-
-
Notifications
You must be signed in to change notification settings - Fork 10.9k
[Compilation fix] add stubs to allow compilation without sm100 #21198
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.
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
[Compilation fix] add stubs to allow compilation without sm100 #21198
Conversation
|
👋 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 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 🚀 |
Signed-off-by: Mickael Seznec <mickael@mistral.ai>
77ef73b to
954b5ed
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Code Review
This pull request fixes a compilation failure when vLLM is built without support for sm100 architectures by adding stub implementations for missing symbols. To fully resolve the issue, I've added a critical comment to include the missing stub for sm100_cutlass_mla_decode.
Signed-off-by: Mickael Seznec <mickael@mistral.ai>
Signed-off-by: Mickael Seznec <mickael@mistral.ai>
|
@LucasWilkinson for viz since you reviewed #16032 (I don't know why the reviewers haven't been set with my PR) |
b29cf72 to
f2fab0b
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Sorry for the delay! is this still needed after: #21020 ?
|
np! Yes I confirm that you fixed it there already :) Closing |
Essential Elements of an Effective PR Description Checklist
supported_models.mdandexamplesfor a new model.Purpose
Previously, when setting
TORCH_CUDA_ARCH_LIST="9.0"for example, vllm would complain because some symbols were missing.sm100_cutlass_mla_get_workspace_sizeis indeed referenced incsrc/torch_bindings.cppAdd stubs to fix this situation.
Test Plan
Test Result
(Optional) Documentation Update