The latest stable release is v0.2.0, including a validated boosted decision tree implementation (arXiv:2002.02534) and binary/ternary neural networks (arXiv:2003.06308).


A list of suppported ML codes and architectures, including a summary table is below. Dependences are given in a dedicated page.

ML code support:

  • Keras/Tensorflow, PyTorch, scikit-learn
  • Planned: xgboost

Neural network architectures:

  • Fully Connected NNs (multi-layer perceptron)
  • Boosted Decision Trees
  • Convolutional NNs (1D/2D), in beta testing
  • Recurrent NN/LSTM, in prototyping

A summary of the on-going status of the hls4ml tool is in the table below.

Architectures/Toolkits Keras/TensorFlow PyTorch scikit-learn
MLP supported supported -
Conv1D/Conv2D supported in development -
BDT - - supported
RNN/LSTM in development - -

Other random feature notes:

  • There is a known Vivado HLS issue where the large loop unrolls create memory issues during synthesis. We are working to solve this issue but you may see errors related to this depending on the memory of your machine. Please feel free to email the hls4ml team if you have any further questions.

Example models

We also provide and documented several example models that have been implemented in hls4ml in this Github repository.

results matching ""

    No results matching ""