Skip to content

[https://nvbugs/5800725][fix] Restore Mistral Large 3 text-only processor#14248

Merged
byshiue merged 6 commits into
mainfrom
fix/nvbug_5800725
May 28, 2026
Merged

[https://nvbugs/5800725][fix] Restore Mistral Large 3 text-only processor#14248
byshiue merged 6 commits into
mainfrom
fix/nvbug_5800725

Conversation

@byshiue
Copy link
Copy Markdown
Collaborator

@byshiue byshiue commented May 18, 2026

Bug

https://nvbugspro.nvidia.com/bug/5800725

Root Cause

For mistral_large_3, self.text_processor was assigned to MistralCommonImageProcessor. That processor requires positional images, so text-only requests failed before generation.

Fix

Restore AutoProcessor as the text-only processor and keep MistralCommonImageProcessor only for text+image processing.

Unwaived Test

Removed the waiver for:
accuracy/test_llm_api_pytorch.py::TestMistralLarge3_675B::test_nvfp4_4gpus[latency_moe_trtllm]

Validation

Summary by CodeRabbit

  • Bug Fixes

    • Improved initialization of the Mistral Large 3 model input processor for more efficient resource handling.
  • Tests

    • Re-enabled accuracy validation test for Mistral Large 3 model that was previously skipped, indicating improved compatibility.

Review Change Stack

byshiue and others added 2 commits May 18, 2026 01:23
Signed-off-by: bhsueh <11360707+byshiue@users.noreply.github.com>
…only path

MistralCommonImageProcessor.__call__ requires an images positional
argument and always applies apply_chat_template, so text-only inputs
(e.g. MMLU/GSM8K prompts) raised TypeError before any inference could
run. Construct AutoProcessor unconditionally and use it as
text_processor for mistral_large_3, matching the intent documented in
the adjacent comment.

Signed-off-by: tensorrt-cicd <90828364+tensorrt-cicd@users.noreply.github.com>
Signed-off-by: bhsueh <11360707+byshiue@users.noreply.github.com>
@byshiue byshiue requested review from a team as code owners May 18, 2026 08:30
@byshiue byshiue requested review from jdebache and symphonylyh May 18, 2026 08:30
@byshiue
Copy link
Copy Markdown
Collaborator Author

byshiue commented May 18, 2026

/bot run

@coderabbitai
Copy link
Copy Markdown
Contributor

coderabbitai Bot commented May 18, 2026

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info
⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: aa8759b5-d442-4644-b65e-ac561f22661a

📥 Commits

Reviewing files that changed from the base of the PR and between f830224 and 494000e.

📒 Files selected for processing (2)
  • tensorrt_llm/_torch/models/modeling_mistral.py
  • tests/integration/test_lists/waives.txt
💤 Files with no reviewable changes (1)
  • tests/integration/test_lists/waives.txt

📝 Walkthrough

Walkthrough

This PR refactors Mistral3InputProcessor initialization to unconditionally create AutoProcessor.from_pretrained upfront, then branches on model_type to assign _processor appropriately. A corresponding test waiver entry is removed, indicating the fix enables a previously-skipped test to pass.

Changes

Mistral3 Input Processor Initialization

Layer / File(s) Summary
Mistral3InputProcessor initialization refactoring
tensorrt_llm/_torch/models/modeling_mistral.py
AutoProcessor.from_pretrained is now instantiated unconditionally before model_type branching. The mistral_large_3 branch sets _processor to MistralCommonImageProcessor, while other mistral types set _processor to auto_processor.
Test waiver removal
tests/integration/test_lists/waives.txt
Removed the waiver entry for TestMistralLarge3_675B::test_nvfp4_4gpus[latency_moe_trtllm], confirming the test now passes with the processor initialization fix.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~8 minutes

Suggested reviewers

  • zhenhuaw-me
  • niukuo
  • jieli-matrix
🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

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.
✅ Passed checks (4 passed)
Check name Status Explanation
Title check ✅ Passed The title clearly identifies the main change: restoring Mistral Large 3 text-only processor for the mistral_large_3 model type, with appropriate bug reference and fix type prefix.
Description check ✅ Passed The description covers root cause, fix, test coverage (waiver removal), and validation steps; it aligns well with the template sections for explanation and test coverage.
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.

✏️ Tip: You can configure your own custom pre-merge checks in the settings.

✨ Finishing Touches
📝 Generate docstrings
  • Create stacked PR
  • Commit on current branch
🧪 Generate unit tests (beta)
  • Create PR with unit tests
  • Commit unit tests in branch fix/nvbug_5800725

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

@tensorrt-cicd
Copy link
Copy Markdown
Collaborator

PR_Github #48874 [ run ] triggered by Bot. Commit: 494000e Link to invocation

@tensorrt-cicd
Copy link
Copy Markdown
Collaborator

PR_Github #48874 [ run ] completed with state SUCCESS. Commit: 494000e
/LLM/main/L0_MergeRequest_PR pipeline #38624 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

@byshiue
Copy link
Copy Markdown
Collaborator Author

byshiue commented May 19, 2026

/bot run

@tensorrt-cicd
Copy link
Copy Markdown
Collaborator

PR_Github #49020 [ run ] triggered by Bot. Commit: 494000e Link to invocation

@tensorrt-cicd
Copy link
Copy Markdown
Collaborator

PR_Github #49020 [ run ] completed with state SUCCESS. Commit: 494000e
/LLM/main/L0_MergeRequest_PR pipeline #38758 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

@byshiue
Copy link
Copy Markdown
Collaborator Author

byshiue commented May 20, 2026

/bot run

@tensorrt-cicd
Copy link
Copy Markdown
Collaborator

PR_Github #49369 [ run ] triggered by Bot. Commit: 494000e Link to invocation

@jdebache
Copy link
Copy Markdown
Collaborator

Hey @byshiue, I think we might have some duplication here with: #13916 (comment). I think your changes might be a better fix in this case. Note that they also fix https://nvbugs/6163033 and https://nvbugs/6157892, which means you should also be able to remove the corresponding lines from the waive list.

Also, to verify the fix, we need to run with --extra-stage "GB200-4_GPUs-PyTorch-Post-Merge-1" I think.

@tensorrt-cicd
Copy link
Copy Markdown
Collaborator

PR_Github #49369 [ run ] completed with state SUCCESS. Commit: 494000e
/LLM/main/L0_MergeRequest_PR pipeline #39022 completed with status: 'SUCCESS'

CI Report

Link to invocation

@byshiue
Copy link
Copy Markdown
Collaborator Author

byshiue commented May 22, 2026

/bot run --extra-stage "GB200-4_GPUs-PyTorch-Post-Merge-1"

@tensorrt-cicd
Copy link
Copy Markdown
Collaborator

PR_Github #49803 [ run ] triggered by Bot. Commit: 494000e Link to invocation

@tensorrt-cicd
Copy link
Copy Markdown
Collaborator

PR_Github #49803 [ run ] completed with state SUCCESS. Commit: 494000e
/LLM/main/L0_MergeRequest_PR pipeline #39390 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: bhsueh <11360707+byshiue@users.noreply.github.com>
@byshiue
Copy link
Copy Markdown
Collaborator Author

byshiue commented May 22, 2026

/bot run --extra-stage "GB200-4_GPUs-PyTorch-Post-Merge-1"

@tensorrt-cicd
Copy link
Copy Markdown
Collaborator

PR_Github #49871 [ run ] triggered by Bot. Commit: fbc643d Link to invocation

@tensorrt-cicd
Copy link
Copy Markdown
Collaborator

PR_Github #49871 [ run ] completed with state FAILURE. Commit: fbc643d
/LLM/main/L0_MergeRequest_PR pipeline #39452 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: bhsueh <11360707+byshiue@users.noreply.github.com>
@byshiue
Copy link
Copy Markdown
Collaborator Author

byshiue commented May 22, 2026

/bot run

@tensorrt-cicd
Copy link
Copy Markdown
Collaborator

PR_Github #49895 [ run ] triggered by Bot. Commit: ed6f4ec Link to invocation

@tensorrt-cicd
Copy link
Copy Markdown
Collaborator

PR_Github #49895 [ run ] completed with state SUCCESS. Commit: ed6f4ec
/LLM/main/L0_MergeRequest_PR pipeline #39474 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

@jdebache
Copy link
Copy Markdown
Collaborator

Hmmm the two failures appear unrelated + the multi-gpu stage did not run. Do you know what happened?

@byshiue
Copy link
Copy Markdown
Collaborator Author

byshiue commented May 25, 2026

Hmmm the two failures appear unrelated + the multi-gpu stage did not run. Do you know what happened?

They are terminated unexpectedly. Let me rerun.

@byshiue
Copy link
Copy Markdown
Collaborator Author

byshiue commented May 25, 2026

/bot run

@tensorrt-cicd
Copy link
Copy Markdown
Collaborator

PR_Github #50116 [ run ] triggered by Bot. Commit: ed6f4ec Link to invocation

@tensorrt-cicd
Copy link
Copy Markdown
Collaborator

PR_Github #50116 [ run ] completed with state SUCCESS. Commit: ed6f4ec
/LLM/main/L0_MergeRequest_PR pipeline #39669 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

@byshiue
Copy link
Copy Markdown
Collaborator Author

byshiue commented May 25, 2026

/bot run

@tensorrt-cicd
Copy link
Copy Markdown
Collaborator

PR_Github #50160 [ run ] triggered by Bot. Commit: 6148ebd Link to invocation

@tensorrt-cicd
Copy link
Copy Markdown
Collaborator

PR_Github #50160 [ run ] completed with state SUCCESS. Commit: 6148ebd
/LLM/main/L0_MergeRequest_PR pipeline #39706 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

@byshiue
Copy link
Copy Markdown
Collaborator Author

byshiue commented May 26, 2026

/bot run

@tensorrt-cicd
Copy link
Copy Markdown
Collaborator

PR_Github #50245 [ run ] triggered by Bot. Commit: 6148ebd Link to invocation

@tensorrt-cicd
Copy link
Copy Markdown
Collaborator

PR_Github #50245 [ run ] completed with state SUCCESS. Commit: 6148ebd
/LLM/main/L0_MergeRequest_PR pipeline #39778 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

@byshiue
Copy link
Copy Markdown
Collaborator Author

byshiue commented May 26, 2026

/bot run

@tensorrt-cicd
Copy link
Copy Markdown
Collaborator

PR_Github #50297 [ run ] triggered by Bot. Commit: 6148ebd Link to invocation

@tensorrt-cicd
Copy link
Copy Markdown
Collaborator

PR_Github #50297 [ run ] completed with state FAILURE. Commit: 6148ebd
/LLM/main/L0_MergeRequest_PR pipeline #39828 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

@byshiue
Copy link
Copy Markdown
Collaborator Author

byshiue commented May 27, 2026

/bot run --disable-fail-fast

@tensorrt-cicd
Copy link
Copy Markdown
Collaborator

PR_Github #50419 [ run ] triggered by Bot. Commit: 6148ebd Link to invocation

@tensorrt-cicd
Copy link
Copy Markdown
Collaborator

PR_Github #50419 [ run ] completed with state SUCCESS. Commit: 6148ebd
/LLM/main/L0_MergeRequest_PR pipeline #39943 completed with status: 'SUCCESS'

CI Report

Link to invocation

Signed-off-by: bhsueh <11360707+byshiue@users.noreply.github.com>

# Conflicts:
#	tests/integration/test_lists/waives.txt
@byshiue
Copy link
Copy Markdown
Collaborator Author

byshiue commented May 27, 2026

/bot run

@tensorrt-cicd
Copy link
Copy Markdown
Collaborator

PR_Github #50498 [ run ] triggered by Bot. Commit: 756cfa5 Link to invocation

@tensorrt-cicd
Copy link
Copy Markdown
Collaborator

PR_Github #50498 [ run ] completed with state SUCCESS. Commit: 756cfa5
/LLM/main/L0_MergeRequest_PR pipeline #40007 completed with status: 'SUCCESS'
Pipeline passed with automatic retried tests. Check the rerun report for details.

CI Report

Link to invocation

@byshiue byshiue merged commit 83ec591 into main May 28, 2026
11 checks passed
@byshiue byshiue deleted the fix/nvbug_5800725 branch May 28, 2026 06:05
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.

3 participants