Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[torch.compile] limit inductor threads and lazy import quant #10482

Merged
merged 14 commits into from
Nov 21, 2024
Merged

[torch.compile] limit inductor threads and lazy import quant #10482

merged 14 commits into from
Nov 21, 2024
+178 −64

Conversation

youkaichao
Copy link
Member

fixes #10480

@youkaichao
fix compile
661072c 
Signed-off-by: youkaichao <youkaichao@gmail.com>
@github-actions GitHub Actions
Copy link

👋 Hi! Thank you for contributing to the vLLM project.
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 do one of these:

  • Add ready label to the PR
  • Enable auto-merge.

🚀

@youkaichao youkaichao mentioned this pull request Nov 20, 2024
Closed
1 task
@youkaichao
fix
d46c60d 
Signed-off-by: youkaichao <youkaichao@gmail.com>
@youkaichao
fix
3449e0a 
Signed-off-by: youkaichao <youkaichao@gmail.com>
@youkaichao
fix lazy import quantization
cba7b7a 
Signed-off-by: youkaichao <youkaichao@gmail.com>
@youkaichao youkaichao changed the title [torch.compile] limit threads for compilation [torch.compile] limit inductor threads and lazy import quant Nov 20, 2024
@youkaichao
Copy link
Member Author

youkaichao commented Nov 20, 2024 •
edited
Loading

the root cause is, quantization configs are imported too early. and when quantization configs are imported, ParallelLMHead is imported, triggering @torch.compile for get_masked_input_and_mask .

every process calling torch.compile will spawn many processes (cpu core number of processes), and having so many processes will cause vscode to hang.

DarkLight1337
DarkLight1337 reviewed Nov 20, 2024
View reviewed changes
Copy link
Member

@DarkLight1337 DarkLight1337 left a comment

Choose a reason for hiding this comment

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

Thanks for fixing, can you add a test to ensure that each config in the list can be successfully resolved? (In case someone forgets to update the list after adding a config)

@youkaichao
add tests
cfbc70a 
Signed-off-by: youkaichao <youkaichao@gmail.com>
@youkaichao
fix
1535490 
Signed-off-by: youkaichao <youkaichao@gmail.com>
@youkaichao
polish message
405fada 
Signed-off-by: youkaichao <youkaichao@gmail.com>
@youkaichao
add tests
fda0e16 
Signed-off-by: youkaichao <youkaichao@gmail.com>
@mergify mergify bot added the ci/build label Nov 20, 2024
@youkaichao
fix ops
585dff9 
Signed-off-by: youkaichao <youkaichao@gmail.com>
@youkaichao
fix import
83d6c1f 
Signed-off-by: youkaichao <youkaichao@gmail.com>
@youkaichao
Merge branch 'main' into fix_compile
2cbfb83
@youkaichao
fix command
4228ac3 
Signed-off-by: youkaichao <youkaichao@gmail.com>
@youkaichao youkaichao enabled auto-merge (squash) November 20, 2024 21:10
@github-actions github-actions bot added the ready ONLY add when PR is ready to merge/full CI is needed label Nov 20, 2024
@youkaichao
fix
5fd5164 
Signed-off-by: youkaichao <youkaichao@gmail.com>
@youkaichao
fix
e991365 
Signed-off-by: youkaichao <youkaichao@gmail.com>
DarkLight1337
DarkLight1337 approved these changes Nov 21, 2024
View reviewed changes
Copy link
Member

@DarkLight1337 DarkLight1337 left a comment

Choose a reason for hiding this comment

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

Thanks for fixing!

@youkaichao youkaichao disabled auto-merge November 21, 2024 02:36
@youkaichao youkaichao merged commit 388ee3d into vllm-project:main Nov 21, 2024
69 of 71 checks passed
@youkaichao youkaichao deleted the fix_compile branch November 21, 2024 02:36
tlrmchlsmth pushed a commit to neuralmagic/vllm that referenced this pull request Nov 23, 2024
@youkaichao @tlrmchlsmth
[torch.compile] limit inductor threads and lazy import quant (vllm-pr…
e658da5 
…oject#10482)

Signed-off-by: youkaichao <youkaichao@gmail.com>
Signed-off-by: Tyler Michael Smith <tyler@neuralmagic.com>
@jeejeelee jeejeelee mentioned this pull request Nov 25, 2024
Closed
1 task
mfournioux pushed a commit to mfournioux/vllm that referenced this pull request Nov 28, 2024
@youkaichao @mfournioux
[torch.compile] limit inductor threads and lazy import quant (vllm-pr…
d866125 
…oject#10482)

Signed-off-by: youkaichao <youkaichao@gmail.com>
Signed-off-by: Maxime Fournioux <55544262+mfournioux@users.noreply.github.com>
sleepwalker2017 pushed a commit to sleepwalker2017/vllm that referenced this pull request Dec 13, 2024
@youkaichao
[torch.compile] limit inductor threads and lazy import quant (vllm-pr…
659bd7d 
…oject#10482)

Signed-off-by: youkaichao <youkaichao@gmail.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@jeejeelee jeejeelee jeejeelee left review comments

@DarkLight1337 DarkLight1337 DarkLight1337 approved these changes

@zhuohan123 zhuohan123 Awaiting requested review from zhuohan123

@alexm-neuralmagic alexm-neuralmagic Awaiting requested review from alexm-neuralmagic

@comaniac comaniac Awaiting requested review from comaniac

@njhill njhill Awaiting requested review from njhill

@mgoin mgoin Awaiting requested review from mgoin mgoin is a code owner

@robertgshaw2-neuralmagic robertgshaw2-neuralmagic Awaiting requested review from robertgshaw2-neuralmagic robertgshaw2-neuralmagic is a code owner

Assignees
No one assigned
Labels
ci/build ready ONLY add when PR is ready to merge/full CI is needed
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.

[Usage]: VSCode debugger is hanging
4 participants
@youkaichao @mgoin @jeejeelee @DarkLight1337