https://github.com/rxwei/swift-tensorflow-profiling

Swift for TensorFlow profiling suite
https://github.com/rxwei/swift-tensorflow-profiling

Last synced: about 2 hours ago
JSON representation

Swift for TensorFlow profiling suite

• Host: GitHub
• URL: https://github.com/rxwei/swift-tensorflow-profiling
• Owner: rxwei
• Created: 2018-07-30T04:51:06.000Z (over 6 years ago)
• Default Branch: master
• Last Pushed: 2018-08-01T12:20:21.000Z (over 6 years ago)
• Last Synced: 2024-11-11T22:34:19.645Z (about 2 months ago)
• Language: Swift
• Homepage:
• Size: 63.5 KB
• Stars: 4
• Watchers: 3
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Unofficial Swift for TensorFlow profiling suite

This repository contains pairs of Python TensorFlow programs and Swift TensorFlow programs for a head-to-head performance comparison.

All tests will report the interval between graph launch and graph finish.

**Note**: The timer in Swift will also include graph deserialization time, so comparisons below are not entirely unfair. I'm working to add hooks in the compiler runtime to just time the call to `TF_SessionRun`.

## My testing environment

- macOS Mojave (18A336e)

- 4.2 GHz Intel Core i7

- TensorFlow 1.9

- Swift for TensorFlow rev [5b36279](https://github.com/apple/swift/tree/5b36279e1cc868248c84ed9445098835c0b138f8)

## Benchmarks

### Graphs with the same ops and control flow constructs

Basic operations are being tested here: matrix multiplication, convolution and loops.

- [axpy_unrolled](axpy_unrolled): Straight-line `Ax+b`

- [axpy_loop](axpy_loop): `Ax+b` in a `while` loop

|                  |  Python |   Swift |

|------------------|---------|---------|

| axpy_unrolled    | 0.7358s | 1.7969s |

| axpy_loop        | 0.3705s | 0.5781s |

### `tf.Variable` vs. functional

Graph Program Extraction currently does not generate mutable variables in the TensorFlow graph for `var` declarations. Instead, each mutation of `var` in Swift source code gets lowered to producing a new value. In Python, however, `tf.Variable` is used everywhere and may be highly optimized. Here we compare purely functional parameter updates with mutation-based parameter updates.

- [parameter_update](parameter_update): Straight-line parameter update

|                  |  Python |   Swift |

|------------------|---------|---------|

| parameter_update | 0.8749s | 1.4375s |

### Vanilla models

_Coming soon_