vllm
vllm copied to clipboard

Published 20 hours ago •

Reame
Issues

Add pipeline parallel support to `TransformersModel`

Open hmellor opened this issue 9 months ago • 7 comments

Merge after:

https://github.com/huggingface/transformers/pull/36091
#13403

Tested with Llama 3.1 70B in pp=4 and pp2,tp2 configurations.

The following command can be used for testing:

CUDA_VISIBLE_DEVICES=0,1 vllm serve meta-llama/Llama-3.2-1B-Instruct --disable-log-requests --model-impl transformers --gpu-memory-utilization 0.25 --pipeline-parallel-size 2

Feb 06 '25 14:02 hmellor

👋 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.

🚀

Feb 06 '25 14:02 github-actions[bot]

This pull request has merge conflicts that must be resolved before it can be merged. Please rebase the PR, @hmellor.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

Feb 12 '25 11:02 mergify[bot]

Can only be merged after Transformers v4.49, but marking as ready for review so this can be merged soon after.

Feb 12 '25 12:02 hmellor

This pull request has merge conflicts that must be resolved before it can be merged. Please rebase the PR, @hmellor.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

Feb 13 '25 21:02 mergify[bot]

This pull request has merge conflicts that must be resolved before it can be merged. Please rebase the PR, @hmellor.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

Feb 14 '25 11:02 mergify[bot]

Please merge the latest changes from main

Feb 18 '25 05:02 DarkLight1337

This pull request has merge conflicts that must be resolved before it can be merged. Please rebase the PR, @hmellor.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

Feb 18 '25 10:02 mergify[bot]

Let's update this after #13634 to ensure that nothing is broken.

Feb 22 '25 13:02 DarkLight1337

This pull request has merge conflicts that must be resolved before it can be merged. Please rebase the PR, @hmellor.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

Mar 02 '25 17:03 mergify[bot]

The test passes locally, but I've commented it out so that we are not blocked by https://github.com/vllm-project/vllm/pull/13905

I'd be keen to get this merged because it makes quite a few changes to the TransformersModel without breaking it for non-PP cases. Merging would allow me to commit smaller QoL changes and pave the way for VLM support.

Mar 24 '25 20:03 hmellor