[None][feat] EPD for Qwen3 VL #10470
Conversation
2ez4bz
force-pushed
the
dev-qwen3-vl-epd
branch
from
60d965e to
de3a321
Compare
📝 WalkthroughWalkthroughThe pull request refactors head dimension handling from precomputed storage to on-demand computation via properties across weight mappers, introduces multimodal token processing methods to the Qwen3VL model, applies a multimodal support decorator, and extends test coverage to include the new Qwen3-VL-2B-Instruct model variant. Changes
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~22 minutes Pre-merge checks and finishing touches❌ Failed checks (1 warning)
✅ Passed checks (2 passed)
✨ Finishing touches
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 2
🤖 Fix all issues with AI agents
In @tensorrt_llm/_torch/models/checkpoints/hf/qwen3vl_weight_mapper.py:
- Around line 12-19: The _head_dim property currently returns head counts
(config.num_attention_heads or config.num_heads) but should return per-head
dimension; update the _head_dim property in qwen3vl_weight_mapper to return
config.head_dim for Qwen3VLTextConfig and compute config.hidden_size //
config.num_heads for Qwen3VLVisionConfig so downstream calculations (e.g.,
num_kv_heads = kv_shape // self._head_dim in the weight mapping logic) use the
actual head dimension rather than the head count; keep the TypeError for
unexpected config types.
In @tensorrt_llm/_torch/models/modeling_qwen3vl.py:
- Around line 387-389: The RuntimeError string in modeling_qwen3vl.py is missing
an f-string prefix so {expected_size} and {hidden_size} are not interpolated;
update the raise RuntimeError call (the line that currently reads "Expected
multimodal embedding to have hidden size {expected_size}, got {hidden_size}.")
to use an f-string (prefix with f) so the values of expected_size and
hidden_size are injected into the message.
📜 Review details
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (4)
tensorrt_llm/_torch/models/checkpoints/base_weight_mapper.pytensorrt_llm/_torch/models/checkpoints/hf/qwen3vl_weight_mapper.pytensorrt_llm/_torch/models/modeling_qwen3vl.pytests/unittest/_torch/multimodal/test_mm_encoder_standalone.py
🧰 Additional context used
📓 Path-based instructions (2)
**/*.py
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
**/*.py: The code developed for TensorRT-LLM should conform to Python 3.8+
Indent Python code with 4 spaces. Do not use tabs
Always maintain the namespace when importing Python modules, even if only one class or function from a module is used
Python filenames should use snake_case (e.g.,some_file.py)
Python classes should use PascalCase (e.g.,class SomeClass)
Python functions and methods should use snake_case (e.g.,def my_awesome_function():)
Python local variables should use snake_case, with prefixkfor variable names that start with a number (e.g.,k_99th_percentile)
Python global variables should use upper snake_case with prefixG(e.g.,G_MY_GLOBAL)
Python constants should use upper snake_case (e.g.,MY_CONSTANT)
Avoid shadowing variables declared in an outer scope in Python
Initialize all externally visible members of a Python class in the constructor
For Python interfaces that may be used outside a file, prefer docstrings over comments
Use comments in Python for code within a function, or interfaces that are local to a file
Use Google-style docstrings for Python classes and functions, which can be parsed by Sphinx
Python attributes and variables can be documented inline with the format"""<type>: Description"""
Avoid using reflection in Python when functionality can be easily achieved without reflection
When using try-except blocks in Python, limit the except clause to the smallest set of errors possible
When using try-except blocks in Python to handle multiple possible variable types (duck-typing), keep the body of the try as small as possible and use the else block for the main logic
Files:
tensorrt_llm/_torch/models/checkpoints/hf/qwen3vl_weight_mapper.pytensorrt_llm/_torch/models/modeling_qwen3vl.pytests/unittest/_torch/multimodal/test_mm_encoder_standalone.pytensorrt_llm/_torch/models/checkpoints/base_weight_mapper.py
**/*.{cpp,cc,cxx,h,hpp,hxx,cu,cuh,py}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
All TensorRT-LLM source files (.cpp, .h, .cu, .py, and other source files) should contain an NVIDIA copyright header with the year of latest meaningful modification
Files:
tensorrt_llm/_torch/models/checkpoints/hf/qwen3vl_weight_mapper.pytensorrt_llm/_torch/models/modeling_qwen3vl.pytests/unittest/_torch/multimodal/test_mm_encoder_standalone.pytensorrt_llm/_torch/models/checkpoints/base_weight_mapper.py
🧠 Learnings (4)
📚 Learning: 2025-09-29T15:14:28.503Z
Learnt from: amitz-nv
Repo: NVIDIA/TensorRT-LLM PR: 8063
File: tensorrt_llm/lora_manager.py:1080-1112
Timestamp: 2025-09-29T15:14:28.503Z
Learning: In tensorrt_llm/lora_manager.py, when calculating part_sizes for attn_qkv fused LoRA modules, the sizes are correctly multiplied by tp_size because model_config.num_heads and model_config.num_kv_heads are already divided by tp_size (per-TP-rank values), so multiplication is needed to get the original full concatenated dimension size. The interleave_fused_lora_weights_for_tp function provides proper validation with asserts for total size and TP divisibility.
Applied to files:
tensorrt_llm/_torch/models/checkpoints/hf/qwen3vl_weight_mapper.pytensorrt_llm/_torch/models/checkpoints/base_weight_mapper.py
📚 Learning: 2025-09-29T15:14:28.503Z
Learnt from: amitz-nv
Repo: NVIDIA/TensorRT-LLM PR: 8063
File: tensorrt_llm/lora_manager.py:1080-1112
Timestamp: 2025-09-29T15:14:28.503Z
Learning: In tensorrt_llm/lora_manager.py, when calculating part_sizes for attn_qkv fused LoRA modules, the sizes are correctly multiplied by tp_size because model_config.num_heads and model_config.num_kv_heads are already divided by tp_size (per-TP-rank values), so multiplication is needed to get the original full concatenated dimension size. The interleave_fused_lora_weights_for_tp function provides proper validation.
Applied to files:
tensorrt_llm/_torch/models/checkpoints/hf/qwen3vl_weight_mapper.pytensorrt_llm/_torch/models/checkpoints/base_weight_mapper.py
📚 Learning: 2025-07-22T09:22:14.726Z
Learnt from: yechank-nvidia
Repo: NVIDIA/TensorRT-LLM PR: 6254
File: tensorrt_llm/_torch/pyexecutor/model_engine.py:1201-1204
Timestamp: 2025-07-22T09:22:14.726Z
Learning: In TensorRT-LLM's multimodal processing pipeline, shared tensor recovery using `from_shared_tensor()` is only needed during the context phase. Generation requests reuse the already-recovered tensor data and only need to call `strip_for_generation()` to remove unnecessary multimodal data while preserving the recovered tensors. This avoids redundant tensor recovery operations during generation.
Applied to files:
tensorrt_llm/_torch/models/modeling_qwen3vl.py
📚 Learning: 2025-07-28T17:06:08.621Z
Learnt from: moraxu
Repo: NVIDIA/TensorRT-LLM PR: 6303
File: tests/integration/test_lists/qa/examples_test_list.txt:494-494
Timestamp: 2025-07-28T17:06:08.621Z
Learning: In TensorRT-LLM testing, it's common to have both CLI flow tests (test_cli_flow.py) and PyTorch API tests (test_llm_api_pytorch.py) for the same model. These serve different purposes: CLI flow tests validate the traditional command-line workflow, while PyTorch API tests validate the newer LLM API backend. Both are legitimate and should coexist.
Applied to files:
tests/unittest/_torch/multimodal/test_mm_encoder_standalone.py
🧬 Code graph analysis (2)
tensorrt_llm/_torch/models/checkpoints/hf/qwen3vl_weight_mapper.py (1)
tensorrt_llm/_torch/models/checkpoints/base_weight_mapper.py (3)
_head_dim(175-180)config(163-166)model(169-172)
tensorrt_llm/_torch/models/checkpoints/base_weight_mapper.py (3)
tensorrt_llm/_torch/models/checkpoints/hf/qwen3vl_weight_mapper.py (1)
_head_dim(13-19)tensorrt_llm/_torch/models/modeling_qwen3vl.py (1)
config(88-89)tensorrt_llm/_torch/models/modeling_utils.py (1)
config(526-527)
🪛 Ruff (0.14.10)
tensorrt_llm/_torch/models/checkpoints/hf/qwen3vl_weight_mapper.py
19-19: Avoid specifying long messages outside the exception class
(TRY003)
tensorrt_llm/_torch/models/modeling_qwen3vl.py
374-374: Avoid specifying long messages outside the exception class
(TRY003)
377-377: Avoid specifying long messages outside the exception class
(TRY003)
387-389: Avoid specifying long messages outside the exception class
(TRY003)
⏰ 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
🔇 Additional comments (7)
tensorrt_llm/_torch/models/checkpoints/base_weight_mapper.py (1)
174-180: LGTM! Clean on-demand computation of head dimension.The property correctly handles the fallback logic when
head_dimis not explicitly defined in the config. This approach is more flexible than precomputing during initialization, allowing subclasses likeQwen3VLHfWeightMapperto override with config-specific logic.tests/unittest/_torch/multimodal/test_mm_encoder_standalone.py (2)
24-29: LGTM! Good test coverage extension for Qwen3 VL.The new model variant is properly integrated into both test functions with appropriate parameterization.
185-186: LGTM! Appropriate batch size selection.Using
encoder_max_batch_size=3for Qwen3 VL (matching LLAVA's configuration) is appropriate, as the Qwen2.5 VL limitation noted in the comment appears to be model-specific.tensorrt_llm/_torch/models/modeling_qwen3vl.py (4)
354-430: Overall logic is sound for the EPD token expansion.The method correctly handles single-image cases with proper validation. The TODO at line 393 appropriately flags that video token support is deferred.
Consider extracting the validation block (lines 372-389) into a separate helper for cleaner separation of concerns when extending to multi-image/video support.
1036-1048: LGTM! Clean refactoring of multimodal data filtering.Extracting the filtering logic into
_get_requests_with_mm_dataimproves readability and allows for consistent MM data detection across the class.
1085-1099: LGTM! Robust filtering logic for multimodal data detection.The method correctly handles all three cases: image data, video data, and pre-populated embeddings for disaggregated inference. The nested
.get()pattern with empty dict defaults is a safe approach.
1102-1103: Decorator placement and naming appear correct for multimodal disaggregated support.The
@support_multimodal_disaggregateddecorator is properly applied before the class definition. However, I cannot verify the decorator's actual implementation, the attribute it sets, or its interaction with the forward method logic without access to the repository.
tensorrt_llm/_torch/models/checkpoints/hf/qwen3vl_weight_mapper.py
Outdated
Show resolved
Hide resolved
2ez4bz
force-pushed
the
dev-qwen3-vl-epd
branch
from
de3a321 to
bdb9b04
Compare
yechank-nvidia
added
the
Multimodal
yechank-nvidia
left a comment
yechank-nvidia
left a comment
There was a problem hiding this comment.
Moving @jaedeok-nvidia's commnets from #10435's EPD part.
|
/bot run |
|
PR_Github #30845 [ run ] triggered by Bot. Commit: |
* Why? We would like to support EPD disaggregated serving for Qwen3 VL. * What? This commit adds such support, and extends existing unit tests for correctness checks. Some minor (protected) interface changes had to be made to the weight mapper as a side-effect. Signed-off-by: William Zhang <[email protected]>
2ez4bz
force-pushed
the
dev-qwen3-vl-epd
branch
from
bdb9b04 to
76db627
Compare
|
/bot kill |
|
/bot run |
|
PR_Github #30848 [ run ] triggered by Bot. Commit: |
|
PR_Github #30845 [ run ] completed with state |
|
PR_Github #30849 [ kill ] triggered by Bot. Commit: |
|
PR_Github #30848 [ run ] completed with state |
|
PR_Github #30849 [ kill ] completed with state |
|
/bot run |
|
PR_Github #30850 [ run ] triggered by Bot. Commit: |
|
PR_Github #30850 [ run ] completed with state
|
|
/bot run |
|
PR_Github #30920 [ run ] triggered by Bot. Commit: |
jaedeok-nvidia
left a comment
jaedeok-nvidia
left a comment
There was a problem hiding this comment.
Thanks for your work! Although there is one question, overall it looks good to me.
|
PR_Github #30920 [ run ] completed with state
|
|
/bot run |
|
PR_Github #31004 [ run ] triggered by Bot. Commit: |
|
PR_Github #31004 [ run ] completed with state |
4 participants
Summary by CodeRabbit
New Features
Refactor
Tests
✏️ Tip: You can customize this high-level summary in your review settings.
Description
We would like to support EPD disaggregated serving for Qwen3 VL.
This commit adds such support, and extends existing unit tests for
correctness checks.
Some minor (protected) interface changes had to be made to the
weight mapper as a side-effect.
Test Coverage
test_mm_encoder_standalone.pywith a Qwen3 VL model.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)
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
/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.
Details
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: Does NOT update GitHub check status.--stage-list "A10-PyTorch-1, xxx"(OPTIONAL) : Only run the specified test stages. Examples: "A10-PyTorch-1, xxx". Note: Does NOT 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: Does NOT 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: Does NOT update GitHub pipeline status.--only-multi-gpu-test(OPTIONAL) : Only run the multi-GPU tests. Note: Does NOT update GitHub check status.--disable-multi-gpu-test(OPTIONAL) : Disable the multi-GPU tests. Note: Does NOT 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-listparameter to access the appropriate container environment. Note: Does NOT update GitHub check status.For guidance on mapping tests to stage names, see
docs/source/reference/ci-overview.mdand the
scripts/test_to_stage_mapping.pyhelper.kill
killKill all running builds associated with pull request.
skip
skip --comment COMMENTSkip 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-pipelineReuse 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.