Make seed type to be Int32 so that it is compatible with TPUs.

Sources/TensorFlow/Initializers.swift

@@ -376,12 +376,12 @@ public extension Tensor where Scalar: BinaryFloatingPoint {
     ///
     init(
         randomUniform shape: TensorShape,
-        seed: (Int64, Int64) = (Int64.random(in: Int64.min..<Int64.max),
-                                Int64.random(in: Int64.min..<Int64.max))
+        seed: (Int32, Int32) = (Int32.random(in: Int32.min..<Int32.max),
+                                Int32.random(in: Int32.min..<Int32.max))
     ) {
         self = Raw.statelessRandomUniform(
           shape: Tensor<Int32>((0..<shape.rank).map { Int32(shape[$0]) }),
-          seed: Tensor<Int64>([seed.0, seed.1])
+          seed: Tensor<Int32>([seed.0, seed.1])
         )
     }
 
@@ -394,12 +394,12 @@ public extension Tensor where Scalar: BinaryFloatingPoint {
     ///
     init(
         randomNormal shape: TensorShape,
-        seed: (Int64, Int64) = (Int64.random(in: Int64.min..<Int64.max),
-                                Int64.random(in: Int64.min..<Int64.max))
+        seed: (Int32, Int32) = (Int32.random(in: Int32.min..<Int32.max),
+                                Int32.random(in: Int32.min..<Int32.max))
     ) {
         self = Raw.statelessRandomNormal(
             shape: Tensor<Int32>((0..<shape.rank).map { Int32(shape[$0]) }),
-            seed: Tensor<Int64>([seed.0, seed.1])
+            seed: Tensor<Int32>([seed.0, seed.1])
         )
     }
 }
@@ -475,8 +475,8 @@ public extension Tensor where Scalar: TensorFlowFloatingPoint {
     ///
     init(
         glorotUniform shape: TensorShape,
-        seed: (Int64, Int64) = (Int64.random(in: Int64.min..<Int64.max),
-                                Int64.random(in: Int64.min..<Int64.max))
+        seed: (Int32, Int32) = (Int32.random(in: Int32.min..<Int32.max),
+                                Int32.random(in: Int32.min..<Int32.max))
     ) {
         let uniform = Tensor(randomUniform: shape, seed: seed)
         self = Tensor.glorot(fromStandardUniform: uniform, shape: shape)

Sources/TensorFlow/Layers/Convolutional.swift

@@ -116,8 +116,8 @@ public extension Conv1D {
         stride: Int = 1,
         padding: Padding = .valid,
         activation: @escaping Activation = identity,
-        seed: (Int64, Int64) = (Int64.random(in: Int64.min..<Int64.max),
-                                Int64.random(in: Int64.min..<Int64.max))
+        seed: (Int32, Int32) = (Int32.random(in: Int32.min..<Int32.max),
+                                Int32.random(in: Int32.min..<Int32.max))
     ) {
         let filterTensorShape = TensorShape([
             filterShape.0, filterShape.1, filterShape.2])
@@ -232,8 +232,8 @@ public extension Conv2D {
         strides: (Int, Int) = (1, 1),
         padding: Padding = .valid,
         activation: @escaping Activation = identity,
-        seed: (Int64, Int64) = (Int64.random(in: Int64.min..<Int64.max),
-                                Int64.random(in: Int64.min..<Int64.max))
+        seed: (Int32, Int32) = (Int32.random(in: Int32.min..<Int32.max),
+                                Int32.random(in: Int32.min..<Int32.max))
     ) {
         let filterTensorShape = TensorShape([
             filterShape.0, filterShape.1, filterShape.2, filterShape.3])
@@ -348,8 +348,8 @@ public extension Conv3D {
         strides: (Int, Int, Int) = (1, 1, 1),
         padding: Padding = .valid,
         activation: @escaping Activation = identity,
-        seed: (Int64, Int64) = (Int64.random(in: Int64.min..<Int64.max),
-                                Int64.random(in: Int64.min..<Int64.max))
+        seed: (Int32, Int32) = (Int32.random(in: Int32.min..<Int32.max),
+                                Int32.random(in: Int32.min..<Int32.max))
     ) {
         let filterTensorShape = TensorShape([
             filterShape.0, filterShape.1, filterShape.2, filterShape.3, filterShape.4])
@@ -473,8 +473,8 @@ public extension TransposedConv2D {
         strides: (Int, Int) = (1, 1),
         padding: Padding = .valid,
         activation: @escaping Activation = identity,
-        seed: (Int64, Int64) = (Int64.random(in: Int64.min..<Int64.max),
-                                Int64.random(in: Int64.min..<Int64.max))
+        seed: (Int32, Int32) = (Int32.random(in: Int32.min..<Int32.max),
+                                Int32.random(in: Int32.min..<Int32.max))
     ) {
         let filterTensorShape = TensorShape([
             filterShape.0, filterShape.1, filterShape.2, filterShape.3])

Sources/TensorFlow/Layers/Core.swift

@@ -214,8 +214,8 @@ public extension Dense {
         inputSize: Int,
         outputSize: Int,
         activation: @escaping Activation = identity,
-        seed: (Int64, Int64) = (Int64.random(in: Int64.min..<Int64.max),
-                                Int64.random(in: Int64.min..<Int64.max))
+        seed: (Int32, Int32) = (Int32.random(in: Int32.min..<Int32.max),
+                                Int32.random(in: Int32.min..<Int32.max))
     ) {
         self.init(weight: Tensor(glorotUniform: [inputSize, outputSize],
                                  seed: seed),

Sources/TensorFlow/Layers/Recurrent.swift

@@ -101,8 +101,8 @@ public struct SimpleRNNCell<Scalar: TensorFlowFloatingPoint>: RNNCell, VectorNum
     ///   - hiddenSize: The number of features in 2-D hidden states.
     ///   - seed: The random seed for initialization. The default value is random.
     public init(inputSize: Int, hiddenSize: Int,
-                seed: (Int64, Int64) = (Int64.random(in: Int64.min..<Int64.max),
-                                        Int64.random(in: Int64.min..<Int64.max))) {
+                seed: (Int32, Int32) = (Int32.random(in: Int32.min..<Int32.max),
+                                        Int32.random(in: Int32.min..<Int32.max))) {
         let concatenatedInputSize = inputSize + hiddenSize
         self.weight = Tensor(glorotUniform: [concatenatedInputSize, hiddenSize], seed: seed)
         self.bias = Tensor(zeros: [hiddenSize])
@@ -144,8 +144,8 @@ public struct LSTMCell<Scalar: TensorFlowFloatingPoint>: RNNCell, VectorNumeric
     ///   - inputSize: The number of features in 2-D input tensors.
     ///   - hiddenSize: The number of features in 2-D hidden states.
     public init(inputSize: Int, hiddenSize: Int,
-                seed: (Int64, Int64) = (Int64.random(in: Int64.min..<Int64.max),
-                                        Int64.random(in: Int64.min..<Int64.max))) {
+                seed: (Int32, Int32) = (Int32.random(in: Int32.min..<Int32.max),
+                                        Int32.random(in: Int32.min..<Int32.max))) {
         let concatenatedInputSize = inputSize + hiddenSize
         let gateWeightShape = TensorShape([concatenatedInputSize, hiddenSize])
         let gateBiasShape = TensorShape([hiddenSize])

Tests/TensorFlowTests/LayerTests.swift

@@ -214,15 +214,15 @@ final class LayerTests: XCTestCase {
         let x = Tensor<Float>(rangeFrom: 0.0, to: 0.4, stride: 0.1).rankLifted()
         let inputs: [Tensor<Float>] = Array(repeating: x, count: 4)
         let rnn = RNN(SimpleRNNCell<Float>(inputSize: 4, hiddenSize: 4,
-                                           seed: (0xFeedBeef, 0xDeadBeef)))
+                                           seed: (0xFeed, 0xBeef)))
         let (outputs, _) = rnn.valueWithPullback(at: inputs) { rnn, inputs in
             return rnn(inputs)
         }
         XCTAssertEqual(outputs.map { $0.value },
-                       [[[ -0.00262943,  -0.005866742, 0.044919778,  0.20036437]],
-                        [[ 0.066890605,   0.049586136, 0.024610005,  0.09341654]],
-                        [[ 0.065792546,   0.009325638, 0.06439907,  0.114802904]],
-                        [[ 0.055909205, 0.00035158166, 0.054020774,  0.09812111]]])
+                       [[[ 0.20775771,  0.20080023, -0.13768704, -0.18534681]],
+                        [[ 0.22666009,  0.30019346, -0.19720285, -0.14683801]],
+                        [[ 0.23758979,  0.32101023, -0.20359215,  -0.1787096]],
+                        [[ 0.24337786,   0.3389194, -0.21143384,  -0.1675081]]])
         // TODO: Figure out why the following is numerically unstable.
         // let (𝛁rnn, _) = pullback(.init(inputs.map { SimpleRNNCell<Float>.State($0) }))
         // XCTAssertEqual(𝛁rnn.cell.weight,

Tests/TensorFlowTests/SequentialTests.swift

@@ -21,7 +21,7 @@ final class SequentialTests: XCTestCase {
             var dense1 = Dense<Float>(inputSize: 2, outputSize: 4, activation: relu,
                                       seed: (0xfffffff, 0xfeeff))
             var dense2 = Dense<Float>(inputSize: 4, outputSize: 1, activation: relu,
-                                      seed: (0xfeffeffe, 0xfffe))
+                                      seed: (0xeffeffe, 0xfffe))
 
             @differentiable
             func call(_ input: Tensor<Float>) -> Tensor<Float> {
@@ -41,7 +41,7 @@ final class SequentialTests: XCTestCase {
             optimizer.update(&model.allDifferentiableVariables, along: 𝛁model)
         }
         XCTAssertEqual(model.inferring(from: [[0, 0], [0, 1], [1, 0], [1, 1]]),
-                       [[  0.491493], [ 0.5063815], [0.49968663], [0.50133944]])
+                       [[ 0.4904838], [0.49942452], [0.49740878], [ 0.5106092]])
     }
 
     static var allTests = [

Tests/TensorFlowTests/TrivialModelTests.swift

@@ -30,7 +30,7 @@ final class TrivialModelTests: XCTestCase {
                     inputSize: hiddenSize,
                     outputSize: 1,
                     activation: relu,
-                    seed: (0xfeffeffe, 0xfffe)
+                    seed: (0xffeffe, 0xfffe)
                 )
             }
             @differentiable