Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Follow-up on analog info #336

Open
maljoras opened this issue Dec 15, 2021 · 3 comments
Open

Follow-up on analog info #336

maljoras opened this issue Dec 15, 2021 · 3 comments
Assignees
Labels
enhancement New feature or request

Comments

@maljoras
Copy link
Collaborator

Description and motivation

#316 introduced an analog info functionality. Going forward, we might want to enhance it further by adding the following aspects:

  • Add some more summary info like remaining digital-MACs, total number of analog tile mat-vecs
  • User-definable "columns" to select only some of the columns to display
  • Improve on layout, e.g. the (- - - ) first entry of the tile section is a little confusing
  • Count also the peripheral digital ops (e.g noise management, bound management)
  • Currently only forward pass is counted, but backward / update would be interesting to add for analog training
  • Add other specifics for analog training with analog optimizers (tiki-taka / mixed precision) which need additional tiles internally and additional digital ops
@maljoras maljoras added the enhancement New feature or request label Dec 15, 2021
@kaoutar55
Copy link
Collaborator

Hi @IHIaadj will you be able to add these enhacements.

@kaoutar55
Copy link
Collaborator

@IHIaadj can you please let us know if you are interested in incorporating some of this feedback to analog info.

@anissbslh
Copy link

anissbslh commented Jan 25, 2025

Hi,
we are working on these new enhancements

For now, we added:

  • An improved layout
  • Display of total digital MACs
  • Display of total analog tile matrix-vector multiplications (Mat-Vecs)
  • User-selectable columns for a customizable display

Regarding the remaining features, I've been wondering:

Count also the peripheral digital ops (e.g noise management, bound management)

  • How can these operations be accurately estimated?

Add other specifics for analog training with analog optimizers (tiki-taka / mixed precision) which need additional tiles internally and additional digital ops

  • How should the additional digital operations and tiles be calculated in this context?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
enhancement New feature or request
Projects
None yet
Development

No branches or pull requests

4 participants