Fast inference using hls4ml

Following up on the great week at the NGTHackathon, a few highlights of what we achieved:

  • Two students made solid progress on converting their ML model from PyTorch to hls4ml. Thanks to close interactions with hls4ml developers, they gained a much clearer understanding of the challenges they faced and even helped improve the ecosystem by reporting several bugs.
  • They also got to test PQuant, a new tool provided by NGT for pruning and quantizing models, and shared valuable user feedback that will help guide future improvements.
  • We explored a new workflow enabling parallel synthesis of hls4ml models — a promising direction for speeding up development.

But just as importantly, the week was full of great exchanges, new project ideas, and meaningful connections that will carry forward beyond the hackathon!