[CI/Build] Add Dockerfile.cuda for NVIDIA GPU users#1439
[CI/Build] Add Dockerfile.cuda for NVIDIA GPU users#1439loveysuby wants to merge 2 commits intovllm-project:mainfrom
Conversation
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 986f39bf65
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
docker/Dockerfile.cuda
Outdated
|
|
||
| RUN ln -sf /usr/bin/python3 /usr/bin/python | ||
|
|
||
| CMD ["/bin/bash"] |
There was a problem hiding this comment.
Remove inherited CMD from vllm-openai stage
The vllm-openai stage is built FROM final, so it inherits CMD ["/bin/bash"] from this stage; with ENTRYPOINT ["vllm", "serve", "--omni"], Docker will run vllm serve --omni /bin/bash by default, which passes /bin/bash as an extra argument and can break container startup unless users override the command. This affects the default build target because vllm-openai is the last stage in this Dockerfile.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Updated to use empty entrypoint for consistency with Dockerfile.rocm.
I was wondering if it might be helpful to use ENTRYPOINT ["vllm", "serve", "--omni"] as a common pattern for both Dockerfiles.rocm & .cuda.
27187cc to
1ea6145
Compare
Signed-off-by: Hyoseop Song <[email protected]>
1ea6145 to
e90721d
Compare
e90721d to
20682ee
Compare
Signed-off-by: Hyoseop Song <[email protected]>
20682ee to
ecc9813
Compare
|
@vllm-omni-reviewer |
PLEASE FILL IN THE PR DESCRIPTION HERE ENSURING ALL CHECKLIST ITEMS (AT THE BOTTOM) HAVE BEEN CONSIDERED.
Purpose
Add
Dockerfile.cudafor NVIDIA GPU users to build custom Docker images.This addresses review feedback on PR #1386 that
Dockerfile.ciis not suitable for end users:.[dev])vllm serve --omniThe new Dockerfile follows the same structure as
Dockerfile.rocm. (#280)Build with a specific vLLM base image version (e.g. v0.14.0 - default: v0.15.0):
DOCKER_BUILDKIT=1 docker build -f docker/Dockerfile.cuda \ --build-arg BASE_IMAGE=vllm/vllm-openai:v0.14.0 \ -t vllm-omni-cuda .Test Plan
Verified Dockerfile syntax with
docker build --checkacross all official vLLM release versions:Test Result
Essential Elements of an Effective PR Description Checklist
supported_models.mdandexamplesfor a new model. Please runmkdocs serveto sync the documentation editions to./docs.BEFORE SUBMITTING, PLEASE READ https://github.com/vllm-project/vllm-omni/blob/main/CONTRIBUTING.md (anything written below this line will be removed by GitHub Actions)