Skip to content

1. Parallelize grouped low-rank output projection (SumHeads) across NUMA - #1009

Merged
copybara-service[bot] merged 1 commit into
devfrom
test_974965810
Sep 3, 2026
Merged

1. Parallelize grouped low-rank output projection (SumHeads) across NUMA#1009
copybara-service[bot] merged 1 commit into
devfrom
test_974965810

Conversation

@copybara-service

Copy link
Copy Markdown
  1. Parallelize grouped low-rank output projection (SumHeads) across NUMA
    clusters via ParallelFor over o_groups.
  2. Direct strided matrix view (C_g) into mla_o_mid destination buffer,
    eliminating intermediate copy per group.
  3. Demote att_out directly from float to BF16 into per-cluster mla_o_in
    buffers, switching mla_o_mid to BF16 to reduce memory footprint and
    bandwidth.

   clusters via ParallelFor over o_groups.
2. Direct strided matrix view (C_g) into mla_o_mid destination buffer,
   eliminating intermediate copy per group.
3. Demote att_out directly from float to BF16 into per-cluster mla_o_in
   buffers, switching mla_o_mid to BF16 to reduce memory footprint and
   bandwidth.

PiperOrigin-RevId: 975630365
@copybara-service
copybara-service Bot merged commit 30b178e into dev Sep 3, 2026
11 checks passed
@copybara-service
copybara-service Bot deleted the test_974965810 branch September 3, 2026 10:20
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.

0 participants