Skip to content
This repository has been archived by the owner on Oct 23, 2023. It is now read-only.

finn-base v0.0.2b

Pre-release
Pre-release
Compare
Choose a tag to compare
@maltanar maltanar released this 04 Jun 23:01
· 33 commits to dev since this release
d4b80dd

Highlights:

  • Generic partitioning + ExtendPartition transformations for working with large models
  • Fast-mode data packing for binary values
  • Changes to support non-equal image dimensions / kernel / stride / .. for convolutions

PRs:

  • Add ZCU111 board to part map #32 by fpjentzsch
  • Support infer_datatype for flatten layer #30 by fpjentzsch
  • Changes for supporting non-equal dilation #29 by mmrahorovic
  • Changes for supporting vitis_hls #28 by maltanar
  • Added ExtendPartition transformation #27 by mmrahorovic
  • Remove redundant output value_info entries #26 by mmrahorovic
  • Added support for non-equal strides along different axes #25 by mmrahorovic
  • Experimental: fast_mode data packing for binary #24 by maltanar
  • Generic partitioning feature #23 by fpjentzsch
  • Update quantavgpool2d.py #22 by jalezeta
  • Changed how the epsilon value is set #21 by mmrahorovic
  • Support for non-square input images and kernels for im2col node #20 by mmrahorovic
  • Added 3D to 4D (tensor) transformation #19 by mmrahorovic
  • Modified set_nodeattr to allow using it on repeated fields #18 by mmrahorovic
  • Added support for dilation value = 2 for 1D and 2D images/kernels #17 by mmrahorovic
  • Support for non-square input images and kernels in LowerConvsToMatMul transformation #16 by mmrahorovic