8000 GitHub - RyanTomich/photonic_compiler: compilation with optimization and code generation that can take neural network inference requests and divide the load between classical hardware and novel photonic hardware.
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content

compilation with optimization and code generation that can take neural network inference requests and divide the load between classical hardware and novel photonic hardware.

Notifications You must be signed in to change notification settings

RyanTomich/photonic_compiler

Repository files navigation

GitHub commit activity

Automated Compiler Software for Emerging Photonic Computing Hardware

drawing

*Contributions highlighted in green

Overview

This project aims to make a compiler with optimization and code generation that can take neural network inference requests and divide the load between classical hardware (CPU/ GPU) and novel photonic hardware. We take advantage of the TVM compiler to translate Tensorflow and Pytorch models into their internal Relay IR, at which point this compiler conducts the next layer of scheduling and translation.

File sections and definitions

  • inference_pratice: recreating popular modles from scrach using numpy

  • json_parse: IN PROGRESS. Everything related to code generation and parsing of TVM Relay IR .json files

    • parser.py: script for instruction generation and file structure
    • simple_LeNet_parsed.txt: generated instructions
  • ONNX-ResNet: model files/parameters for ResNet ML model in the ONNX format

  • Pytorch-LeNet: model files/parameters and code for loading LeNet from pytorch

  • Transformer-GPT2 Compile GPT2 to Relay IR

About

compilation with optimization and code generation that can take neural network inference requests and divide the load between classical hardware and novel photonic hardware.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 2

  •  
  •  
0