Skip to content
This repository was archived by the owner on Apr 23, 2025. It is now read-only.

transformer: upstream api changes #171

Merged
merged 1 commit into from
Jun 19, 2019
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
10 changes: 5 additions & 5 deletions Transformer/Model.swift
Original file line number Diff line number Diff line change
Expand Up @@ -51,9 +51,9 @@ struct FeedForward: Layer {
}

struct AttentionInput: Differentiable {
let query: Tensor<Float>
let key: Tensor<Float>
let value: Tensor<Float>
var query: Tensor<Float>
var key: Tensor<Float>
var value: Tensor<Float>
}

@differentiable(wrt: (query, key, value), vjp: _vjpMakeAttentionInput)
Expand All @@ -69,8 +69,8 @@ func _vjpMakeAttentionInput(query: Tensor<Float>, key: Tensor<Float>, value: Ten
}

struct AttentionContext: Differentiable {
let key: Tensor<Float>
let value: Tensor<Float>
var key: Tensor<Float>
var value: Tensor<Float>
}

@differentiable(wrt: (key, value), vjp: _vjpMakeAttentionContext)
Expand Down
4 changes: 2 additions & 2 deletions Transformer/Operators.swift
Original file line number Diff line number Diff line change
Expand Up @@ -29,7 +29,7 @@ func gelu<Scalar: TensorFlowFloatingPoint>(_ x: Tensor<Scalar>) -> Tensor<Scalar
@differentiable(
wrt: (left, right),
vjp: _vjpBatchedMatmul
where Scalar : Differentiable & FloatingPoint
where Scalar : Differentiable & TensorFlowFloatingPoint
)
func batchedMatmul<Scalar : Numeric>(
_ left: Tensor<Scalar>,
Expand All @@ -41,7 +41,7 @@ func batchedMatmul<Scalar : Numeric>(
}

@usableFromInline
func _vjpBatchedMatmul<Scalar : Differentiable & FloatingPoint>(
func _vjpBatchedMatmul<Scalar : Differentiable & TensorFlowFloatingPoint>(
_ left: Tensor<Scalar>,
_ right: Tensor<Scalar>,
adjointLeft: Bool,
Expand Down