lite/schema/schema_v3.fbs

// Copyright 2017 The TensorFlow 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.

// Revision History
// Version 0: Initial version.
// Version 1: Add subgraphs to schema.
// Version 2: Rename operators to conform to NN API.
// Version 3: Move buffer data from Model.Subgraph.Tensors to Model.Buffers.

namespace tflite;

// This corresponds to the version (4).
file_identifier "TFL3";
// File extension of any written files.
file_extension "tflite";

// The type of data stored in a tensor.
enum TensorType : byte {
  FLOAT32 = 0,
  FLOAT16 = 1,
  INT32 = 2,
  UINT8 = 3,
  INT64 = 4,
  STRING = 5,
}

// Parameters for converting a quantized tensor back to float. Given a
// quantized value q, the corresponding float value f should be:
//   f = scale * (q - zero_point)
table QuantizationParameters {
  min:[float];  // For importing back into tensorflow.
  max:[float];  // For importing back into tensorflow.
  scale:[float];
  zero_point:[long];
}

table Tensor {
  // The tensor shape. The meaning of each entry is operator-specific but
  // builtin ops use: [batch size, number of channels, height, width] (That's
  // Tensorflow's NCHW).
  shape:[int];
  type:TensorType;
  // An index that refers to the buffers table at the root of the model. Or,
  // if there is no data buffer associated (i.e. intermediate results), then
  // this is 0 (which refers to an always existant empty buffer).
  //
  // The data_buffer itself is an opaque container, with the assumption that the
  // target device is little-endian. In addition, all builtin operators assume
  // the memory is ordered such that if `shape` is [4, 3, 2], then index
  // [i, j, k] maps to data_buffer[i*3*2 + j*3 + k].
  buffer:uint;
  name:string;  // For debugging and importing back into tensorflow.
  quantization:QuantizationParameters;  // Optional.
}

// A list of builtin operators. Builtin operators a slighlty faster than custom
// ones, but not by much. Moreover, while custom operators accept an opaque
// object containing configuration parameters, builtins have a predetermined
// set of acceptable options.
enum BuiltinOperator : byte {
  ADD = 0,
  AVERAGE_POOL_2D = 1,
  CONCATENATION = 2,
  CONV_2D = 3,
  DEPTHWISE_CONV_2D = 4,
  // DEPTH_TO_SPACE = 5,
  // DEQUANTIZE = 6,
  EMBEDDING_LOOKUP = 7,
  // FLOOR = 8,
  FULLY_CONNECTED = 9,
  HASHTABLE_LOOKUP = 10,
  L2_NORMALIZATION = 11,
  L2_POOL_2D = 12,
  LOCAL_RESPONSE_NORMALIZATION = 13,
  LOGISTIC = 14,
  LSH_PROJECTION = 15,
  LSTM = 16,
  MAX_POOL_2D = 17,
  // MUL = 18,
  RELU = 19,
  // RELU1=20,
  RELU6 = 21,
  RESHAPE = 22,
  RESIZE_BILINEAR = 23,
  RNN = 24,
  SOFTMAX = 25,
  SPACE_TO_DEPTH = 26,
  SVDF = 27,
  TANH = 28,
  // TODO(aselle): Consider rename to CONCATENATE_EMBEDDINGS
  CONCAT_EMBEDDINGS = 29,
  SKIP_GRAM = 30,
  CALL = 31,
  CUSTOM = 32,

}

// Options for the builtin operators.
union BuiltinOptions {
  Conv2DOptions,
  DepthwiseConv2DOptions,
  ConcatEmbeddingsOptions,
  LSHProjectionOptions,
  Pool2DOptions,
  SVDFOptions,
  RNNOptions,
  FullyConnectedOptions,
  SoftmaxOptions,
  ConcatenationOptions,
  AddOptions,
  L2NormOptions,
  LocalResponseNormalizationOptions,
  LSTMOptions,
  ResizeBilinearOptions,
  CallOptions,
  ReshapeOptions,
  SkipGramOptions,
  SpaceToDepthOptions,
}

enum Padding : byte { SAME, VALID }

enum ActivationFunctionType : byte {
  NONE = 0,
  RELU = 1,
  RELU1 = 2,
  RELU6 = 3,
  TANH = 4,
  SIGN_BIT = 5,
}

table Conv2DOptions {
  padding:Padding;
  stride_w:int;
  stride_h:int;
  fused_activation_function:ActivationFunctionType;
}

table Pool2DOptions {
  padding:Padding;
  stride_w:int;
  stride_h:int;
  filter_width:int;
  filter_height:int;
  fused_activation_function:ActivationFunctionType;
}

table DepthwiseConv2DOptions {
  padding:Padding;
  stride_w:int;
  stride_h:int;
  depth_multiplier:int;
  fused_activation_function:ActivationFunctionType;
}

table ConcatEmbeddingsOptions {
  num_channels:int;
  num_columns_per_channel:[int];
  embedding_dim_per_channel:[int]; // This could be inferred from parameters.
}

enum LSHProjectionType: byte {
  UNKNOWN = 0,
  SPARSE = 1,
  DENSE = 2,
}

table LSHProjectionOptions {
  type: LSHProjectionType;
}

table SVDFOptions {
  rank:int;
  fused_activation_function:ActivationFunctionType;
}

// An implementation of TensorFlow RNNCell.
table RNNOptions {
  fused_activation_function:ActivationFunctionType;
}

// An implementation of TensorFlow fully_connected (a.k.a Dense) layer.
table FullyConnectedOptions {
  fused_activation_function:ActivationFunctionType;
}

table SoftmaxOptions {
  beta: float;
}

// An implementation of TensorFlow concat.
table ConcatenationOptions {
  axis:int;
  fused_activation_function:ActivationFunctionType;
}

table AddOptions {
  fused_activation_function:ActivationFunctionType;
}

table L2NormOptions {
  fused_activation_function:ActivationFunctionType;
}

table LocalResponseNormalizationOptions {
  radius:int;
  bias:float;
  alpha:float;
  beta:float;
}

// An implementation of TensorFlow LSTMCell and CoupledInputForgetGateLSTMCell
table LSTMOptions {
  fused_activation_function:ActivationFunctionType;
  cell_clip: float; // Optional, 0.0 means no clipping
  proj_clip: float; // Optional, 0.0 means no clipping
}

table ResizeBilinearOptions {
  new_height:int;
  new_width:int;
}

// A call operation options
table CallOptions {
  // The subgraph index that needs to be called.
  subgraph:uint;
}

table ReshapeOptions {
  new_shape:[int];
}

table SkipGramOptions {
  ngram_size: int;
  max_skip_size: int;
  include_all_ngrams: bool;
}

table SpaceToDepthOptions {
  block_size: int;
}

// An OperatorCode can be an enum value (BuiltinOperator) if the operator is a
// builtin, or a string if the operator is custom.
table OperatorCode {
  builtin_code:BuiltinOperator;
  custom_code:string;
}

// An operator takes tensors as inputs and outputs. The type of operation being
// performed is determined by an index into the list of valid OperatorCodes,
// while the specifics of each operations is configured using builtin_options
// or custom_options.
table Operator {
  // Index into the operator_codes array. Using an integer here avoids
  // complicate map lookups.
  opcode_index:uint;

  inputs:[int];
  outputs:[int];

  builtin_options:BuiltinOptions;
  custom_options:[ubyte];
}

// The root type, defining a model.
table SubGraph {
  // A list of all tensors used in this model.
  tensors:[Tensor];

  // Indices of the input tensors.
  inputs:[int];

  // Indices of the output tensors.
  outputs:[int];

  // All operators, in execution order.
  operators:[Operator];

  // Name of subgraph (used for debugging).
  name:string;
}

// Table of raw data buffers (used for constant tensors). Referenced by tensors
// by index.
table Buffer {
  data:[ubyte];
}

table Model {
  // Version of the schema.
  version:uint;

  // A list of all operator codes used in this model. This is
  // kept in order because operators carry an index into this
  // vector.
  operator_codes:[OperatorCode];

  // All the subgraphs of the model. The 0th is assumed to be the main
  // model.
  subgraphs:[SubGraph];

  // A description of the model.
  description:string;

  // Buffers of the model.
  // NOTE: It is required that the first entry in here is always an empty
  // buffer. This is so that the default buffer index of zero in Tensor
  // will always refer to a valid empty buffer.
  buffers:[Buffer];

}

root_type Model;