Add test case for compiling multiple graphs #21044

sarckk · 2025-07-16T09:33:21Z

Essential Elements of an Effective PR Description Checklist

The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
The test plan, such as providing test command.
The test results, such as pasting the results comparison before and after, or e2e results
(Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.

Purpose

In some cases, we might want to compile multiple subgraphs, e.g. if some submodules in the model are not traceable with torch dynamo. See #14913 and #19719 for motivating scenarios.

This PR adds a test case showcasing how this can be supported in vLLM for piecewise compilation. This requires several components:

Each subgraph in the model being compiled needs to be initialized with set_model_tag(tag) to ensure compile cache is not shared between these subgraphs (initially added in [torch.compile] reorganize the cache directory to support compiling multiple models #19064 to support compiling multiple models)
Pre-allocate static buffers to be used as arg inputs to each compiled subgraph as CUDA Graph expects static tensor addresses between runs
Added ignore_torch_compile helper decorator in scenarios where a class inherits a parent module that has support_torch_compile decorator applied but this class does not want to compile its forward(). Added test_ignore_torch_compile_decorator that tests this.

Also disables AOTAutogradCache in StandaloneInductor adaptor which is needed for compilation caching to work. StandaloneInductorAdaptor in PyTorch 2.8+ will fix this problem (thanks @zou3519)

Test Plan

pytest tests/compile/piecewise/test_multiple_graphs.py

Test Result

Both newly added unit tests (test_ignore_torch_compile_decorator and test_multi_graph_piecewise_compile) pass

github-actions · 2025-07-16T09:33:30Z

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can either: Add ready label to the PR or enable auto-merge.

🚀

sarckk · 2025-07-16T09:34:26Z

cc: @zou3519 @youkaichao

gemini-code-assist

Code Review

This pull request introduces a new ignore_torch_compile decorator and adds test cases for compiling multiple graphs. There is a potential issue with the inheritance of the IGNORE_COMPILE_KEY that needs to be addressed.

vllm/compilation/decorators.py

zou3519

Looks reasonable to me. One minor comment

zou3519 · 2025-07-16T14:43:55Z

cc @BoyuanFeng

tests/compile/piecewise/test_multiple_graphs.py

BoyuanFeng

looks good! a few minor comments

tests/compile/piecewise/test_multiple_graphs.py

sarckk · 2025-07-22T01:39:46Z

@houseroad the failing tests don't seem related to this PR, can we merge it? e.g. for model tests (voxtral unrecognized), the test also fails on main

Signed-off-by: Yong Hoon Shin <yhshin@meta.com>

… and compile Signed-off-by: Yong Hoon Shin <yhshin@meta.com>

zou3519 · 2025-07-22T18:47:26Z

vllm/compilation/compiler_interface.py

+                # InductorAdaptor (unfortunately) requires AOTAutogradCache
+                # to be turned off to run. It will fail to acquire the hash_str
+                # and error if not.
+                # StandaloneInductorAdaptor (PyTorch 2.8+) fixes this problem.
+                stack.enter_context(
+                    torch._functorch.config.patch(enable_autograd_cache=False))


btw @houseroad @yeqcharlotte this will fix a lot of the compiler cache issues people report

Signed-off-by: Yong Hoon Shin <yhshin@meta.com> Signed-off-by: qizixi <qizixi@meta.com>

Signed-off-by: Yong Hoon Shin <yhshin@meta.com> Signed-off-by: x22x22 <wadeking@qq.com>

Signed-off-by: Yong Hoon Shin <yhshin@meta.com>

Signed-off-by: Yong Hoon Shin <yhshin@meta.com> Signed-off-by: Jinzhen Lin <linjinzhen@hotmail.com>

Signed-off-by: Yong Hoon Shin <yhshin@meta.com> Signed-off-by: Paul Pak <paulpak58@gmail.com>

Signed-off-by: Yong Hoon Shin <yhshin@meta.com> Signed-off-by: Diego-Castan <diego.castan@ibm.com>

Signed-off-by: Yong Hoon Shin <yhshin@meta.com>

sarckk requested review from youkaichao and zou3519 as code owners July 16, 2025 09:33

sarckk requested review from zou3519 and removed request for zou3519 July 16, 2025 09:33

gemini-code-assist bot reviewed Jul 16, 2025

View reviewed changes

vllm/compilation/decorators.py Outdated Show resolved Hide resolved

zou3519 reviewed Jul 16, 2025

View reviewed changes

vllm/compilation/decorators.py Outdated Show resolved Hide resolved

zou3519 requested a review from ProExpertProg July 16, 2025 14:40

zou3519 reviewed Jul 16, 2025

View reviewed changes

sarckk commented Jul 16, 2025

View reviewed changes

tests/compile/piecewise/test_multiple_graphs.py Outdated Show resolved Hide resolved

BoyuanFeng reviewed Jul 16, 2025

View reviewed changes

tests/compile/piecewise/test_multiple_graphs.py Outdated Show resolved Hide resolved

tests/compile/piecewise/test_multiple_graphs.py Outdated Show resolved Hide resolved

tests/compile/piecewise/test_multiple_graphs.py Outdated Show resolved Hide resolved

sarckk added the torch.compile label Jul 16, 2025

github-project-automation bot added this to torch.compile integration Jul 16, 2025

github-project-automation bot moved this to To triage in torch.compile integration Jul 16, 2025

sarckk removed the torch.compile label Jul 16, 2025

zou3519 approved these changes Jul 18, 2025

View reviewed changes

BoyuanFeng approved these changes Jul 18, 2025

View reviewed changes

sarckk added the ready ONLY add when PR is ready to merge/full CI is needed label Jul 18, 2025

sarckk force-pushed the compile-multiple-graphs branch from c178193 to 0080859 Compare July 19, 2025 04:00

sarckk added 8 commits July 22, 2025 11:17

Add test case for compiling multiple graphs

c5e1b7b

Signed-off-by: Yong Hoon Shin <yhshin@meta.com>

Lint and minor fixes

811ebd1

Signed-off-by: Yong Hoon Shin <yhshin@meta.com>

Fix grandchild support_torch_compile ignored

764a879

Signed-off-by: Yong Hoon Shin <yhshin@meta.com>

Update comment

f7e1489

Signed-off-by: Yong Hoon Shin <yhshin@meta.com>

Randomize test inputs

3907b6e

Signed-off-by: Yong Hoon Shin <yhshin@meta.com>

Disable AOTAutograd cache in InductorAdaptor

e130e0f

Signed-off-by: Yong Hoon Shin <yhshin@meta.com>

Add torch.equal check for use_cudagraph=False

d8f9c71

Signed-off-by: Yong Hoon Shin <yhshin@meta.com>

Make model slightly more complex to expose numeric diff between eager…

7ece902

… and compile Signed-off-by: Yong Hoon Shin <yhshin@meta.com>

sarckk force-pushed the compile-multiple-graphs branch from 0080859 to 7ece902 Compare July 22, 2025 18:18

zou3519 reviewed Jul 22, 2025

View reviewed changes

houseroad enabled auto-merge (squash) July 23, 2025 00:48

vllm-bot merged commit 4ac7713 into vllm-project:main Jul 23, 2025
65 of 67 checks passed

github-project-automation bot moved this from To triage to Done in torch.compile integration Jul 23, 2025

zixi-qi pushed a commit to zixi-qi/vllm that referenced this pull request Jul 23, 2025

Add test case for compiling multiple graphs (vllm-project#21044)

49b48be

Signed-off-by: Yong Hoon Shin <yhshin@meta.com> Signed-off-by: qizixi <qizixi@meta.com>

sarckk deleted the compile-multiple-graphs branch July 24, 2025 00:08

sarckk mentioned this pull request Aug 5, 2025

Support conditional torch.compile per module #22269

Merged

4 tasks

x22x22 pushed a commit to x22x22/vllm that referenced this pull request Aug 5, 2025

Add test case for compiling multiple graphs (vllm-project#21044)

c80b511

Signed-off-by: Yong Hoon Shin <yhshin@meta.com> Signed-off-by: x22x22 <wadeking@qq.com>

sarckk mentioned this pull request Aug 5, 2025

Add option to disable weakref conversion for last piecewise cudagraph in a module #22282

Open

4 tasks

Pradyun92 pushed a commit to Pradyun92/vllm that referenced this pull request Aug 6, 2025

Add test case for compiling multiple graphs (vllm-project#21044)

923f637

Signed-off-by: Yong Hoon Shin <yhshin@meta.com>

npanpaliya pushed a commit to odh-on-pz/vllm-upstream that referenced this pull request Aug 6, 2025

Add test case for compiling multiple graphs (vllm-project#21044)

2ffcf9d

Signed-off-by: Yong Hoon Shin <yhshin@meta.com>

jinzhen-lin pushed a commit to jinzhen-lin/vllm that referenced this pull request Aug 9, 2025

Add test case for compiling multiple graphs (vllm-project#21044)

f07db3a

Signed-off-by: Yong Hoon Shin <yhshin@meta.com> Signed-off-by: Jinzhen Lin <linjinzhen@hotmail.com>

paulpak58 pushed a commit to paulpak58/vllm that referenced this pull request Aug 13, 2025

Add test case for compiling multiple graphs (vllm-project#21044)

ac49f2b

Signed-off-by: Yong Hoon Shin <yhshin@meta.com> Signed-off-by: Paul Pak <paulpak58@gmail.com>

diegocastanibm pushed a commit to diegocastanibm/vllm that referenced this pull request Aug 15, 2025

Add test case for compiling multiple graphs (vllm-project#21044)

3cc1001

Signed-off-by: Yong Hoon Shin <yhshin@meta.com> Signed-off-by: Diego-Castan <diego.castan@ibm.com>

epwalsh pushed a commit to epwalsh/vllm that referenced this pull request Aug 28, 2025

Add test case for compiling multiple graphs (vllm-project#21044)

799a54d

Signed-off-by: Yong Hoon Shin <yhshin@meta.com>

Uh oh!

Add test case for compiling multiple graphs #21044

Add test case for compiling multiple graphs #21044

Uh oh!

Conversation

sarckk commented Jul 16, 2025 • edited by github-actions bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Essential Elements of an Effective PR Description Checklist

Purpose

Test Plan

Test Result

Uh oh!

github-actions bot commented Jul 16, 2025

Uh oh!

sarckk commented Jul 16, 2025

Uh oh!

gemini-code-assist bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

Uh oh!

Uh oh!

zou3519 left a comment

Choose a reason for hiding this comment

Uh oh!

zou3519 commented Jul 16, 2025

Uh oh!

Uh oh!

BoyuanFeng left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

sarckk commented Jul 22, 2025

Uh oh!

zou3519 Jul 22, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

sarckk commented Jul 16, 2025 •

edited by github-actions bot

Loading

zou3519 Jul 22, 2025 •

edited

Loading