package paddle.framework.proto

Get desktop application:
View/edit binary Protocol Buffers messages

Used in: OpDesc.Attr, OpProto.Attr, VarDesc.Attr

• INT = 0

• FLOAT = 1

• STRING = 2

• INTS = 3

• FLOATS = 4

• STRINGS = 5

• BOOLEAN = 6

• BOOLEANS = 7

• BLOCK = 8

• LONG = 9

• BLOCKS = 10

• LONGS = 11

• FLOAT64S = 12

• VAR = 13

• VARS = 14

• FLOAT64 = 15

Used in: ProgramDesc

• required int32 idx = 1

• required int32 parent_idx = 2

• repeated VarDesc vars = 3

• repeated OpDesc ops = 4

• optional int32 forward_block_idx = 5

OpDesc describes an instance of a C++ framework::OperatorBase derived class type.

Used in: BlockDesc

• required string type = 3

• repeated OpDesc.Var inputs = 1

• repeated OpDesc.Var outputs = 2

• repeated OpDesc.Attr attrs = 4

• optional bool is_target = 5

Used in: OpDesc

• required string name = 1

• required AttrType type = 2

• optional int32 i = 3

• optional float f = 4

• optional string s = 5

• repeated int32 ints = 6

• repeated float floats = 7

• repeated string strings = 8

• optional bool b = 10

• repeated bool bools = 11

• optional int32 block_idx = 12

• optional int64 l = 13

• repeated int32 blocks_idx = 14

• repeated int64 longs = 15

• repeated double float64s = 16

• optional string var_name = 17

• repeated string vars_name = 18

• optional double float64 = 19

Used in: OpDesc

• required string parameter = 1

• repeated string arguments = 2

OpProto describes a C++ framework::OperatorBase derived class.

• required string type = 1

• repeated OpProto.Var inputs = 2

• repeated OpProto.Var outputs = 3

• repeated OpProto.Attr attrs = 4

• required string comment = 5

AttrProto describes the C++ type Attribute.

Used in: OpProto

• required string name = 1

• required AttrType type = 2

• required string comment = 3

• optional bool generated = 4

  If that attribute is generated, it means the Paddle third language binding has responsibility to fill that attribute. End-User should not set that attribute.

• optional bool extra = 5

• optional bool quant = 6

• optional bool support_tensor = 7

VarProto describes the C++ type framework::Variable.

Used in: OpProto

• required string name = 1

• required string comment = 2

• optional bool duplicable = 3

• optional bool intermediate = 4

• optional bool dispensable = 5

• optional bool extra = 6

• optional bool quant = 7

In some cases, Paddle may perform operator definition iterations, and the operator uses OpVersionMap for compatibility testing.

Used in: OpVersionMap.OpVersionPair

• required int32 version = 1

Used in: ProgramDesc

• repeated OpVersionMap.OpVersionPair pair = 1

Used in: OpVersionMap

• required string op_name = 1

• required OpVersion op_version = 2

Please refer to https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/program.md for more details. TODO(panyx0718): A model can have multiple programs. Need a way to distinguish them. Maybe ID or name?

• repeated BlockDesc blocks = 1

• optional Version version = 4

• optional OpVersionMap op_version_map = 5

Used in: BlockDesc

• required string name = 1

• required VarType type = 2

• optional bool persistable = 3

• optional bool need_check_feed = 4

  True if the variable is an input data and have to check the feed data shape and dtype

• optional bool is_parameter = 5

• optional bool stop_gradient = 6

• repeated VarDesc.Attr attrs = 7

Used in: VarDesc

• required string name = 1

• required AttrType type = 2

• optional int32 i = 3

• optional string s = 4

• repeated int32 ints = 5

Used in: VarDesc

• required VarType.Type type = 1

• optional VarType.TensorDesc selected_rows = 2

• optional VarType.DenseTensorDesc dense_tensor = 3

• optional VarType.DenseTensorArrayDesc tensor_array = 4

• optional VarType.ReaderDesc reader = 5

• optional VarType.Tuple tuple = 7

• optional VarType.TensorDesc string = 8

• optional VarType.TensorDesc strings = 9

• optional VarType.TensorDesc vocab = 10

Used in: VarType

• required TensorDesc tensor = 1

• optional int32 lod_level = 2

Used in: VarType, ReaderDesc

• required TensorDesc tensor = 1

• optional int32 lod_level = 2

Used in: VarType

• repeated DenseTensorDesc dense_tensor = 1

Used in: VarType, DenseTensorArrayDesc, DenseTensorDesc

• required Type data_type = 1

  Should only be PODType. Is enforced in C++

• repeated int64 dims = 2

  [UNK, 640, 480] is saved as [-1, 640, 480]

Used in: VarType

• repeated Type element_type = 1

Used in: VarType, TensorDesc, Tuple

• BOOL = 0

  Pod Types

• INT16 = 1

• INT32 = 2

• INT64 = 3

• FP16 = 4

• FP32 = 5

• FP64 = 6

• SIZE_T = 19

  Tensor<size_t> is used in C++.

• UINT8 = 20

• INT8 = 21

• BF16 = 22

• COMPLEX64 = 23

• COMPLEX128 = 24

• DENSE_TENSOR = 7

  Other types that may need additional descriptions

• SELECTED_ROWS = 8

• FEED_MINIBATCH = 9

• FETCH_LIST = 10

• STEP_SCOPES = 11

• LOD_RANK_TABLE = 12

• DENSE_TENSOR_ARRAY = 13

• PLACE_LIST = 14

• READER = 15

• RAW = 17

  Any runtime decided variable type is raw raw variables should manage their own allocations in operators like nccl_op

• TUPLE = 18

• STRING = 25

• STRINGS = 26

• VOCAB = 27

• FEED_LIST = 28

• PSTRING = 29

  The data type of phi::StringTensor

Any incompatible changes to ProgramDesc and its dependencies should raise the version defined version.h. Serailization and Deserialization codes should be modified in a way that supports old versions following the version and compatibility policy.

Used in: ProgramDesc

• optional int64 version = 1