Refactor
- Multimodal images are now padded to the maximum height/width across the batch for consistent processing.
- A batching helper was made public and the image-processing path updated to use it; returns batched pixel data and corresponding image sizes.
Tests
- Added parameterized tests validating batched shapes and padding preservation across heterogeneous and uniform image-size scenarios.

Description

Prior to this commit, if multiple requests with images were in the same batch, the batching logic for the images would fail.

This commit fixes it, and adds unit tests for it that were verified to fail prior to the fix.

Test Coverage

Added a new unit tests intests/unittest/_torch/modeling/test_modeling_mistral.py.

GitHub Bot Help

/bot [-h] ['run', 'kill', 'skip', 'reuse-pipeline'] ...

Provide a user friendly way for developers to interact with a Jenkins server.

Run/bot [-h|--help] to print this help message.

See details below for each supported subcommand.

run [--reuse-test (optional)pipeline-id --disable-fail-fast --skip-test --stage-list "A10-PyTorch-1, xxx" --gpu-type "A30, H100_PCIe" --test-backend "pytorch, cpp" --add-multi-gpu-test --only-multi-gpu-test --disable-multi-gpu-test --post-merge --extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx" --detailed-log --debug(experimental)]

Launch build/test pipelines. All previously running jobs will be killed.

--reuse-test (optional)pipeline-id(OPTIONAL) : Allow the new pipeline to reuse build artifacts and skip successful test stages from a specified pipeline or the last pipeline if no pipeline-id is indicated. If the Git commit ID has changed, this option will be always ignored. The DEFAULT behavior of the bot is to reuse build artifacts and successful test results from the last pipeline.

--disable-reuse-test(OPTIONAL) : Explicitly prevent the pipeline from reusing build artifacts and skipping successful test stages from a previous pipeline. Ensure that all builds and tests are run regardless of previous successes.

--disable-fail-fast(OPTIONAL) : Disable fail fast on build/tests/infra failures.

--skip-test(OPTIONAL) : Skip all test stages, but still run build stages, package stages and sanity check stages. Note: DoesNOT update GitHub check status.

--stage-list "A10-PyTorch-1, xxx"(OPTIONAL) : Only run the specified test stages. Examples: "A10-PyTorch-1, xxx". Note: DoesNOT update GitHub check status.

--gpu-type "A30, H100_PCIe"(OPTIONAL) : Only run the test stages on the specified GPU types. Examples: "A30, H100_PCIe". Note: DoesNOT update GitHub check status.

--test-backend "pytorch, cpp"(OPTIONAL) : Skip test stages which don't match the specified backends. Only support [pytorch, cpp, tensorrt, triton]. Examples: "pytorch, cpp" (does not run test stages with tensorrt or triton backend). Note: DoesNOT update GitHub pipeline status.

--only-multi-gpu-test(OPTIONAL) : Only run the multi-GPU tests. Note: DoesNOT update GitHub check status.

--disable-multi-gpu-test(OPTIONAL) : Disable the multi-GPU tests. Note: DoesNOT update GitHub check status.

--add-multi-gpu-test(OPTIONAL) : Force run the multi-GPU tests in addition to running L0 pre-merge pipeline.

--post-merge(OPTIONAL) : Run the L0 post-merge pipeline instead of the ordinary L0 pre-merge pipeline.

--extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx"(OPTIONAL) : Run the ordinary L0 pre-merge pipeline and specified test stages. Examples: --extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx".

--detailed-log(OPTIONAL) : Enable flushing out all logs to the Jenkins console. This will significantly increase the log volume and may slow down the job.

--debug(OPTIONAL) :Experimental feature. Enable access to the CI container for debugging purpose. Note: Specify exactly one stage in thestage-list parameter to access the appropriate container environment. Note: DoesNOT update GitHub check status.

For guidance on mapping tests to stage names, seedocs/source/reference/ci-overview.md
and thescripts/test_to_stage_mapping.py helper.

kill

kill

Kill all running builds associated with pull request.

skip

skip --comment COMMENT

Skip testing for latest commit on pull request.--comment "Reason for skipping build/test" is required. IMPORTANT NOTE: This is dangerous since lack of user care and validation can cause top of tree to break.

reuse-pipeline

reuse-pipeline

Reuse a previous pipeline to validate current commit. This action will also kill all currently running builds associated with the pull request. IMPORTANT NOTE: This is dangerous since lack of user care and validation can cause top of tree to break.

2ez4bz requested review froma team ascode owners

August 12, 2025 22:28

2ez4bz requested review fromhypdeb andyechank-nvidia

August 12, 2025 22:28

2ez4bz force-pushed thedev-mistral3-batching-fix branch froma46a745 to42e00e7Compare

August 12, 2025 22:31

Copy link

CollaboratorAuthor

2ez4bz commentedAug 12, 2025

/bot run

Copy link

Collaborator

tensorrt-cicd commentedAug 12, 2025

PR_Github #15022 [ run ] triggered by Bot

coderabbitaibot reviewed

Aug 12, 2025

View reviewed changes

Copy link

Contributor

coderabbitaibot left a comment

Choose a reason for hiding this comment

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

Actionable comments posted: 1

🔭 Outside diff range comments (3)

tensorrt_llm/_torch/models/modeling_mistral.py (2)
1-1:Missing NVIDIA copyright header (required by repo guidelines).
Please prepend the current-year NVIDIA copyright header to this source file.
Apply this diff at the top of the file:
+# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.
374-379:Fix Tensor→int conversion in Mistral3PatchMerger.forward
We verified that batching all image features into onemm_embeds list entry works withfuse_input_embeds (it simply concatenates and masks correctly), butMistral3PatchMerger.forward currently buildstokens_per_image from scalar Tensors—causing a runtime error inTensor.split.
Please updateforward intensorrt_llm/_torch/models/modeling_mistral.py (around line 496) to convertimage_sizes to Python ints before computing splits. For example:
     @torch.inference_mode()     def forward(self, image_features: torch.Tensor,-                image_sizes: torch.Tensor) -> torch.Tensor:-        image_sizes = [(image_size[0] // self._patch_size,-                        image_size[1] // self._patch_size)-                       for image_size in image_sizes]+                image_sizes: torch.Tensor) -> torch.Tensor:+        # Ensure we work with Python ints, not 0-d Tensors+        if isinstance(image_sizes, torch.Tensor):+            # [[H1, W1], [H2, W2], …]+            image_sizes = image_sizes.tolist()+        # Divide by patch size using Python ints+        image_sizes = [+            (h // self._patch_size, w // self._patch_size)+            for h, w in image_sizes+        ]          tokens_per_image = [h * w for h, w in image_sizes]         d = image_features.shape[-1]
This change prevents passing a list of scalar Tensors intosplit, which expects integers.
tests/unittest/_torch/modeling/test_modeling_mistral.py (1)
1-1:Missing NVIDIA copyright header (required by repo guidelines).
Please prepend the current-year NVIDIA copyright header to this test file.
Apply this diff at the top of the file:
+# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.

🧹 Nitpick comments (2)

tensorrt_llm/_torch/models/modeling_mistral.py (1)
439-470:Add a short docstring for the new public helper.
This has become part of the public API; document input shapes, device/dtype expectations, and return shapes. Keeps usage stable for downstreams and tests.
Suggested docstring:
@staticmethod@torch.inference_mode()defbatch_pixel_values(pixel_values:List[torch.Tensor],image_sizes:List[torch.Tensor],)->Tuple[torch.Tensor,torch.Tensor]:"""Batch and pad image tensors to a per-batch max size.    Args:        pixel_values: List of tensors with shapes [B_i, C, H_i, W_i]. Each tensor            corresponds to one request in the batch and may have different H/W.        image_sizes: List of tensors with shapes [B_i, 2] containing original (H, W)            prior to any processor padding.    Returns:        A tuple:          - batched_pixels: Tensor [sum(B_i), C, H_max, W_max] with all images            padded to (H_max, W_max) across the batch.          - batched_image_sizes: Tensor [sum(B_i), 2] of original per-image sizes.    """
tests/unittest/_torch/modeling/test_modeling_mistral.py (1)
443-496:LGTM on coverage; add dtype specificity for image_sizes to mirror model expectations.
Great parameterized test. One suggestion: enforce integer dtype forimage_sizes to match the model path’s expectations and avoid accidental float tensors if test data changes.
Apply this diff:
-    image_sizes = [torch.tensor(size) for size in image_sizes]+    image_sizes = [torch.tensor(size, dtype=torch.long) for size in image_sizes]

📜 Review details

Configuration used: .coderabbit.yaml
Review profile: CHILL
Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between3d169bf and42e00e7.

📒 Files selected for processing (2)

tensorrt_llm/_torch/models/modeling_mistral.py (2 hunks)
tests/unittest/_torch/modeling/test_modeling_mistral.py (1 hunks)

🧰 Additional context used

📓 Path-based instructions (2)

**/*.py

📄 CodeRabbit Inference Engine (CODING_GUIDELINES.md)

**/*.py: Python code must target Python 3.8+
Python indentation: 4 spaces, no tabs
Maintain module namespace in imports (from package.subpackage import foo; then use foo.SomeClass())
Python file names use snake_case
Python class names use PascalCase
Python functions/methods and local variables use snake_case; variables starting with a number get k_ prefix (e.g., k_99th_percentile)
Global variables use G_ prefixed UPPER_SNAKE_CASE (e.g., G_MY_GLOBAL)
Constants use UPPER_SNAKE_CASE in Python
Avoid shadowing variables from outer scopes in Python
Initialize all externally visible members of a Python class ininit
Prefer docstrings for interfaces used outside a file; comments for local code
Use Google-style docstrings for classes and functions (Sphinx-parsable)
Document attributes/variables inline with short docstrings
Avoid reflection when simple alternatives exist (e.g., prefer explicit parameters over dict(**locals()))
In try/except, catch the narrowest exceptions possible
For duck-typing with try/except, keep try body minimal and put logic in else

Files:

tests/unittest/_torch/modeling/test_modeling_mistral.py
tensorrt_llm/_torch/models/modeling_mistral.py

**/*.{cpp,cxx,cc,cu,h,hpp,hxx,hh,cuh,py}

📄 CodeRabbit Inference Engine (CODING_GUIDELINES.md)

Prepend NVIDIA copyright header (current year) to all source files

Files:

tests/unittest/_torch/modeling/test_modeling_mistral.py
tensorrt_llm/_torch/models/modeling_mistral.py

⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (1)

GitHub Check: Pre-commit Check

tensorrt_llm/_torch/models/modeling_mistral.pyShow resolvedHide resolved

2ez4bz force-pushed thedev-mistral3-batching-fix branch from42e00e7 to00b7718Compare

August 12, 2025 22:48

[None][fix] Fix batching bug in Mistral3 model

fa9e19a

Prior to this commit, if multiple requests with images were in the samebatch, the batching logic for the images would fail.This commit fixes it, and adds unit tests for it that were verified tofail prior to the fix.Signed-off-by: William Zhang <133824995+2ez4bz@users.noreply.github.com>

2ez4bz force-pushed thedev-mistral3-batching-fix branch from00b7718 tofa9e19aCompare

PR_Github #15027 [ run ] triggered by Bot

Copy link

Collaborator

tensorrt-cicd commentedAug 13, 2025

PR_Github #15027 [ run ] completed with stateSUCCESS
/LLM/release-1.0/L0_MergeRequest_PR pipeline #71 completed with status: 'FAILURE'

yechank-nvidia approved these changes

Aug 13, 2025

View reviewed changes

Copy link

Collaborator

yechank-nvidia left a comment

Choose a reason for hiding this comment

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

Thx for the fix.

tensorrt_llm/_torch/models/modeling_mistral.pyShow resolvedHide resolved

NVIDIA deleted a comment fromcoderabbitaibot

Aug 13, 2025

Copy link

CollaboratorAuthor

2ez4bz commentedAug 13, 2025

/bot run

2ez4bzenabled auto-merge (squash)

August 13, 2025 04:57

Copy link

Collaborator

tensorrt-cicd commentedAug 13, 2025

PR_Github #15073 [ run ] triggered by Bot

Copy link

Collaborator

tensorrt-cicd commentedAug 13, 2025

PR_Github #15073 [ run ] completed with stateFAILURE
/LLM/release-1.0/L0_MergeRequest_PR pipeline #80 completed with status: 'FAILURE'

hypdeb approved these changes

Aug 13, 2025

View reviewed changes

Copy link

Collaborator

hypdeb left a comment

Choose a reason for hiding this comment

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

dominicshanshan pushed a commit to dominicshanshan/TensorRT-LLM that referenced this pull request

Aug 22, 2025