ElasticAI.creator

Design, train and compile neural networks optimized specifically for FPGAs. Obtaining a final model is typically a three stage process.

• design and train it using the layers provided in the elasticai.creator.nn package.
• translate the model to a target representation, e.g. VHDL
• compile the intermediate representation with a third party tool, e.g. Xilinx Vivado (TM)

This version currently only supports parts of VHDL as target representations.

The project is part of the elastic ai ecosystem developed by the Embedded Systems Department of the University Duisburg-Essen. For more details checkout the slides at researchgate.

Table of contents

• Users Guide
  • Install
  • Minimal Example
  • Features
    • Supported Network Architectures and Layers
    • Planned Network Architectures
    • Modules in Development
    • Deprecated Modules
    • General Limitations
  • Structure of the Project

Users Guide

Install

You can install the ElasticAI.creator as a dependency using pip:

python3 -m pip install "elasticai.creator"

On PyPi the latest tagged version is published.

Currently, we do not automatically pack and push the code to PyPi. If you want to make sure to use the latest version from the main branch, you can install the ElasticAI.creator as a dependency via git:

python3 -m pip install git+https://github.com/es-ude/elastic-ai.creator.git@main

Minimal Example

In examples you can find a minimal example. It shows how to use the ElasticAI.creator to define and translate a machine learning model to VHDL. It will save the generated VHDL code to a directory called build_dir. Furthermore, it will generate a skeleton for the Elastic Node V5 that you can use to interface with your machine learning model on the FPGA via a C stub (defined in the elastic-ai.runtime.enV5).

Features

Supported network architectures and layers

• all sequential network architectures representable with torch.nn.Sequential
• fixed-point quantized:
  • layers: linear, linear with batch normalization, LSTM
  • activations: hard sigmoid, hard tanh, ReLU
    • precomputed: sigmoid, tanh, adaptable SiLU

Planned network architectures and layers supported in the future

• integer-only linear quantization
• 1D convolutional layers (fixed-point)
• gated recurrent unit (fixed-point)

Modules in development:

• elasticai.creator.nn.fixed_point.conv1d

Deprecated modules (removal up to discussion):

• elasticai.creator.nn.binary (binary quantization)
• elasticai.creator.nn.float (limited-precision floating-point quantization)
• elasticai.creator.nn.fixed_point.mac

General limitations

By now we only support sequential models for our translations. That excludes skip and residual connections.

Structure of the Project

The structure of the project is as follows. The creator folder includes all main concepts of our project, especially the qtorch implementation which is our implementation of quantized PyTorch layer. It also includes the supported target representations, like the subfolder nn is for the translation to vhdl. Additionally, we have unit and integration tests in the tests folder.