rxwei

Swift Charts Framework Internship at Apple

Note: Applications are now closed.

Apple's Swift Charts team is now looking for interns for 2024!

This is a paid, in-person internship in Cupertino, California. While most internships last 3 months, starting in May or June, the start date and internship length are flexible. Internships are restricted to students pursuing a Bachelor's, Master's, or PhD degree. Students must be enrolled in school in

rxwei

// To be shared on Swift Forums.
// Compile with:
//     swiftc -Xllvm -enable-experimental-cross-file-derivative-registration -enable-experimental-forward-mode-differentiation main.swift

// MARK: - Make integers differentiable

extension Int: Differentiable {
    public typealias TangentVector = Int
}

rxwei

import TensorFlow

protocol Differentiable {
    ...
    var zeroTangentVectorInitializer: () -> Self { get }
}

extension Tensor where Scalar: TensorFlowFloatingPoint {
    var zeroTangentVectorInitializer: () -> Self {
        { [shape = self.shape] in Tensor(zeros: shape) }

rxwei

Upstreaming Swift AutoDiff

Author: Richard Wei ([email protected]) on behalf of the Swift for TensorFlow team

Last updated: October 2, 2019

Overview

The differentiable programming feature (AutoDiff) has been incubated in the 'tensorflow' branch of apple/swift since December 2017 and released as part of the Swift for TensorFlow toolchains. The Differentiable Programming Mega-Proposal, which serves as a manifesto, received general positive feedback from the community, but there is a long way between receiving conceptual approval and obtaining Swift Evolution approval of such a large feature. We would like to merge the pieces into the 'master' branch under a gate to further development and bake the feature on master, just like Apple develops its major features

rxwei

@differentiable
func nested_loop(_ x: Float, count: Int) -> Float {
  var outer = x
  outerLoop: for _ in 1..<count {
    outer = outer * x

    var inner = outer
    var i = 1
    while i < count {
      inner = inner + x

rxwei

// AD__dot__pullback_src_0_wrt_0_1
sil hidden [ossa] @AD__dot__pullback_src_0_wrt_0_1 : $@convention(thin) (@in_guaranteed Float, @owned _AD__dot_bb0__PB__src_0_wrt_0_1) -> (@out Vector, @out Vector) {
// %0                                             // user: %202
// %1                                             // user: %203
// %2                                             // user: %65
// %3                                             // user: %66
bb0(%0 : $*Vector, %1 : $*Vector, %2 : $*Float, %3 : @owned $_AD__dot_bb0__PB__src_0_wrt_0_1):
  %4 = alloc_stack $Tracked<Float>                // users: %233, %232, %196, %188, %7
  %5 = witness_method $Tracked<Float>, #AdditiveArithmetic.zero!getter.1 : <Self where Self : AdditiveArithmetic> (Self.Type) -> () -> Self : $@convention(witness_method: AdditiveArithmetic) <τ_0_0 where τ_0_0 : AdditiveArithmetic> (@thick τ_0_0.Type) -> @out τ_0_0 // user: %7
  %6 = metatype $@thick Tracked<Float>.Type       // user: %7

rxwei

%0 = differentiable_function (%f, %f_jvp) : $sil_differentiable(1) {(T) -> T, differential: (T) -> T, pullback: (T) -> T}
%0 = differentiable_function (%f, %f_transpose) : $sil_differentiable(linear) {(T) -> T, transpose: (T) -> T}

rxwei

import TensorFlow
import Dispatch

func foo(x: Tensor<Float>, y: Tensor<Float>) -> Tensor<Float> {
  return (x + x + x * y * y * y).sum()
}

func time(_ body: () -> Void) {
  let divisor: Float = 1_000_000_000
  let start = Float(DispatchTime.now().uptimeNanoseconds) / divisor

rxwei

- (Anthony) Switch swift-bindings to eager and `TFE_Execute`.
  - (Anthony) Remove all `#tfop`s from stdlib/TensorFlow.
- (Marc) Remove local constexpr.
- (Parker) Remove IRGen logic from `graph_op`.
- (Parker) Remove GPE files.
- (Parker) Remove `#tfop`.
- (Parker) Remove TensorFlow Bazel build rules from build-script.
- (Parker) Set up GPU CI in tensorflow/swift-apis.

rxwei