[TRTLLM-7758][feat] Optimize phi4-mm image modality inference #7918
Conversation
Wanli-Jiang
requested review from
2ez4bz,
amukkara,
chang-l and
yechank-nvidia
|
/bot run --disable-fail-fast |
📝 WalkthroughWalkthroughIntroduces runtime torch.compile toggle, new optimized paths for SigLIP image embedding and vision encoder, dynamic and full image preprocessing utilities, public NoOp layer and InputMode enum, bindings/injections into encoder and processor, batch-oriented encoding in forward, and weight-loading skip for removed vision head, all within modeling_phi4mm.py. Changes
Sequence Diagram(s)sequenceDiagram
autonumber
participant App as Caller
participant LM as Phi4MMForCausalLM.forward
participant Enc as HFPhi4MultimodalEncoder.forward
participant Proc as ImageProcessor (optimized_*)
participant Vis as Vision Encoder (optimized)
participant Txt as Text Embeddings/LM Head
App->>LM: input_ids, images, audios, masks
note right of LM: @torch.compile (runtime toggle)
LM->>Enc: batch inputs (ids, image tensors/masks, audio)
Enc->>Proc: optimized_preprocess / dynamic_preprocess
Proc-->>Enc: input_image_embeds, image_attention_mask, sizes
Enc->>Vis: optimized_vision_encoder_forward(...)
Vis-->>Enc: fused image features (SigLIP path)
Enc->>Txt: integrate image/audio features with text
Txt-->>LM: logits
LM-->>App: logits/output
sequenceDiagram
autonumber
participant Loader as load_weights
participant Store as State Dict
participant Model as Phi4MMForCausalLM
note over Loader,Model: Skip replaced vision head weights
Loader->>Store: iterate keys
alt key startswith "model.embed_tokens_extend.image_embed.img_processor.head."
Loader-->>Model: skip load
else
Loader->>Model: load weight
end
Estimated code review effort🎯 4 (Complex) | ⏱️ ~60 minutes Pre-merge checks and finishing touches❌ Failed checks (2 warnings)
✅ Passed checks (1 passed)
✨ Finishing touches
🧪 Generate unit tests
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: 5
🧹 Nitpick comments (6)
tensorrt_llm/_torch/models/modeling_phi4mm.py (6)
111-116: Avoid hyphens in module names when using importlib.Hyphens in
spec_from_file_locationmodule name can cause oddities. Use an identifier.- spec = importlib.util.spec_from_file_location( - "Phi-4-multimodal-instruct.hf_modeling_phi4mm", - modeling_phi4mm_path) + spec = importlib.util.spec_from_file_location( + "phi4mm_hf_modeling", + modeling_phi4mm_path)
491-498: Nit: avoid list concatenation inside torch.cat.- torch.cat([_global_image] + [_im], dim=0) + torch.cat([_global_image, _im], dim=0)- torch.cat([_global_mask] + [_mask], + torch.cat([_global_mask, _mask], dim=0) ...
528-529: Silence unused args in NoOp.forward.- def forward(self, *args, **kwargs): + def forward(self, *_args, **_kwargs): return None
615-623: Use boolean dtype and correct device for image attention masks.Allocating masks with embed dtype is misleading and wastes memory.
- batched_image_attn_mask = torch.zeros( - (total_b, max_p, h_i_attn, w_i_attn), - dtype=input_image_embeds_list[0].dtype, - device=input_image_embeds_list[0].device) + batched_image_attn_mask = torch.zeros( + (total_b, max_p, h_i_attn, w_i_attn), + dtype=torch.bool, + device=input_image_embeds_list[0].device) ... - else: - batched_image_attn_mask[b_offset:b_offset + b, :p] = 1 + else: + batched_image_attn_mask[b_offset:b_offset + b, :p] = TrueAlso applies to: 627-633
671-675: Same as above for audio attention masks.- batched_audio_attn_mask = torch.zeros( - (total_b, max_p), - dtype=input_audio_embeds_list[0].dtype, - device=input_audio_embeds_list[0].device) + batched_audio_attn_mask = torch.zeros( + (total_b, max_p), + dtype=torch.bool, + device=input_audio_embeds_list[0].device) ... - else: - batched_audio_attn_mask[b_offset:b_offset + b, :p] = 1 + else: + batched_audio_attn_mask[b_offset:b_offset + b, :p] = TrueAlso applies to: 679-685
187-237: Minor: avoid CPU sync via Python int conversions in hot path.
int(tensor)forces D2H sync. If feasible, keep in tensor form until the final reshape/index step.Consider computing
useful_height/widthwith tensor ops and slicing via tensor indices where possible.
📜 Review details
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (1)
tensorrt_llm/_torch/models/modeling_phi4mm.py(15 hunks)
🧰 Additional context used
📓 Path-based instructions (3)
**/*.{h,hpp,hh,hxx,cpp,cxx,cc,cu,cuh,py}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
Use only spaces, no tabs; indent with 4 spaces.
Files:
tensorrt_llm/_torch/models/modeling_phi4mm.py
**/*.py
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
**/*.py: Python code must target Python 3.8+.
Indent Python code with 4 spaces; do not use tabs.
Maintain module namespace when importing; prefer 'from package.subpackage import foo' then 'foo.SomeClass()' instead of importing the class directly.
Python filenames should be snake_case (e.g., some_file.py).
Python classes use PascalCase names.
Functions and methods use snake_case names.
Local variables use snake_case; prefix 'k' for variables that start with a number (e.g., k_99th_percentile).
Global variables use upper SNAKE_CASE prefixed with 'G' (e.g., G_MY_GLOBAL).
Constants use upper SNAKE_CASE (e.g., MY_CONSTANT).
Avoid shadowing variables from an outer scope.
Initialize all externally visible members of a class in the constructor.
Prefer docstrings for interfaces that may be used outside a file; comments for in-function or file-local interfaces.
Use Google-style docstrings for classes and functions (Sphinx-parsable).
Document attributes and variables inline so they render under the class/function docstring.
Avoid reflection when a simpler, explicit approach suffices (e.g., avoid dict(**locals()) patterns).
In try/except, catch the most specific exceptions possible.
For duck-typing try/except, keep the try body minimal and use else for the main logic.
Files:
tensorrt_llm/_torch/models/modeling_phi4mm.py
**/*.{cpp,cxx,cc,h,hpp,hh,hxx,cu,cuh,py}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
Prepend the NVIDIA Apache-2.0 copyright header with current year to the top of all source files (e.g., .cpp, .h, .cu, .py).
Files:
tensorrt_llm/_torch/models/modeling_phi4mm.py
🧬 Code graph analysis (1)
tensorrt_llm/_torch/models/modeling_phi4mm.py (4)
tensorrt_llm/_utils.py (1)
nvtx_range(857-876)tensorrt_llm/_torch/models/modeling_multimodal_utils.py (2)
find_closest_aspect_ratio(367-381)dynamic_preprocess(384-435)tensorrt_llm/inputs/multimodal.py (1)
MultimodalParams(196-520)tensorrt_llm/_torch/models/modeling_qwen2vl.py (2)
multimodal_data_device_paths(978-983)multimodal_data_device_paths(1030-1041)
🪛 Ruff (0.13.1)
tensorrt_llm/_torch/models/modeling_phi4mm.py
375-376: Avoid specifying long messages outside the exception class
(TRY003)
440-442: Avoid specifying long messages outside the exception class
(TRY003)
491-491: Consider [_global_image, _im] instead of concatenation
Replace with [_global_image, _im]
(RUF005)
495-495: Consider [_global_mask, _mask] instead of concatenation
Replace with [_global_mask, _mask]
(RUF005)
528-528: Unused method argument: args
(ARG002)
528-528: Unused method argument: kwargs
(ARG002)
⏰ 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 (2)
tensorrt_llm/_torch/models/modeling_phi4mm.py (2)
743-747: Confirm torch.compile decorator order with inference_mode.
@torch.compilewrapping@torch.inference_mode()can affect graph capture. Verify perf/correctness for variable crop counts.Would you like me to benchmark both orders and report compile success rates?
955-960: Head replaced with NoOp — confirm no call sites expect a Tensor.Repo search for "img_processor.head(" returned no matches; if any callers invoke img_processor.head(...) returning None will crash them — ensure no callers exist or change the NoOp to return a Tensor (e.g., an identity tensor). File: tensorrt_llm/_torch/models/modeling_phi4mm.py lines 955–960.
|
PR_Github #19651 [ run ] triggered by Bot |
|
PR_Github #19651 [ run ] completed with state |
Wanli-Jiang
force-pushed
the
user/williamj/phi4mm-opt
branch
from
2bb6a97 to
a88d73d
Compare
Wanli-Jiang
force-pushed
the
user/williamj/phi4mm-opt
branch
from
a88d73d to
6e6f246
Compare
|
@Wanli-Jiang Can you add a summary of perf improvement in PR description? Just the TLLM speedup compared to previous implementation. |
Signed-off-by: Wanli Jiang <[email protected]>
Wanli-Jiang
force-pushed
the
user/williamj/phi4mm-opt
branch
from
6e6f246 to
bf956f1
Compare
|
/bot run |
|
PR_Github #19859 [ run ] triggered by Bot |
|
PR_Github #19859 [ run ] completed with state |
|
/bot run |
|
PR_Github #19898 [ run ] triggered by Bot |
|
PR_Github #19898 [ run ] completed with state |
5 participants
(img=512x512, ISL=256, concurrency=16)
Summary by CodeRabbit
New Features
Performance
Chores
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)
Any new dependencies have been scanned for license and vulnerabilities
CODEOWNERS updated if ownership changes
Documentation updated as needed
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.
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.