Skip to content

Custom Kernel for MORR linear layer#5

Open
Johnny1882 wants to merge 2 commits into
DeepWok:masterfrom
Johnny1882:morr_linear
Open

Custom Kernel for MORR linear layer#5
Johnny1882 wants to merge 2 commits into
DeepWok:masterfrom
Johnny1882:morr_linear

Conversation

@Johnny1882

Copy link
Copy Markdown

Custom CUDA kernel for MORR linear layer

What’s inside

  • Fused forward + backward kernel for Multi-Operand Ring Resonators (MORR) linear layer simulation module.
  • Optional activation checkpointing → significant less peak GPU memory.

Trade-off

  • Training throughput ↓ ~50 %.

Memory Evaluation:
Memory bar chart

Inference & Training Runtime:

Training + Inference Inference-only

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.

1 participant