Skip to content

Conversation

@sajadn
Copy link
Contributor

@sajadn sajadn commented Jan 14, 2026

  • introduced ConditionalChainedOptimizer which makes it possible to have optimizers dedicated to only a particular submodule within the model.
  • Implemented WanDMD which has additional arguments in the forward path such as feature_indices and fwd_pred_type that are necessary for dmd.
  • Implemented WanDMDCombined which creates three submodules of fake_score, teacher and student (net).
  • Added DMDOptimizerProvider which instantiate the conditionalChainedOptimizer and set the conditions for each optimizer.

Signed-off-by: sajadn <snorouzi@nvidia.com>
Signed-off-by: sajadn <snorouzi@nvidia.com>
@copy-pr-bot
Copy link

copy-pr-bot bot commented Jan 14, 2026

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

Signed-off-by: sajadn <snorouzi@nvidia.com>
Signed-off-by: sajadn <snorouzi@nvidia.com>
…pt, remove 4 step generation of the student and just call fastgen approach.

Signed-off-by: sajadn <snorouzi@nvidia.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants