack* Lambda* *means this op doesn't have a corresponding one at ONNX op sets, therefore, it needs a converter function by using basic op sets. > Autograd operators for ONNX > --- > > Key: SINGA-476 > URL: https://issues.apache.org/jira/browse/SINGA-476 > Project: Singa > Issue Type: New Feature >Reporter: zhangzhaoqi >Priority: Critical > > For the demo purpose, we need to implement these three models, and these are > their components: > h2. [Tiny yolov2|https://arxiv.org/pdf/1612.08242.pdf] > MaxPooling2D > Conv2D > BatchNormalization > LeakyReLU > Reshape > h2. [Arcface]|https://arxiv.org/pdf/1801.07698.pdf] > Conv2D > BatchNormalization > relu > MaxPooling2D > Dropout > Flatten > Dense > Softmax > l2_normalize > acos > cos > h2. [BIDAF]]|https://arxiv.org/pdf/1611.01603.pdf] > K.stack > Softmax > K.expand_dims > K.sum > Constant > Dense > Lambda(lambda x: 1.0 - x, output_shape=(dim,)) > Multiply > Add > K.concatenate > K.shape > K.max > K.tile > K.squeeze > linear > TimeDistributed > Bidirectional(LSTM > > > In summary, we already implemented 12 ops, and there still are 16 ops needed > to be implemented: > h2. Already implemented: > -LSTM- > -Multiply- > -Add- > -linear- > -relu- > -acos- > -cos- > -LeakyReLU- > -Softmax- > -MaxPooling2D- > -Conv2D- > -BatchNormalization- > h2. To be implemented: > Reshape > Flatten > Dropout > max > shape > concatenate > Constant > L2Normalization > Expand > tile > squeeze > Dense* > TimeDistributed* > Bidirectional* > Stack* > Lambda* > *means this op doesn't have a corresponding one at ONNX op sets, therefore, > it needs a converter function by using basic op sets. > -- This message was sent by Atlassian JIRA (v7.6.14#76016)
[jira] [Updated] (SINGA-476) Autograd operators for ONNX
[ https://issues.apache.org/jira/browse/SINGA-476?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] zhangzhaoqi updated SINGA-476: -- Description: For the demo purpose, we need to implement these three models, and these are their components: h2. [[Tiny yolov2|]|https://arxiv.org/pdf/1612.08242.pdf]] MaxPooling2D Conv2D BatchNormalization LeakyReLU Reshape h2. Arcface[link title|https://arxiv.org/abs/1801.07698] Conv2D BatchNormalization relu MaxPooling2D Dropout Flatten Dense Softmax l2_normalize acos cos h2. BIDAF[link title|https://arxiv.org/pdf/1611.01603] K.stack Softmax K.expand_dims K.sum Constant Dense Lambda(lambda x: 1.0 - x, output_shape=(dim,)) Multiply Add K.concatenate K.shape K.max K.tile K.squeeze linear TimeDistributed Bidirectional(LSTM In summary, we already implemented 12 ops, and there still are 16 ops needed to be implemented: h2. Already implemented: -LSTM- -Multiply- -Add- -linear- -relu- -acos- -cos- -LeakyReLU- -Softmax- -MaxPooling2D- -Conv2D- -BatchNormalization- h2. To be implemented: Reshape Flatten Dropout max shape concatenate Constant L2Normalization Expand tile squeeze Dense* TimeDistributed* Bidirectional* Stack* Lambda* *means this op doesn't have a corresponding one at ONNX op sets, therefore, it needs a converter function by using basic op sets. was: For the demo purpose, we need to implement these three models, and these are their components: h2. Tiny yolov2[link title|https://arxiv.org/pdf/1612.08242.pdf] MaxPooling2D Conv2D BatchNormalization LeakyReLU Reshape h2. Arcface[link title|https://arxiv.org/abs/1801.07698] Conv2D BatchNormalization relu MaxPooling2D Dropout Flatten Dense Softmax l2_normalize acos cos h2. BIDAF[link title|https://arxiv.org/pdf/1611.01603] K.stack Softmax K.expand_dims K.sum Constant Dense Lambda(lambda x: 1.0 - x, output_shape=(dim,)) Multiply Add K.concatenate K.shape K.max K.tile K.squeeze linear TimeDistributed Bidirectional(LSTM In summary, we already implemented 12 ops, and there still are 16 ops needed to be implemented: h2. Already implemented: -LSTM- -Multiply- -Add- -linear- -relu- -acos- -cos- -LeakyReLU- -Softmax- -MaxPooling2D- -Conv2D- -BatchNormalization- h2. To be implemented: Reshape Flatten Dropout max shape concatenate Constant L2Normalization Expand tile squeeze Dense* TimeDistributed* Bidirectional* Stack* Lambda* *means this op doesn't have a corresponding one at ONNX op sets, therefore, it needs a converter function by using basic op sets. > Autograd operators for ONNX > --- > > Key: SINGA-476 > URL: https://issues.apache.org/jira/browse/SINGA-476 > Project: Singa > Issue Type: New Feature >Reporter: zhangzhaoqi >Priority: Critical > > For the demo purpose, we need to implement these three models, and these are > their components: > h2. [[Tiny yolov2|]|https://arxiv.org/pdf/1612.08242.pdf]] > MaxPooling2D > Conv2D > BatchNormalization > LeakyReLU > Reshape > h2. Arcface[link title|https://arxiv.org/abs/1801.07698] > Conv2D > BatchNormalization > relu > MaxPooling2D > Dropout > Flatten > Dense > Softmax > l2_normalize > acos > cos > h2. BIDAF[link title|https://arxiv.org/pdf/1611.01603] > K.stack > Softmax > K.expand_dims > K.sum > Constant > Dense > Lambda(lambda x: 1.0 - x, output_shape=(dim,)) > Multiply > Add > K.concatenate > K.shape > K.max > K.tile > K.squeeze > linear > TimeDistributed > Bidirectional(LSTM > > > In summary, we already implemented 12 ops, and there still are 16 ops needed > to be implemented: > h2. Already implemented: > -LSTM- > -Multiply- > -Add- > -linear- > -relu- > -acos- > -cos- > -LeakyReLU- > -Softmax- > -MaxPooling2D- > -Conv2D- > -BatchNormalization- > h2. To be implemented: > Reshape > Flatten > Dropout > max > shape > concatenate > Constant > L2Normalization > Expand > tile > squeeze > Dense* > TimeDistributed* > Bidirectional* > Stack* > Lambda* > *means this op doesn't have a corresponding one at ONNX op sets, therefore, > it needs a converter function by using basic op sets. > -- This message was sent by Atlassian JIRA (v7.6.14#76016)
[jira] [Updated] (SINGA-476) Autograd operators for ONNX
[ https://issues.apache.org/jira/browse/SINGA-476?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] zhangzhaoqi updated SINGA-476: -- Description: For the demo purpose, we need to implement these three models, and these are their components: h2. Tiny yolov2[link title|https://arxiv.org/pdf/1612.08242.pdf] MaxPooling2D Conv2D BatchNormalization LeakyReLU Reshape h2. Arcface[link title|https://arxiv.org/abs/1801.07698] Conv2D BatchNormalization relu MaxPooling2D Dropout Flatten Dense Softmax l2_normalize acos cos h2. BIDAF[link title|https://arxiv.org/pdf/1611.01603] K.stack Softmax K.expand_dims K.sum Constant Dense Lambda(lambda x: 1.0 - x, output_shape=(dim,)) Multiply Add K.concatenate K.shape K.max K.tile K.squeeze linear TimeDistributed Bidirectional(LSTM In summary, we already implemented 12 ops, and there still are 16 ops needed to be implemented: h2. Already implemented: -LSTM- -Multiply- -Add- -linear- -relu- -acos- -cos- -LeakyReLU- -Softmax- -MaxPooling2D- -Conv2D- -BatchNormalization- h2. To be implemented: Reshape Flatten Dropout max shape concatenate Constant L2Normalization Expand tile squeeze Dense* TimeDistributed* Bidirectional* Stack* Lambda* *means this op doesn't have a corresponding one at ONNX op sets, therefore, it needs a converter function by using basic op sets. was: For the demo purpose, we need to implement these three models, and these are their components: h2. Tiny yolov2[link title|https://arxiv.org/pdf/1612.08242.pdf] MaxPooling2D Conv2D BatchNormalization LeakyReLU Reshape h2. Arcface[link title|https://arxiv.org/abs/1801.07698] Conv2D BatchNormalization relu MaxPooling2D Dropout Flatten Dense Softmax l2_normalize acos cos h2. BIDAF[link title|https://arxiv.org/pdf/1611.01603] K.stack Softmax K.expand_dims K.sum Constant Dense Lambda(lambda x: 1.0 - x, output_shape=(dim,)) Multiply Add K.concatenate K.shape K.max K.tile K.squeeze linear TimeDistributed Bidirectional(LSTM h2. In summary, h2. Already implemented: -LSTM- -Multiply- -Add- -linear- -relu- -acos- -cos- -LeakyReLU- -Softmax- -MaxPooling2D- -Conv2D- -BatchNormalization- h2. To be implemented: Reshape Flatten Dropout max shape concatenate Constant L2Normalization Expand tile squeeze Dense* TimeDistributed* Bidirectional* Stack* Lambda* *means this op doesn't have a corresponding one at ONNX op sets, therefore, it needs a converter function by using basic op sets. > Autograd operators for ONNX > --- > > Key: SINGA-476 > URL: https://issues.apache.org/jira/browse/SINGA-476 > Project: Singa > Issue Type: New Feature >Reporter: zhangzhaoqi >Priority: Critical > > For the demo purpose, we need to implement these three models, and these are > their components: > h2. Tiny yolov2[link title|https://arxiv.org/pdf/1612.08242.pdf] > MaxPooling2D > Conv2D > BatchNormalization > LeakyReLU > Reshape > h2. Arcface[link title|https://arxiv.org/abs/1801.07698] > Conv2D > BatchNormalization > relu > MaxPooling2D > Dropout > Flatten > Dense > Softmax > l2_normalize > acos > cos > h2. BIDAF[link title|https://arxiv.org/pdf/1611.01603] > K.stack > Softmax > K.expand_dims > K.sum > Constant > Dense > Lambda(lambda x: 1.0 - x, output_shape=(dim,)) > Multiply > Add > K.concatenate > K.shape > K.max > K.tile > K.squeeze > linear > TimeDistributed > Bidirectional(LSTM > > > In summary, we already implemented 12 ops, and there still are 16 ops needed > to be implemented: > h2. Already implemented: > -LSTM- > -Multiply- > -Add- > -linear- > -relu- > -acos- > -cos- > -LeakyReLU- > -Softmax- > -MaxPooling2D- > -Conv2D- > -BatchNormalization- > h2. To be implemented: > Reshape > Flatten > Dropout > max > shape > concatenate > Constant > L2Normalization > Expand > tile > squeeze > Dense* > TimeDistributed* > Bidirectional* > Stack* > Lambda* > *means this op doesn't have a corresponding one at ONNX op sets, therefore, > it needs a converter function by using basic op sets. > -- This message was sent by Atlassian JIRA (v7.6.14#76016)
[jira] [Updated] (SINGA-476) Autograd operators for ONNX
[ https://issues.apache.org/jira/browse/SINGA-476?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] zhangzhaoqi updated SINGA-476: -- Description: For the demo purpose, we need to implement these three models, and these are their components: h2. Tiny yolov2[link title|https://arxiv.org/pdf/1612.08242.pdf] MaxPooling2D Conv2D BatchNormalization LeakyReLU Reshape h2. Arcface[link title|https://arxiv.org/abs/1801.07698] Conv2D BatchNormalization relu MaxPooling2D Dropout Flatten Dense Softmax l2_normalize acos cos h2. BIDAF[link title|https://arxiv.org/pdf/1611.01603] K.stack Softmax K.expand_dims K.sum Constant Dense Lambda(lambda x: 1.0 - x, output_shape=(dim,)) Multiply Add K.concatenate K.shape K.max K.tile K.squeeze linear TimeDistributed Bidirectional(LSTM h2. In summary, h2. Already implemented: -LSTM- -Multiply- -Add- -linear- -relu- -acos- -cos- -LeakyReLU- -Softmax- -MaxPooling2D- -Conv2D- -BatchNormalization- h2. To be implemented: Reshape Flatten Dropout max shape concatenate Constant L2Normalization Expand tile squeeze Dense* TimeDistributed* Bidirectional* Stack* Lambda* *means this op doesn't have a corresponding one at ONNX op sets, therefore, it needs a converter function by using basic op sets. was: Already implemented: -LSTM- -Multiply- -Add- -linear- -relu- -acos- -cos- -LeakyReLU- -Softmax- -MaxPooling2D- -Conv2D- -BatchNormalization- To be implemented: Reshape Flatten Dropout max shape concatenate Constant L2Normalization Expand tile squeeze Dense* TimeDistributed* Bidirectional* Stack* Lambda* *means this op doesn't have a corresponding one at ONNX op sets, therefore, it needs a converter function by using basic op sets. > Autograd operators for ONNX > --- > > Key: SINGA-476 > URL: https://issues.apache.org/jira/browse/SINGA-476 > Project: Singa > Issue Type: New Feature >Reporter: zhangzhaoqi >Priority: Critical > > For the demo purpose, we need to implement these three models, and these are > their components: > h2. Tiny yolov2[link title|https://arxiv.org/pdf/1612.08242.pdf] > MaxPooling2D > Conv2D > BatchNormalization > LeakyReLU > Reshape > h2. Arcface[link title|https://arxiv.org/abs/1801.07698] > Conv2D > BatchNormalization > relu > MaxPooling2D > Dropout > Flatten > Dense > Softmax > l2_normalize > acos > cos > h2. BIDAF[link title|https://arxiv.org/pdf/1611.01603] > K.stack > Softmax > K.expand_dims > K.sum > Constant > Dense > Lambda(lambda x: 1.0 - x, output_shape=(dim,)) > Multiply > Add > K.concatenate > K.shape > K.max > K.tile > K.squeeze > linear > TimeDistributed > Bidirectional(LSTM > h2. In summary, > h2. Already implemented: > -LSTM- > -Multiply- > -Add- > -linear- > -relu- > -acos- > -cos- > -LeakyReLU- > -Softmax- > -MaxPooling2D- > -Conv2D- > -BatchNormalization- > h2. To be implemented: > Reshape > Flatten > Dropout > max > shape > concatenate > Constant > L2Normalization > Expand > tile > squeeze > Dense* > TimeDistributed* > Bidirectional* > Stack* > Lambda* > *means this op doesn't have a corresponding one at ONNX op sets, therefore, > it needs a converter function by using basic op sets. > -- This message was sent by Atlassian JIRA (v7.6.14#76016)
[jira] [Created] (SINGA-476) Autograd operators for ONNX
zhangzhaoqi created SINGA-476: - Summary: Autograd operators for ONNX Key: SINGA-476 URL: https://issues.apache.org/jira/browse/SINGA-476 Project: Singa Issue Type: New Feature Reporter: zhangzhaoqi Already implemented: -LSTM- -Multiply- -Add- -linear- -relu- -acos- -cos- -LeakyReLU- -Softmax- -MaxPooling2D- -Conv2D- -BatchNormalization- To be implemented: Reshape Flatten Dropout max shape concatenate Constant L2Normalization Expand tile squeeze Dense* TimeDistributed* Bidirectional* Stack* Lambda* *means this op doesn't have a corresponding one at ONNX op sets, therefore, it needs a converter function by using basic op sets. -- This message was sent by Atlassian JIRA (v7.6.14#76016)
[jira] [Updated] (SINGA-476) Autograd operators for ONNX
[ https://issues.apache.org/jira/browse/SINGA-476?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] zhangzhaoqi updated SINGA-476: -- Description: Already implemented: -LSTM- -Multiply- -Add- -linear- -relu- -acos- -cos- -LeakyReLU- -Softmax- -MaxPooling2D- -Conv2D- -BatchNormalization- To be implemented: Reshape Flatten Dropout max shape concatenate Constant L2Normalization Expand tile squeeze Dense* TimeDistributed* Bidirectional* Stack* Lambda* *means this op doesn't have a corresponding one at ONNX op sets, therefore, it needs a converter function by using basic op sets. was: Already implemented: -LSTM- -Multiply- -Add- -linear- -relu- -acos- -cos- -LeakyReLU- -Softmax- -MaxPooling2D- -Conv2D- -BatchNormalization- To be implemented: Reshape Flatten Dropout max shape concatenate Constant L2Normalization Expand tile squeeze Dense* TimeDistributed* Bidirectional* Stack* Lambda* *means this op doesn't have a corresponding one at ONNX op sets, therefore, it needs a converter function by using basic op sets. > Autograd operators for ONNX > --- > > Key: SINGA-476 > URL: https://issues.apache.org/jira/browse/SINGA-476 > Project: Singa > Issue Type: New Feature >Reporter: zhangzhaoqi >Priority: Critical > > Already implemented: > -LSTM- > -Multiply- > -Add- > -linear- > -relu- > -acos- > -cos- > -LeakyReLU- > -Softmax- > -MaxPooling2D- > -Conv2D- > -BatchNormalization- > > To be implemented: > Reshape > Flatten > Dropout > max > shape > concatenate > Constant > L2Normalization > Expand > tile > squeeze > Dense* > TimeDistributed* > Bidirectional* > Stack* > Lambda* > *means this op doesn't have a corresponding one at ONNX op sets, therefore, > it needs a converter function by using basic op sets. -- This message was sent by Atlassian JIRA (v7.6.14#76016)
[GitHub] [incubator-singa] chrishkchris commented on a change in pull request #493: SINGA-473 Autograd Trigonometry: Backward Test
chrishkchris commented on a change in pull request #493: SINGA-473 Autograd Trigonometry: Backward Test URL: https://github.com/apache/incubator-singa/pull/493#discussion_r309006617 ## File path: test/python/test_operation.py ## @@ -65,6 +65,17 @@ def prepare_inputs_targets_for_rnn_test(): targets = [t0, t1, t2] return inputs, targets, h0 +def numpy_unary_ops_backward(func, x, dy, h=0.0005): Review comment: OK, I will change the code using one of your two methods (the link provided or compute the gradient explicitly). The code I did was only the diagonal of the gradient matrix, which is only suitable for unary operator where one output is determined by one input, without the coupling from other inputs. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-singa] pinpom opened a new pull request #495: SINGA-475 add SoftSign operator
pinpom opened a new pull request #495: SINGA-475 add SoftSign operator URL: https://github.com/apache/incubator-singa/pull/495 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-singa] nudles commented on a change in pull request #492: Make singa use multiple memory pools
nudles commented on a change in pull request #492: Make singa use multiple
memory pools
URL: https://github.com/apache/incubator-singa/pull/492#discussion_r308742630
##
File path: include/singa/core/device.h
##
@@ -295,7 +295,9 @@ class Platform {
/// Create a set of CudaGPU Device using given GPU IDs.
static const std::vector>
CreateCudaGPUsOn(const std::vector , size_t init_size = 0);
-
+
+static std::vector > allRet;
Review comment:
what is allRet and retUsed?
pls add some comments.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-singa] nudles commented on a change in pull request #492: Make singa use multiple memory pools
nudles commented on a change in pull request #492: Make singa use multiple
memory pools
URL: https://github.com/apache/incubator-singa/pull/492#discussion_r308745750
##
File path: include/singa/core/device.h
##
@@ -295,7 +295,9 @@ class Platform {
/// Create a set of CudaGPU Device using given GPU IDs.
static const std::vector>
CreateCudaGPUsOn(const std::vector , size_t init_size = 0);
-
+
+static std::vector > allRet;
Review comment:
should they be static members of Platform?
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-singa] nudles commented on a change in pull request #494: SINGA-475 add SoftPlus operator
nudles commented on a change in pull request #494: SINGA-475 add SoftPlus operator URL: https://github.com/apache/incubator-singa/pull/494#discussion_r308742158 ## File path: test/python/test_operation.py ## @@ -610,6 +610,17 @@ def test_Atanh_gpu(self): np.testing.assert_array_almost_equal(tensor.to_numpy(result), XT, decimal=5) self.check_shape(dx.shape(), (3, 2)) +def test_SoftPlus(self): +X=np.array([1.0,2.0,3.0,4.0,5.0,6.0]).reshape(3,2).astype(np.float32) +XT=np.log(np.exp(X) + 1) +x=tensor.from_numpy(X) +x.to_device(gpu_dev) + +result=autograd.softplus(x) +dx=result.creator.backward(x.data) Review comment: pls also test the correctness of the gradients. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-singa] nudles commented on a change in pull request #493: SINGA-473 Autograd Trigonometry: Backward Test
nudles commented on a change in pull request #493: SINGA-473 Autograd Trigonometry: Backward Test URL: https://github.com/apache/incubator-singa/pull/493#discussion_r308740781 ## File path: test/python/test_operation.py ## @@ -65,6 +65,17 @@ def prepare_inputs_targets_for_rnn_test(): targets = [t0, t1, t2] return inputs, targets, h0 +def numpy_unary_ops_backward(func, x, dy, h=0.0005): Review comment: This is not correct. You can either use this [one](http://cs231n.github.io/optimization-1/#gradcompute), or compute the gradient explicitly, e.g., gradient of cos() is sin() This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-singa] pinpom opened a new pull request #494: SINGA-475 add SoftPlus operator
pinpom opened a new pull request #494: SINGA-475 add SoftPlus operator URL: https://github.com/apache/incubator-singa/pull/494 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
