Skip to content

feat: add vllm pipeline parallelism and multi node rollout #460

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
Merged
merged 48 commits into from
Jun 3, 2025
Merged

feat: add vllm pipeline parallelism and multi node rollout #460

merged 48 commits into from
Jun 3, 2025

Conversation

parthchadha
Copy link
Contributor

What does this PR do ?

Add a one line overview of what this PR aims to accomplish.

Issues

List issues that this PR closes (syntax):

Usage

  • You can potentially add a usage example below
# Add a code snippet demonstrating how to use this

Before your PR is "Ready for review"

Pre checks:

  • Make sure you read and followed Contributor guidelines
  • Did you write any new necessary tests?
  • Did you run the unit tests and functional tests locally? Visit our Testing Guide for how to run tests
  • Did you add or update any necessary documentation? Visit our Document Development Guide for how to write, build and test the docs.

Additional Information

  • ...

parthchadha and others added 30 commits May 19, 2025 17:03
@parthchadha
feat: add async vllm engine for verifying its correctness in unit tests
3b6e457 
Signed-off-by: Parth Chadha <pchadha@nvidia.com>
@parthchadha
Merge remote-tracking branch 'origin/main' into pchadha/async-vllm
c1d61a3
@parthchadha
fix: Fix failing unit test
fb47ba9 
Signed-off-by: Parth Chadha <pchadha@nvidia.com>
@parthchadha
Basic pipeline config
2aebfa2 
Signed-off-by: Parth Chadha <pchadha@nvidia.com>
@parthchadha
Add more pp logic (with failing tests)
0434c15 
Signed-off-by: Parth Chadha <pchadha@nvidia.com>
@parthchadha
Make single placement group instead of one per node
3a5098c 
Signed-off-by: Parth Chadha <pchadha@nvidia.com>
@parthchadha
WIP changes to allow pp loading of model
5eb7a95 
Signed-off-by: Parth Chadha <pchadha@nvidia.com>
@parthchadha
add unit tests for larger model testing (failing unit tests with 32b)
298ee41 
Signed-off-by: Parth Chadha <pchadha@nvidia.com>
@parthchadha
Fix to get tp 4 working with async policy and single placement group
899131b 
Signed-off-by: Parth Chadha <pchadha@nvidia.com>
@parthchadha
pp test working
abce21d 
Signed-off-by: Parth Chadha <pchadha@nvidia.com>
@parthchadha
Unify tp and pp code
e7b2df6 
Signed-off-by: Parth Chadha <pchadha@nvidia.com>
@parthchadha
fix large model test
f47366f 
Signed-off-by: Parth Chadha <pchadha@nvidia.com>
@parthchadha
Merge remote-tracking branch 'origin/main' into pchadha/async-vllm
973fcdb
@parthchadha
Add back commented tests
d4910ce 
Signed-off-by: Parth Chadha <pchadha@nvidia.com>
@parthchadha
Create separate async vs sync apis
cca95cc 
Signed-off-by: Parth Chadha <pchadha@nvidia.com>
@parthchadha
Add async flag to recipes
14e821c 
Signed-off-by: Parth Chadha <pchadha@nvidia.com>
@parthchadha
Merge remote-tracking branch 'origin/pchadha/async-vllm' into pchadha…
adceb51 
…/multi-node-rollouts
@terrykong
wip
840532d 
Signed-off-by: Terry Kong <terryk@nvidia.com>

fix it

Signed-off-by: Terry Kong <terryk@nvidia.com>

make it match

Signed-off-by: Terry Kong <terryk@nvidia.com>

fix it all

Signed-off-by: Terry Kong <terryk@nvidia.com>

cleanup

Signed-off-by: Terry Kong <terryk@nvidia.com>
@parthchadha
Merge remote-tracking branch 'origin/use-default-uvcache' into pchadh…
4415596 
…a/multi-node-rollouts
@parthchadha
Update uv.lock
8840fef 
Signed-off-by: Parth Chadha <pchadha@nvidia.com>
@parthchadha
Merge remote-tracking branch 'origin/pchadha/async-vllm' into pchadha…
6ccd72a 
…/multi-node-rollouts
@parthchadha
Create a separate test for large model multi node
bbd9dbb 
Signed-off-by: Parth Chadha <pchadha@nvidia.com>
@parthchadha
Fix failing unit test due to wrong access of called_workers
4abd54e 
Signed-off-by: Parth Chadha <pchadha@nvidia.com>
@parthchadha
Add a comment about not using async as of now
b0cedec 
Signed-off-by: Parth Chadha <pchadha@nvidia.com>
@parthchadha
Merge branch 'pchadha/async-vllm' into pchadha/multi-node-rollouts
e524179
@parthchadha
Merge remote-tracking branch 'origin/main' into pchadha/multi-node-ro…
f3bc77f 
…llouts
@parthchadha
Remove dead code
48e9237 
Signed-off-by: Parth Chadha <pchadha@nvidia.com>
@parthchadha
Merge remote-tracking branch 'origin/main' into pchadha/async-vllm
15ca5be
@parthchadha
Insert back missing type hints
2aa6d2c 
Signed-off-by: Parth Chadha <pchadha@nvidia.com>
@parthchadha
Merge branch 'pchadha/async-vllm' into pchadha/multi-node-rollouts
86c8222
@parthchadha
Fix the wrong type annotation for ray placement group
001d1a8 
Signed-off-by: Parth Chadha <pchadha@nvidia.com>
@parthchadha parthchadha force-pushed the pchadha/multi-node-rollouts branch from b071796 to 001d1a8 Compare June 2, 2025 02:12
@parthchadha parthchadha added CI:L0 Run doctests and unit tests and removed CI:L0 Run doctests and unit tests labels Jun 2, 2025
@parthchadha parthchadha added CI:L0 Run doctests and unit tests and removed CI:L0 Run doctests and unit tests labels Jun 2, 2025
@parthchadha
Use one pg per node by default unless vllm requires multiple nodes
0d4bc16 
Signed-off-by: Parth Chadha <pchadha@nvidia.com>
@parthchadha parthchadha added CI:L0 Run doctests and unit tests and removed CI:L0 Run doctests and unit tests labels Jun 2, 2025
@parthchadha parthchadha added CI:L0 Run doctests and unit tests and removed CI:L0 Run doctests and unit tests labels Jun 3, 2025
terrykong
terrykong previously approved these changes Jun 3, 2025
View reviewed changes
Copy link
Contributor

@terrykong terrykong left a comment

Choose a reason for hiding this comment

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

approved modulo small typing comment

@parthchadha
Add missing type annotations
2088c9d 
Signed-off-by: Parth Chadha <pchadha@nvidia.com>
@parthchadha parthchadha added this pull request to the merge queue Jun 3, 2025
Merged via the queue into main with commit 03610df Jun 3, 2025
13 of 14 checks passed
@parthchadha parthchadha deleted the pchadha/multi-node-rollouts branch June 3, 2025 02:35
yfw pushed a commit that referenced this pull request Jun 6, 2025
@parthchadha @terrykong @yfw
feat: add vllm pipeline parallelism and multi node rollout (#460)
91219a2 
Signed-off-by: Parth Chadha <pchadha@nvidia.com>
Signed-off-by: Terry Kong <terryk@nvidia.com>
Co-authored-by: Terry Kong <terryk@nvidia.com>
YzjiaoNvd pushed a commit to YzjiaoNvd/NeMo-RL that referenced this pull request Jun 10, 2025
@parthchadha @terrykong
feat: add vllm pipeline parallelism and multi node rollout (NVIDIA-Ne…
4c41903 
…Mo#460)

Signed-off-by: Parth Chadha <pchadha@nvidia.com>
Signed-off-by: Terry Kong <terryk@nvidia.com>
Co-authored-by: Terry Kong <terryk@nvidia.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@terrykong terrykong terrykong approved these changes

Assignees
No one assigned
Labels
CI:L0 Run doctests and unit tests
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
2 participants
@parthchadha @terrykong