Add proper support for Binarized and binary neural nets in the QKeras converter#141

Open

jurevreca12 wants to merge 29 commits intofastmachinelearning:mainfrom jurevreca12:bnn_support_in_qkeras_converter

+411-100

Commits on Mar 22, 2024

Added test, docs/, and updated resolve_rounding_mode function to return new rounding modes.
jvreca
committed
Fix table visualization.
jvreca
committed
Fix table visualization again.
jvreca
committed

Commits on Aug 9, 2024

Fixed converter to allow alpha/scale to be a tensor
jvreca
committed
Added a check to see if tensor is representable by the quantization parameters.
jvreca
committed

Commits on Aug 12, 2024

Extra Mul node inserted only when neccessary
jvreca
committed
Added parameterized test for tensor style alpha.
jvreca
committed

Commits on Aug 14, 2024

Updated QKeras converter to support auto_po2
jvreca
committed
Added check for none.
jvreca
committed
Fixed pre-commit issues.
jvreca
committed
Added check if tensor is repsentable with quant setting.s
jvreca
committed

Commits on Aug 22, 2024

Commits on Aug 23, 2024

Addded del_initializer to modelwrapper.
jvreca
committed

Commits on Aug 26, 2024

Commits on Aug 27, 2024

Commits on Aug 28, 2024

Simplified qlayer converter.
jvreca
committed
Fixed bug in _add_quant_node_on_input func.
jvreca
committed
Outputing bipolarQuant node for binary inputs.
jvreca
committed

Commits on Aug 29, 2024

Binarized networks now seem to work (with BinaryQuant)
jvreca
committed
Further cleaned up converter.
jvreca
committed
Input quantizer logic now working also for qkeras.binary quantizer.
jvreca
committed