Adding pruning to Brevitas

[seniorgoogles]

Hi,

i am working on a project in which pruning is to be used. In my opinion, there are two options for this. Either you add a quantizer that does the pruning or you implement pruning in the same way as quantizing was implemented by implementing the same structure (injectors, proxies, enums, etc.).

I have implemented both concepts "prototypically", whereby embedding pruning in a quantizer was associated with significantly less effort, but is less "nice" in some places.

What do you think, which variant would be preferable or are there any other ideas in this regard?

[Giuseppe5]

Hello,

Without taking a closer look at your proposal it is difficult to say what is the "best" solution.

Ultimately, Brevitas is mostly focus on quantization for now, so it all depends on what solution fits and solves your problem best.