A package for machine learning inference in FPGAs. We create firmware implementations of machine learning algorithms using high level synthesis language (HLS). We translate traditional open-source machine learning package models into HLS that can be configured for your use-case!

If you have any questions, comments, or ideas regarding hls4ml or just want to show us how you use hls4ml, don't hesitate to reach us through the discussions tab.

Documentation & Tutorial

For more information visit the webpage: https://fastmachinelearning.org/hls4ml/

Detailed tutorials on how to use hls4ml's various functionalities can be found here.

Installation

bash pip install hls4ml

To install the extra dependencies for profiling:

bash pip install hls4ml[profiling]

Getting Started

Creating an HLS project

```Python import hls4ml

Fetch a keras model from our example repository

This will download our example model to your working directory and return an example configuration file

config = hls4ml.utils.fetchexamplemodel('KERAS_3layer.json')

You can print the configuration to see some default parameters

print(config)

Convert it to a hls project

hlsmodel = hls4ml.converters.kerasto_hls(config)

Print full list of example models if you want to explore more

hls4ml.utils.fetchexamplelist() ```

Building a project with Xilinx Vivado HLS (after downloading and installing from here)

Note: Vitis HLS is not yet supported. Vivado HLS versions between 2018.2 and 2020.1 are recommended.

```Python

Use Vivado HLS to synthesize the model

This might take several minutes

hls_model.build()

Print out the report if you want

hls4ml.report.readvivadoreport('my-hls-test') ```

Citation

If you use this software in a publication, please cite the software bibtex @software{fastml_hls4ml, author = {{FastML Team}}, title = {fastmachinelearning/hls4ml}, year = 2023, publisher = {Zenodo}, version = {v0.7.0}, doi = {10.5281/zenodo.1201549}, url = {https://github.com/fastmachinelearning/hls4ml} } and first publication: bibtex @article{Duarte:2018ite, author = "Duarte, Javier and others", title = "{Fast inference of deep neural networks in FPGAs for particle physics}", eprint = "1804.06913", archivePrefix = "arXiv", primaryClass = "physics.ins-det", reportNumber = "FERMILAB-PUB-18-089-E", doi = "10.1088/1748-0221/13/07/P07027", journal = "JINST", volume = "13", number = "07", pages = "P07027", year = "2018" } Additionally, if you use specific features developed in later papers, please cite those as well. For example, CNNs: bibtex @article{Aarrestad:2021zos, author = "Aarrestad, Thea and others", title = "{Fast convolutional neural networks on FPGAs with hls4ml}", eprint = "2101.05108", archivePrefix = "arXiv", primaryClass = "cs.LG", reportNumber = "FERMILAB-PUB-21-130-SCD", doi = "10.1088/2632-2153/ac0ea1", journal = "Mach. Learn. Sci. Tech.", volume = "2", number = "4", pages = "045015", year = "2021" } @article{Ghielmetti:2022ndm, author = "Ghielmetti, Nicol\`{o} and others", title = "{Real-time semantic segmentation on FPGAs for autonomous vehicles with hls4ml}", eprint = "2205.07690", archivePrefix = "arXiv", primaryClass = "cs.CV", reportNumber = "FERMILAB-PUB-22-435-PPD", doi = "10.1088/2632-2153/ac9cb5", journal ="Mach. Learn. Sci. Tech.", year = "2022" } binary/ternary networks: bibtex @article{Loncar:2020hqp, author = "Ngadiuba, Jennifer and others", title = "{Compressing deep neural networks on FPGAs to binary and ternary precision with HLS4ML}", eprint = "2003.06308", archivePrefix = "arXiv", primaryClass = "cs.LG", reportNumber = "FERMILAB-PUB-20-167-PPD-SCD", doi = "10.1088/2632-2153/aba042", journal = "Mach. Learn. Sci. Tech.", volume = "2", pages = "015001", year = "2021" }