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

[AMD][ROCm] Enable DeepSeek model on ROCm #12662

Merged
merged 6 commits into from
Feb 4, 2025
Merged

[AMD][ROCm] Enable DeepSeek model on ROCm #12662

merged 6 commits into from
Feb 4, 2025

Conversation

hongxiayang
Copy link
Collaborator

@hongxiayang hongxiayang commented Feb 2, 2025 •
edited by github-actions bot
Loading

Thanks for great work from vLLM community to enable DeepSeek model.
This PR is to use the Triton MLA backend on ROCm to enable DeepSeek model on AMD GPUs.

Tests:

  • Tested on DeepSeek V2 (DeepSeek-Coder-V2-Lite-Instruct) model on MI300x and MI210.
  • Tested on DeepSeek V3 dummy weights on MI300x.
  • Added two unit tests.

Note: to use DeepSeek model, we should ensure to have the "trust_remote_code" flag on.

@github-actions GitHub Actions
Copy link

github-actions bot commented Feb 2, 2025

👋 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.

🚀

mawong-amd and others added 2 commits February 2, 2025 11:10
@mawong-amd @hongxiayang
Fix MLA and logic for using triton scaled_mm on ROCm as blockwise FP8…
2fa8a9d 
… quant is not supported

Signed-off-by: Hongxia Yang <hongxia.yang@amd.com>
@hongxiayang
use MLA on rocm
3523ce5 
Signed-off-by: Hongxia Yang <hongxia.yang@amd.com>
@hongxiayang
pre-commit format
3930fdd 
Signed-off-by: Hongxia Yang <hongxia.yang@amd.com>
@hongxiayang hongxiayang marked this pull request as ready for review February 2, 2025 18:27
@hongxiayang hongxiayang added ready ONLY add when PR is ready to merge/full CI is needed rocm labels Feb 3, 2025
mgoin
mgoin reviewed Feb 3, 2025
View reviewed changes
Copy link
Member

@mgoin mgoin left a comment

Choose a reason for hiding this comment

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

LGTM! Would you able to add DeepSeek-Coder-V2-Lite-Instruct as a model test for the AMD runner with a check that TritonMLA is chosen for the attention module?

@hongxiayang
add a rocm_attention_selector test for MLA as test_attention_selector…
6459100 
….py has to be ignored today

Signed-off-by: Hongxia Yang <hongxia.yang@amd.com>
@hongxiayang
license header
0d0190e 
Signed-off-by: Hongxia Yang <hongxia.yang@amd.com>
@hongxiayang
add a triton mla attn_backend test for deepseek v2 model
277e860 
Signed-off-by: Hongxia Yang <hongxia.yang@amd.com>
mgoin
mgoin approved these changes Feb 4, 2025
View reviewed changes
Copy link
Member

@mgoin mgoin left a comment

Choose a reason for hiding this comment

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

Thank you

@mgoin mgoin enabled auto-merge (squash) February 4, 2025 01:32
@hongxiayang
Copy link
Collaborator Author

LGTM! Would you able to add DeepSeek-Coder-V2-Lite-Instruct as a model test for the AMD runner with a check that TritonMLA is chosen for the attention module?

@mgoin

Thank you

@mgoin Thank your for your review and approval.

@mgoin mgoin merged commit c36ac98 into vllm-project:main Feb 4, 2025
54 checks passed
fxmarty-amd pushed a commit to fxmarty-amd/vllm that referenced this pull request Feb 7, 2025
@hongxiayang @mawong-amd @fxmarty-amd
[AMD][ROCm] Enable DeepSeek model on ROCm (vllm-project#12662)
22c5c67 
Signed-off-by: Hongxia Yang <hongxia.yang@amd.com>
Co-authored-by: Matthew Wong <Matthew.Wong2@amd.com>
Signed-off-by: Felix Marty <felmarty@amd.com>
ShangmingCai pushed a commit to ShangmingCai/vllm that referenced this pull request Feb 10, 2025
@hongxiayang @mawong-amd @ShangmingCai
[AMD][ROCm] Enable DeepSeek model on ROCm (vllm-project#12662)
deea142 
Signed-off-by: Hongxia Yang <hongxia.yang@amd.com>
Co-authored-by: Matthew Wong <Matthew.Wong2@amd.com>
@tlrmchlsmth tlrmchlsmth mentioned this pull request Feb 14, 2025
Merged
panf2333 pushed a commit to yottalabsai/vllm that referenced this pull request Feb 18, 2025
@hongxiayang @mawong-amd @panf2333
[AMD][ROCm] Enable DeepSeek model on ROCm (vllm-project#12662)
659d02d 
Signed-off-by: Hongxia Yang <hongxia.yang@amd.com>
Co-authored-by: Matthew Wong <Matthew.Wong2@amd.com>
kerthcet pushed a commit to kerthcet/vllm that referenced this pull request Feb 21, 2025
@hongxiayang @mawong-amd @kerthcet
[AMD][ROCm] Enable DeepSeek model on ROCm (vllm-project#12662)
cedf9c0 
Signed-off-by: Hongxia Yang <hongxia.yang@amd.com>
Co-authored-by: Matthew Wong <Matthew.Wong2@amd.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@mgoin mgoin mgoin approved these changes

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

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

@DarkLight1337 DarkLight1337 Awaiting requested review from DarkLight1337

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

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

Successfully merging this pull request may close these issues.
3 participants
@hongxiayang @mgoin @mawong-amd