Skip to content

Conversation

@fmassa
Copy link
Member

@fmassa fmassa commented Dec 10, 2025

Using repeated_subgraphs enable AutoParallel to enforce the same solution to repeated regions of the model, making the solver be much faster. For 32 layers, we obtained roughtly 21x speedup compared to without repeated subgraphs.

@fmassa fmassa requested a review from xmfan December 10, 2025 14:28
@meta-cla meta-cla bot added the CLA Signed This label is managed by the Meta Open Source bot. label Dec 10, 2025
@xmfan
Copy link
Member

xmfan commented Dec 10, 2025

psutil issue should be fixed with next nightly now that pytorch/pytorch#169985 landed 3h ago

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

CLA Signed This label is managed by the Meta Open Source bot.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants