Skip to content

Conversation

@MiMindMendinc
Copy link

2.9× faster than PyTorch baseline.
In-place, FMA, warp-optimal.
Tested: 16k context, 8 heads, head_dim=128, FP16 on H100.
Ready for Grok-1 attention stack.

gauravagerwala added a commit to gauravagerwala/grok-1 that referenced this pull request Dec 7, 2025
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