Jovan Mitrevski comments

Results 189 comments of


                                            Jovan Mitrevski

Automatic type inference for `param_t` in Parametrised Activations

I just put Nicolo's test in the pytests instead of the one I had (with minor pre-commit changes). I also updated to the latest main so ideally the test failures...

update keras activation parsing, especially leaky relu

I also added support for the new v3 API parameter names for leaky relu.

update keras activation parsing, especially leaky relu

I want to understand the need for the assert change, so putting it to draft for now.

Add an optimizer to replace SeparableConv by Depthwise + Conv (pointwise)

I am somewhat torn as to whether to remove the regular accum_t and reuse_factor, which are not used if the separable is split, which is the default.

Add an optimizer to replace SeparableConv by Depthwise + Conv (pointwise)

I decided that it was cleaner to remove the unused attributes.

Add an optimizer to replace SeparableConv by Depthwise + Conv (pointwise)

Comment for reviewers: Note the changes in graph.py for easier passing of the configuration from one layer to child layers after transformations. This is used also in the qonnx PR....

Add an optimizer to replace SeparableConv by Depthwise + Conv (pointwise)

There are some template changes in the Vivado backend from when I thought I would make this PR handle multiplier factors != 1. In particular, I changed n_chan to n_filt...

oneAPI backend update: kernel and layer optimizations

I noticed, by the way, that ReLU uses blocking reads, and all the components use blocking writes. Is there a requirement to use nonblocking reads and writes? Note, we do...

Average pooling result_t

Ideally one would just increase the fractional component but not the integer component, with rounding left to the user if they want to reduce the width.

make auto the default for layer config

Added maximum precision to the keras options. In `InferPrecisionTypes`, `_infer_common_precision` now uses the maximum precision to not make the size too big. Note: for the maximum precision, basically the total...