Skip to content

[TRTLLM-13458][feat] Support Minimax M3 MXFP8 checkpoint - #15687

Merged
WeiHaocheng merged 2 commits into
NVIDIA:mainfrom
WeiHaocheng:feat/minimax_m3_mxfp8_support
Jul 1, 2026
Merged

[TRTLLM-13458][feat] Support Minimax M3 MXFP8 checkpoint#15687
WeiHaocheng merged 2 commits into
NVIDIA:mainfrom
WeiHaocheng:feat/minimax_m3_mxfp8_support

Conversation

@WeiHaocheng

@WeiHaocheng WeiHaocheng commented Jun 27, 2026

Copy link
Copy Markdown
Collaborator

Summary by CodeRabbit

  • New Features

    • Added support for the MiniMax-M3 MXFP8 checkpoint across deployment examples, curated configs, and accuracy coverage.
    • Expanded guidance for using BF16 and MXFP8, including recommended settings and supported serving limits.
  • Bug Fixes

    • Updated sparse-attention and runtime settings to better match MiniMax-M3 serving requirements and avoid CUDA graph capture issues.
  • Documentation

    • Revised the deployment guide and examples to use the new MXFP8 model reference and updated test/benchmark commands.
  • Tests

    • Added MXFP8 accuracy validation and updated expected reference thresholds for MiniMax-M3.

Description

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.

@coderabbitai

coderabbitai Bot commented Jun 27, 2026

Copy link
Copy Markdown
Contributor

Review Change Stack

📝 Walkthrough

Walkthrough

The PR updates MiniMax M3 references, configs, docs, routing, and tests to add MXFP8 alongside BF16. It changes deployment examples, throughput settings, accuracy references, and adds an MXFP8 integration test case.

Changes

MiniMax M3 MXFP8 support

Layer / File(s) Summary
Model metadata
docs/source/_static/config_db.json, examples/configs/curated/lookup.yaml
Curated MiniMax M3 model entries and lookup mappings switch from MiniMaxAI/MiniMax-M3 to nvidia/MiniMax-M3-MXFP8, with matching display and URL fields updated.
Serving config and routing
examples/configs/curated/minimax-m3-throughput.yaml, tensorrt_llm/_torch/modules/fused_moe/routing.py
The throughput config adds max_seq_len, sparse-attention selection, and cache-fraction changes, and MiniMaxM3MoeRoutingMethod.apply accepts and forwards optional input_ids.
Deployment guide
docs/source/deployment-guide/deployment-guide-for-minimax-m3-on-trtllm.md
The guide adds BF16/MXFP8 sections, expands feature notes, switches the launch example to MXFP8 by default, and updates the test and benchmark snippets to use the MXFP8 model id.
Accuracy coverage
tests/integration/defs/accuracy/references/gsm8k.yaml, tests/integration/defs/accuracy/references/mmlu.yaml, tests/integration/defs/accuracy/test_llm_api_pytorch.py, tests/integration/test_lists/qa/llm_function_core.txt
GSM8K and MMLU references add MXFP8 entries and updated thresholds, TestMiniMaxM3 adds test_mxfp8, and the QA list includes the new timeout case.

Sequence Diagram(s)

sequenceDiagram
  participant ComponentA
  participant ComponentB
  ComponentA->>ComponentB: observable interaction
Loading

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~25 minutes

Possibly related PRs

  • NVIDIA/TensorRT-LLM#15587: Updates the same MiniMax M3 deployment guide with MXFP8/BF16 content and related launch/test examples.

Suggested reviewers

  • nv-guomingz
  • xinhe-nv
  • chang-l
🚥 Pre-merge checks | ✅ 3 | ❌ 2

❌ Failed checks (2 warnings)

Check name Status Explanation Resolution
Docstring Coverage ⚠️ Warning Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%. Write docstrings for the functions missing them to satisfy the coverage threshold.
Description check ⚠️ Warning The PR body only contains the template and checklist checkbox, with no actual issue, solution, or test coverage details filled in. Add a short Description and Test Coverage section that explains the change, why it was needed, and which tests verify the new MXFP8 path.
✅ Passed checks (3 passed)
Check name Status Explanation
Title check ✅ Passed The title follows the repository template and clearly summarizes the MXFP8 MiniMax M3 support change.
Linked Issues check ✅ Passed Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check ✅ Passed Check skipped because no linked issues were found for this pull request.
✨ Finishing Touches
🧪 Generate unit tests (beta)
  • Create PR with unit tests

Comment @coderabbitai help to get the list of available commands.

@coderabbitai coderabbitai Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Actionable comments posted: 2

🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@docs/source/deployment-guide/deployment-guide-for-minimax-m3-on-trtllm.md`:
- Line 16: The intra-doc link in the deployment guide points to a non-existent
anchor, so update the Markdown link in the opening description to target a real
heading or add the missing section with a matching slug. Check the referenced
“Choosing the Parallelism Strategy” text and the document’s existing headings to
make sure the fragment matches exactly and the link resolves correctly.
- Line 127: The refreshed serve instructions are using the non-preferred
configuration path for `trtllm-serve`. Update the guidance in the deployment doc
to point users to `--config` instead of `--extra_llm_api_options` and
`$EXTRA_LLM_API_FILE`, and keep the rest of the setup language aligned with the
`trtllm-serve` flow. Make sure the section that discusses tuning YAML-based
settings and selecting the checkpoint directory still references the same serve
workflow, but with `--config` as the documented mechanism.
🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

  • Push a commit to this branch (recommended)
  • Create a new PR with the fixes

ℹ️ Review info
⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 4d9884ea-2c7e-491c-98d1-edc3b12253d7

📥 Commits

Reviewing files that changed from the base of the PR and between eaf5693 and 7fd1c59.

📒 Files selected for processing (9)
  • docs/source/_static/config_db.json
  • docs/source/deployment-guide/deployment-guide-for-minimax-m3-on-trtllm.md
  • examples/configs/curated/lookup.yaml
  • examples/configs/curated/minimax-m3-throughput.yaml
  • tensorrt_llm/_torch/modules/fused_moe/routing.py
  • tests/integration/defs/accuracy/references/gsm8k.yaml
  • tests/integration/defs/accuracy/references/mmlu.yaml
  • tests/integration/defs/accuracy/test_llm_api_pytorch.py
  • tests/integration/test_lists/qa/llm_function_core.txt

@WeiHaocheng

Copy link
Copy Markdown
Collaborator Author

/bot run --disable-fail-fast

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #56178 [ run ] triggered by Bot. Commit: 7fd1c59 Link to invocation

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #56178 [ run ] completed with state SUCCESS. Commit: 7fd1c59
/LLM/main/L0_MergeRequest_PR pipeline #45041 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: WeiHaocheng <20514172+WeiHaocheng@users.noreply.github.com>
Signed-off-by: WeiHaocheng <20514172+WeiHaocheng@users.noreply.github.com>
@WeiHaocheng
WeiHaocheng force-pushed the feat/minimax_m3_mxfp8_support branch from bca2804 to ac4885f Compare July 1, 2026 03:19
@WeiHaocheng

Copy link
Copy Markdown
Collaborator Author

/bot run --disable-fail-fast

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #56808 [ run ] triggered by Bot. Commit: ac4885f Link to invocation

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #56808 [ run ] completed with state SUCCESS. Commit: ac4885f
/LLM/main/L0_MergeRequest_PR pipeline #45624 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

@WeiHaocheng
WeiHaocheng enabled auto-merge (squash) July 1, 2026 15:00
@WeiHaocheng

Copy link
Copy Markdown
Collaborator Author

/bot run

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #56951 [ run ] triggered by Bot. Commit: ac4885f Link to invocation

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #56951 [ run ] completed with state FAILURE. Commit: ac4885f
/LLM/main/L0_MergeRequest_PR pipeline #45757 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

@WeiHaocheng

Copy link
Copy Markdown
Collaborator Author

/bot run

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #56967 [ run ] triggered by Bot. Commit: ac4885f Link to invocation

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #56967 [ run ] completed with state SUCCESS. Commit: ac4885f
/LLM/main/L0_MergeRequest_PR pipeline #45769 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

@WeiHaocheng

Copy link
Copy Markdown
Collaborator Author

/bot run

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #57015 [ run ] triggered by Bot. Commit: ac4885f Link to invocation

@brb-nv

brb-nv commented Jul 1, 2026

Copy link
Copy Markdown
Collaborator

Retriggering as waive MR just merged: #15850

@brb-nv

brb-nv commented Jul 1, 2026

Copy link
Copy Markdown
Collaborator

/bot run --disable-fail-fast

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #57024 [ run ] triggered by Bot. Commit: ac4885f Link to invocation

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #57015 [ run ] completed with state ABORTED. Commit: ac4885f

Link to invocation

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #57024 [ run ] completed with state SUCCESS. Commit: ac4885f
/LLM/main/L0_MergeRequest_PR pipeline #45820 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

@brb-nv

brb-nv commented Jul 1, 2026

Copy link
Copy Markdown
Collaborator

/bot run --disable-fail-fast

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #57038 [ run ] triggered by Bot. Commit: ac4885f Link to invocation

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #57038 [ run ] completed with state SUCCESS. Commit: ac4885f
/LLM/main/L0_MergeRequest_PR pipeline #45833 completed with status: 'SUCCESS'

CI Report

Link to invocation

@WeiHaocheng
WeiHaocheng merged commit 9de2850 into NVIDIA:main Jul 1, 2026
10 of 11 checks passed
evezhier pushed a commit to evezhier/TensorRT-LLM that referenced this pull request Jul 2, 2026
Signed-off-by: WeiHaocheng <20514172+WeiHaocheng@users.noreply.github.com>
BrianLi23 pushed a commit to BrianLi23/TensorRT-LLM that referenced this pull request Jul 9, 2026
Signed-off-by: WeiHaocheng <20514172+WeiHaocheng@users.noreply.github.com>
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.

6 participants