k_propagate CUDA kernel saturation #9

Open

Labels

UtkarshJoshiNtl

opened

when N is small (<50), GPU launch overhead dominates. The auto-backend in accelerator.py doesn't account for this; it always picks CUDA if available. Should fall back to C++ for small N.

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

Metadata

Assignees

No one assigned

Labels

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests