Skip to content

[None][chore] Reduce mhc autotune tactics - #14329

Merged
lfr-0531 merged 2 commits into
NVIDIA:feat/deepseek_v4from
mingyangHao:mingyangh/mhc-v4pro-autotune
May 22, 2026
Merged

lfr-0531 merged 2 commits into
NVIDIA:feat/deepseek_v4from
mingyangHao:mingyangh/mhc-v4pro-autotune

Conversation

@mingyangHao

@mingyangHao mingyangHao commented May 20, 2026 •

Copy link
Copy Markdown
Collaborator

@coderabbitai summary

Description

Summary

Reduce MhcFusedHcRunner autotune time by pruning fused mHC tactics that did not win in the measured H=4096/H=7168 sweeps.

Changes:

  • Keep FMA backend tactics only for M <= 32
  • Use MMA-only candidates for M > 32
  • Collapse MMA tuning to the measured splitK / BigFuse ridge
  • Leave kernel implementations unchanged

Impact

For M <= 4096 with the existing 39 M buckets:

hidden_size Before After Reduction
4096 1386 522 62.3%
7168 1680 522 68.9%

Test Coverage

PR Checklist

Please review the following before submitting your PR:

  • PR description clearly explains what and why. If using CodeRabbit's summary, please make sure it makes sense.

  • PR Follows TRT-LLM CODING GUIDELINES to the best of your knowledge.

  • Test cases are provided for new code paths (see test instructions)

  • If PR introduces API changes, an appropriate PR label is added - either api-compatible or api-breaking. For api-breaking, include BREAKING in the PR title.

  • Any new dependencies have been scanned for license and vulnerabilities

  • CODEOWNERS updated if ownership changes

  • Documentation updated as needed

  • Update tava architecture diagram if there is a significant design change in PR.

  • The reviewers assigned automatically/manually are appropriate for the PR.

  • Please check this after reviewing the above items as appropriate for this PR.

GitHub Bot Help

To see a list of available CI bot commands, please comment /bot help.

@mingyangHao
mingyangHao force-pushed the mingyangh/mhc-v4pro-autotune branch 4 times, most recently from 6f0c641 to 9f9a0ce Compare May 20, 2026 05:10
@mingyangHao
mingyangHao marked this pull request as ready for review May 20, 2026 05:13
@mingyangHao
mingyangHao requested a review from a team as a code owner May 20, 2026 05:13
@mingyangHao
mingyangHao requested review from lfr-0531 and mikeiovine and removed request for a team and mikeiovine May 20, 2026 05:14
@mingyangHao

Copy link
Copy Markdown
Collaborator Author

/bot run --disable-fail-fast

@mingyangHao
mingyangHao force-pushed the mingyangh/mhc-v4pro-autotune branch from 9f9a0ce to 2e78c08 Compare May 20, 2026 06:20
@mingyangHao

Copy link
Copy Markdown
Collaborator Author

/bot run --disable-fail-fast

@mingyangHao
mingyangHao force-pushed the mingyangh/mhc-v4pro-autotune branch from 2e78c08 to afc3b7f Compare May 20, 2026 06:27
@lfr-0531

Copy link
Copy Markdown
Collaborator

/bot run --disable-fail-fast

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #49377 [ run ] triggered by Bot. Commit: afc3b7f Link to invocation

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #49377 [ run ] completed with state SUCCESS. Commit: afc3b7f
/LLM/main/L0_MergeRequest_PR pipeline #39030 completed with status: 'FAILURE'

CI Report

⚠️ Action Required:

  • Please check the failed tests and fix your PR
  • If you cannot view the failures, ask the CI triggerer to share details
  • Once fixed, request an NVIDIA team member to trigger CI again

CI Agent Failure Analysis

Link to invocation

Signed-off-by: Mingyang Hao <mingyangHao@users.noreply.github.com>
Signed-off-by: Mingyang Hao <mingyangHao@users.noreply.github.com>
@mingyangHao
mingyangHao force-pushed the mingyangh/mhc-v4pro-autotune branch from 3810fa3 to 42f661a Compare May 21, 2026 07:44
@lfr-0531

Copy link
Copy Markdown
Collaborator

/bot run --disable-fail-fast

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #49641 [ run ] triggered by Bot. Commit: 42f661a Link to invocation

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #49641 [ run ] completed with state SUCCESS. Commit: 42f661a
/LLM/main/L0_MergeRequest_PR pipeline #39256 completed with status: 'FAILURE'

CI Report

⚠️ Action Required:

  • Please check the failed tests and fix your PR
  • If you cannot view the failures, ask the CI triggerer to share details
  • Once fixed, request an NVIDIA team member to trigger CI again

CI Agent Failure Analysis

Link to invocation

@lfr-0531

Copy link
Copy Markdown
Collaborator

/bot run --disable-fail-fast

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #49714 [ run ] triggered by Bot. Commit: 42f661a Link to invocation

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #49714 [ run ] completed with state SUCCESS. Commit: 42f661a
/LLM/main/L0_MergeRequest_PR pipeline #39319 completed with status: 'SUCCESS'

CI Report

Link to invocation

@lfr-0531
lfr-0531 merged commit 93ed231 into NVIDIA:feat/deepseek_v4 May 22, 2026
6 checks passed
lfr-0531 pushed a commit to lfr-0531/TensorRT-LLM that referenced this pull request Jun 10, 2026
Signed-off-by: Mingyang Hao <mingyangHao@users.noreply.github.com>
Co-authored-by: Mingyang Hao <mingyangHao@users.noreply.github.com>
lfr-0531 pushed a commit to lfr-0531/TensorRT-LLM that referenced this pull request Jun 11, 2026
Signed-off-by: Mingyang Hao <mingyangHao@users.noreply.github.com>
Co-authored-by: Mingyang Hao <mingyangHao@users.noreply.github.com>
lfr-0531 pushed a commit to lfr-0531/TensorRT-LLM that referenced this pull request Jun 12, 2026
Signed-off-by: Mingyang Hao <mingyangHao@users.noreply.github.com>
Co-authored-by: Mingyang Hao <mingyangHao@users.noreply.github.com>
lfr-0531 pushed a commit to lfr-0531/TensorRT-LLM that referenced this pull request Jun 25, 2026
Signed-off-by: Mingyang Hao <mingyangHao@users.noreply.github.com>
Co-authored-by: Mingyang Hao <mingyangHao@users.noreply.github.com>
(cherry picked from commit 93ed231)
Signed-off-by: Mingyang <mhao1999@outlook.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants