[None][feat] AutoDeploy: compiler backends based on nn.Module #8126
Conversation
Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>
|
/bot run --disable-fail-fast |
📝 WalkthroughWalkthroughRefactors the compile subsystem: renames and simplifies the compiler base and registry, removes compile_and_capture, and shifts callers to instantiate backends via CompileBackendRegistry and call compile(). Updates multiple backends to the new API. Reworks CUDA graph capture interfaces and input flattening. Adjusts transform/shim dynamic-shape APIs and adds per-GraphModule transform execution. Updates tests accordingly. Changes
Sequence Diagram(s)sequenceDiagram
autonumber
actor Caller
participant Registry as CompileBackendRegistry
participant Backend as CompilerBackend (subclass)
participant Model as nn.Module
Caller->>Registry: get("<backend>")
Registry-->>Caller: Compiler class
Caller->>Backend: instantiate(model, args, kwargs, **compiler_kwargs)
Backend->>Backend: compile()
note over Backend: Backend-specific compile logic<br/>(e.g., torch.compile, CUDA graph setup)
Backend-->>Caller: Compiled Model
Caller->>Model: use compiled model
sequenceDiagram
autonumber
actor Caller
participant TC as TorchCudagraphCompiler
participant CG as CapturedGraph
participant CUDA as CUDA Graphs
Caller->>TC: instantiate(model, cuda_graph_batch_sizes, num_batched_inputs, max_batch_size)
TC->>TC: compile()
TC->>CG: create CapturedGraph(model, batch_sizes, num_batched_inputs)
CG->>CG: first forward: flatten args, record in_spec
CG->>CUDA: capture per-batch graphs
CG-->>TC: compiled module (wrapper)
Caller->>CG: forward(args, kwargs)
CG->>CG: flatten via recorded in_spec
CG->>CUDA: replay cudagraph matching batch
CG-->>Caller: outputs (unflattened via out_spec)
Estimated code review effort🎯 4 (Complex) | ⏱️ ~75 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: 4
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
tensorrt_llm/_torch/auto_deploy/utils/node_utils.py (1)
1-2: Add the required NVIDIA Apache-2.0 copyright header.The coding guidelines require prepending the NVIDIA Apache-2.0 copyright header with the current year (2025) to all source files, including
.pyfiles.Apply this diff to add the copyright header:
+# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + """Common utils for torch fx graph transformation."""As per coding guidelines.
🧹 Nitpick comments (4)
tensorrt_llm/_torch/auto_deploy/utils/node_utils.py (1)
321-327: LGTM! Output node validation is now more precise.The change correctly tightens the logic by only assigning
output_nodewhen encountering a node withop == "output", and the assertion now directly checks for its presence. This improves clarity and correctness.Optionally, the loop could break early once both
input_id_nodeandoutput_nodeare found to improve efficiency:for node in gm.graph.nodes: if input_id_node is None and node.op == "placeholder": input_id_node = node if node.op == "output": output_node = node + if input_id_node is not None and output_node is not None: + breaktests/unittest/_torch/auto_deploy/unit/singlegpu/compile/test_captured_graph.py (1)
92-94: Drop debug prints from the testThese
tensorrt_llm/_torch/auto_deploy/compile/compiler.py (2)
12-18: UseClassVarfor the shared backend registryAnnotate
_backend_registryas aClassVarso static analyzers (RUF012) stop flagging it and to make the class-level intent explicit. Based on static analysis hints.
35-45: Preserve any extra compiler kwargs
compiler_kwargsare currently swallowed by the base class, so derived compilers (or future extensions) cannot inspect them. Either persist them (e.g.,self.compiler_kwargs = compiler_kwargs) or drop the parameter if it is truly unused; otherwise options passed by callers get lost silently.
📜 Review details
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (16)
examples/auto_deploy/build_and_run_flux.py(2 hunks)tensorrt_llm/_torch/auto_deploy/compile/backends/torch_compile.py(1 hunks)tensorrt_llm/_torch/auto_deploy/compile/backends/torch_cudagraph.py(8 hunks)tensorrt_llm/_torch/auto_deploy/compile/backends/torch_opt.py(2 hunks)tensorrt_llm/_torch/auto_deploy/compile/backends/torch_simple.py(1 hunks)tensorrt_llm/_torch/auto_deploy/compile/compiler.py(2 hunks)tensorrt_llm/_torch/auto_deploy/custom_ops/attention_interface.py(1 hunks)tensorrt_llm/_torch/auto_deploy/shim/interface.py(2 hunks)tensorrt_llm/_torch/auto_deploy/transform/interface.py(8 hunks)tensorrt_llm/_torch/auto_deploy/transform/library/cleanup_input_constraints.py(1 hunks)tensorrt_llm/_torch/auto_deploy/transform/library/compile_model.py(2 hunks)tensorrt_llm/_torch/auto_deploy/transformations/_graph.py(7 hunks)tensorrt_llm/_torch/auto_deploy/utils/node_utils.py(1 hunks)tests/unittest/_torch/auto_deploy/unit/singlegpu/compile/test_captured_graph.py(3 hunks)tests/unittest/_torch/auto_deploy/unit/singlegpu/compile/test_compiler.py(2 hunks)tests/unittest/_torch/auto_deploy/unit/singlegpu/compile/test_cuda_graph_batch_sizes.py(7 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:
tests/unittest/_torch/auto_deploy/unit/singlegpu/compile/test_compiler.pyexamples/auto_deploy/build_and_run_flux.pytests/unittest/_torch/auto_deploy/unit/singlegpu/compile/test_captured_graph.pytensorrt_llm/_torch/auto_deploy/transform/library/cleanup_input_constraints.pytensorrt_llm/_torch/auto_deploy/compile/compiler.pytensorrt_llm/_torch/auto_deploy/shim/interface.pytensorrt_llm/_torch/auto_deploy/custom_ops/attention_interface.pytensorrt_llm/_torch/auto_deploy/transformations/_graph.pytensorrt_llm/_torch/auto_deploy/utils/node_utils.pytensorrt_llm/_torch/auto_deploy/transform/library/compile_model.pytensorrt_llm/_torch/auto_deploy/compile/backends/torch_cudagraph.pytensorrt_llm/_torch/auto_deploy/compile/backends/torch_simple.pytests/unittest/_torch/auto_deploy/unit/singlegpu/compile/test_cuda_graph_batch_sizes.pytensorrt_llm/_torch/auto_deploy/transform/interface.pytensorrt_llm/_torch/auto_deploy/compile/backends/torch_opt.pytensorrt_llm/_torch/auto_deploy/compile/backends/torch_compile.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:
tests/unittest/_torch/auto_deploy/unit/singlegpu/compile/test_compiler.pyexamples/auto_deploy/build_and_run_flux.pytests/unittest/_torch/auto_deploy/unit/singlegpu/compile/test_captured_graph.pytensorrt_llm/_torch/auto_deploy/transform/library/cleanup_input_constraints.pytensorrt_llm/_torch/auto_deploy/compile/compiler.pytensorrt_llm/_torch/auto_deploy/shim/interface.pytensorrt_llm/_torch/auto_deploy/custom_ops/attention_interface.pytensorrt_llm/_torch/auto_deploy/transformations/_graph.pytensorrt_llm/_torch/auto_deploy/utils/node_utils.pytensorrt_llm/_torch/auto_deploy/transform/library/compile_model.pytensorrt_llm/_torch/auto_deploy/compile/backends/torch_cudagraph.pytensorrt_llm/_torch/auto_deploy/compile/backends/torch_simple.pytests/unittest/_torch/auto_deploy/unit/singlegpu/compile/test_cuda_graph_batch_sizes.pytensorrt_llm/_torch/auto_deploy/transform/interface.pytensorrt_llm/_torch/auto_deploy/compile/backends/torch_opt.pytensorrt_llm/_torch/auto_deploy/compile/backends/torch_compile.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:
tests/unittest/_torch/auto_deploy/unit/singlegpu/compile/test_compiler.pyexamples/auto_deploy/build_and_run_flux.pytests/unittest/_torch/auto_deploy/unit/singlegpu/compile/test_captured_graph.pytensorrt_llm/_torch/auto_deploy/transform/library/cleanup_input_constraints.pytensorrt_llm/_torch/auto_deploy/compile/compiler.pytensorrt_llm/_torch/auto_deploy/shim/interface.pytensorrt_llm/_torch/auto_deploy/custom_ops/attention_interface.pytensorrt_llm/_torch/auto_deploy/transformations/_graph.pytensorrt_llm/_torch/auto_deploy/utils/node_utils.pytensorrt_llm/_torch/auto_deploy/transform/library/compile_model.pytensorrt_llm/_torch/auto_deploy/compile/backends/torch_cudagraph.pytensorrt_llm/_torch/auto_deploy/compile/backends/torch_simple.pytests/unittest/_torch/auto_deploy/unit/singlegpu/compile/test_cuda_graph_batch_sizes.pytensorrt_llm/_torch/auto_deploy/transform/interface.pytensorrt_llm/_torch/auto_deploy/compile/backends/torch_opt.pytensorrt_llm/_torch/auto_deploy/compile/backends/torch_compile.py
🧬 Code graph analysis (12)
tests/unittest/_torch/auto_deploy/unit/singlegpu/compile/test_compiler.py (5)
tensorrt_llm/_torch/auto_deploy/compile/backends/torch_compile.py (1)
compile(17-19)tensorrt_llm/_torch/auto_deploy/compile/backends/torch_cudagraph.py (1)
compile(232-243)tensorrt_llm/_torch/auto_deploy/compile/backends/torch_opt.py (1)
compile(26-28)tensorrt_llm/_torch/auto_deploy/compile/backends/torch_simple.py (1)
compile(10-11)tensorrt_llm/_torch/auto_deploy/compile/compiler.py (3)
compile(47-48)CompileBackendRegistry(12-31)get(25-27)
examples/auto_deploy/build_and_run_flux.py (5)
tensorrt_llm/_torch/auto_deploy/compile/backends/torch_compile.py (1)
compile(17-19)tensorrt_llm/_torch/auto_deploy/compile/backends/torch_cudagraph.py (1)
compile(232-243)tensorrt_llm/_torch/auto_deploy/compile/backends/torch_opt.py (1)
compile(26-28)tensorrt_llm/_torch/auto_deploy/compile/backends/torch_simple.py (1)
compile(10-11)tensorrt_llm/_torch/auto_deploy/compile/compiler.py (3)
compile(47-48)CompileBackendRegistry(12-31)get(25-27)
tests/unittest/_torch/auto_deploy/unit/singlegpu/compile/test_captured_graph.py (1)
tensorrt_llm/_torch/auto_deploy/compile/backends/torch_cudagraph.py (3)
compile(232-243)CapturedGraph(28-177)_args_kwargs_flatten_spec(16-19)
tensorrt_llm/_torch/auto_deploy/compile/compiler.py (2)
tensorrt_llm/_torch/auto_deploy/custom_ops/attention_interface.py (3)
register(924-932)decorator(925-930)get(935-937)tensorrt_llm/_torch/auto_deploy/transform/interface.py (2)
register(483-490)get(493-495)
tensorrt_llm/_torch/auto_deploy/shim/interface.py (1)
tensorrt_llm/_torch/auto_deploy/custom_ops/attention_interface.py (4)
GetCacheCallable(773-774)SequenceInfo(35-746)dynamic_shapes(332-334)named_dynamic_shapes(307-329)
tensorrt_llm/_torch/auto_deploy/custom_ops/attention_interface.py (1)
tensorrt_llm/_torch/auto_deploy/shim/interface.py (1)
named_dynamic_shapes(43-47)
tensorrt_llm/_torch/auto_deploy/transform/library/compile_model.py (4)
tensorrt_llm/_torch/auto_deploy/compile/backends/torch_compile.py (1)
compile(17-19)tensorrt_llm/_torch/auto_deploy/compile/backends/torch_cudagraph.py (1)
compile(232-243)tensorrt_llm/_torch/auto_deploy/compile/backends/torch_simple.py (1)
compile(10-11)tensorrt_llm/_torch/auto_deploy/compile/compiler.py (3)
compile(47-48)CompileBackendRegistry(12-31)get(25-27)
tensorrt_llm/_torch/auto_deploy/compile/backends/torch_cudagraph.py (1)
tensorrt_llm/_torch/auto_deploy/compile/compiler.py (4)
CompileBackendRegistry(12-31)CompilerBackend(34-48)register(16-22)compile(47-48)
tensorrt_llm/_torch/auto_deploy/compile/backends/torch_simple.py (4)
tensorrt_llm/_torch/auto_deploy/compile/compiler.py (4)
CompileBackendRegistry(12-31)CompilerBackend(34-48)register(16-22)compile(47-48)tensorrt_llm/_torch/auto_deploy/compile/backends/torch_compile.py (1)
compile(17-19)tensorrt_llm/_torch/auto_deploy/compile/backends/torch_cudagraph.py (1)
compile(232-243)tensorrt_llm/_torch/auto_deploy/compile/backends/torch_opt.py (1)
compile(26-28)
tensorrt_llm/_torch/auto_deploy/transform/interface.py (4)
tensorrt_llm/_torch/auto_deploy/models/factory.py (1)
ModelFactory(23-294)tensorrt_llm/_torch/auto_deploy/shim/interface.py (2)
CachedSequenceInterface(11-88)args(23-25)tensorrt_llm/_torch/auto_deploy/transformations/_graph.py (4)
canonicalize_graph(174-187)lift_to_meta(79-92)named_graphmodules(95-99)run_shape_prop(218-243)tensorrt_llm/logger.py (1)
warning(132-133)
tensorrt_llm/_torch/auto_deploy/compile/backends/torch_opt.py (3)
tensorrt_llm/_torch/auto_deploy/compile/compiler.py (3)
CompileBackendRegistry(12-31)register(16-22)compile(47-48)tensorrt_llm/_torch/auto_deploy/compile/backends/torch_cudagraph.py (2)
TorchCudagraphCompiler(181-257)compile(232-243)tensorrt_llm/_torch/auto_deploy/compile/backends/torch_compile.py (1)
compile(17-19)
tensorrt_llm/_torch/auto_deploy/compile/backends/torch_compile.py (4)
tensorrt_llm/_torch/auto_deploy/compile/compiler.py (4)
CompileBackendRegistry(12-31)CompilerBackend(34-48)register(16-22)compile(47-48)tensorrt_llm/_torch/auto_deploy/compile/backends/torch_cudagraph.py (1)
compile(232-243)tensorrt_llm/_torch/auto_deploy/compile/backends/torch_opt.py (1)
compile(26-28)tensorrt_llm/_torch/auto_deploy/compile/backends/torch_simple.py (1)
compile(10-11)
🪛 Ruff (0.13.2)
tensorrt_llm/_torch/auto_deploy/compile/compiler.py
13-13: Mutable class attributes should be annotated with typing.ClassVar
(RUF012)
40-40: Unused method argument: compiler_kwargs
(ARG002)
tensorrt_llm/_torch/auto_deploy/transform/interface.py
305-305: Do not catch blind exception: Exception
(BLE001)
⏰ 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 (4)
tensorrt_llm/_torch/auto_deploy/custom_ops/attention_interface.py (1)
307-307: LGTM: Type annotation corrected.The return type annotation now correctly reflects the implementation. The property returns a
Dict[str, DynamicShape]whereDynamicShape = Dict[int, Dim](integer keys for tensor dimension indices), consistent with the implementation at lines 315-322 which uses integer indices (0, 1) as keys.tensorrt_llm/_torch/auto_deploy/shim/interface.py (2)
1-1: LGTM: Import updates support type consistency.The removal of
Anyand addition ofDynamicShapefromcustom_ops.attention_interfacealign with the type annotation improvements in this file and maintain type safety.Also applies to: 7-7
38-47: LGTM: Dynamic shapes refactored to use named mapping.The refactoring introduces a named mapping (
named_dynamic_shapes) that aggregates shapes frominfo.named_dynamic_shapesand static shapes for caches. Thedynamic_shapesproperty now derives values from this mapping, consistent with the pattern inattention_interface.py(lines 331-334). Caches use empty dicts{}asDynamicShapeto indicate static (non-dynamic) shapes.tensorrt_llm/_torch/auto_deploy/transform/library/cleanup_input_constraints.py (1)
24-57: Validate placeholder indexing in transform pipeline
Only onefind_nodes(op="placeholder")[1]occurrence was found; manually ensure all graphs have at least two placeholders, that the second is correct for constraint cleanup, document the first placeholder’s role, and update any other dependent transforms.
tensorrt_llm/_torch/auto_deploy/transform/library/cleanup_input_constraints.py
Show resolved
Hide resolved
|
PR_Github #20557 [ run ] triggered by Bot |
tensorrt_llm/_torch/auto_deploy/transform/library/cleanup_input_constraints.py
Show resolved
Hide resolved
|
trying to understand the main change here to focus on: Is it the use of if so, then I find the PR title a bit confusing ( |
|
PR_Github #20557 [ run ] completed with state |
4 participants
Summary by CodeRabbit
New Features
Improvements
Bug Fixes
Refactor
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.