Skip to content

Version History

Jump to bottom
Jordan Welsman edited this page Jan 19, 2023 · 7 revisions

nexport was mostly developed privately, so it didn't have a well-documented history of feature releases until it was officially released with its license.

Table of Contents

0.0.0 (10.13.2022)

  • Project started
    • GitHub repository created
  • Project created
    • Basic README.md written

0.0.1 (10.15.2022)

  • Package files created
    • setup.py and LICENSE.md

0.1.0 (10.27.2022)

  • Initial export function implemented
    • PyTorch 🔥 👉 .txt export
    • Output file 💾 contains a dump of all trained parameters

0.1.1 (11.02.2022)

  • Revised .txt file structure
    • Output file 💾 now contains organized trained parameters formatted for ease of importing into inference-engine 🧠
  • Export function polished
    • PyTorch 🔥 👉 .txt export

0.1.2 (11.16.2022)

  • Added colored output 🌈 to functions
    • Success ✅ messages

0.2.0 (12.23.2022)

  • Export function implemented
    • PyTorch 🔥 👉 .json
    • Output file 💾 contains a crude object dump of trained parameters grouped by layer and neuron

0.2.1 (01.04.2023)

  • Reserved PyPI module page
  • Revised .json data structure
    • Output file 💾 now contains a more organized and human-readable structure grouped by layer only
  • Export function polished
    • PyTorch 🔥 👉 .json
    • Implemented optional metadata flag 🚩 to include model information at the top of the output file
  • Implemented incremental verbose flag 🚩 to report status of model export
    • 0 (No output)
    • 1 (Export complete)
    • 2 (Time taken to export)
    • 3 (Layer-by-layer export status)

0.2.2 (01.08.2023)

  • Elevated module access
    • Module functions can now be called from nexport.<function>
  • Privatized externally irrelevant functionality
    • Removed access ⛔ to functions that should only be used by internal functions

0.3.0 - The Parameter Update (tbc)

  • Implemented trained parameter calculator
    • Usable with nexport.calculate_params()
  • Implemented calculator submodule
    • Implemented layer count function
      • Usable with model_info = nexport.calcualte_layers(model)
    • Implemented neuron count function
      • Usable with model_info = nexport.calculate_neurons(model)
    • Implemented parameter calculation function
      • Usable with model_info = nexport.calculate_params(model), model_info.neuron_count(<layer_number>)
  • Moved example networks to separate class submodule
    • FFNetwork (2-3-3-1)
    • BFNetwork (28^2, 256, 128, 64, 32, 16, 8)
    • ICARNetwork (10, [64*50], 1)

0.4.0 - The TensorFlow Update (tbc)

  • Export function implemented
    • Keras 🟥 👉 .txt
  • nexport output now uses jutils for colored output 🌈

Created and maintained by Jordan Welsman @ LBNL.

nexport Copyright (c) 2022-2023, The Regents of the University of California, through Lawrence Berkeley National Laboratory (subject to receipt of any required approvals from the U.S. Dept. of Energy).

All rights reserved.

nexport

Clone this wiki locally