[AMD] Merge Dockerfiles for ROCm by akao-amd · Pull Request #19203 · sgl-project/sglang

akao-amd · 2026-02-24T01:53:21Z

Motivation

rocm720.Dockerfile and rocm.Dockerfile shares a lot in common. It is time to unify them.

Modifications

Making rocm720.Dockerfile a superset of rocm.Dockerfile was the design choice back in #17799 . This PR mostly replaces rocm.Dockerfile with rocm720.Dockerfile, and modifies all the users of it accordingly. Refactoring workflow files is a reasonable next step but not include in this PR.

Accuracy Tests

Both rocm700-mi35x and rocm720-mi35x pass the accuracy test of DEEPSEEK-R1-MXFP4-Preview model.

Benchmarking and Profiling

N/A.

Checklist

Format your code according to the Format code with pre-commit.
Add unit tests according to the Run and add unit tests.
Update documentation according to Write documentations.
Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.
Follow the SGLang code style guidance.

Review Process

Ping Merge Oncalls to start the PR flow. See the PR Merge Process.
Get approvals from CODEOWNERS and other reviewers.
Trigger CI tests with comments or contact authorized users to do so.
- /tag-run-ci-label, /rerun-failed-ci, /tag-and-rerun-ci
After green CI and required approvals, ask Merge Oncalls to merge.

gemini-code-assist · 2026-02-24T01:53:39Z

Summary of Changes

Hello @akao-amd, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request streamlines the Docker build process for ROCm environments by consolidating two previously separate Dockerfiles, rocm.Dockerfile and rocm720.Dockerfile, into a single, comprehensive rocm.Dockerfile. This change integrates ROCm 7.2 specific configurations and necessary dependency hotfixes directly into the main build definition, reducing redundancy and simplifying the overall build system. The modifications ensure that both ROCm 7.0 and 7.2 environments can be built from a single source, maintaining accuracy and compatibility across different ROCm versions.

Highlights

Dockerfile Unification: The rocm720.Dockerfile has been merged into rocm.Dockerfile, consolidating the build logic for different ROCm versions into a single Dockerfile. This simplifies maintenance and ensures consistency.
ROCm 7.2 Support Enhancements: The unified rocm.Dockerfile now includes specific build arguments, base images, and conditional logic to properly support ROCm 7.2, including installing the AMD SMI Python package and applying hot patches for AITER and PyTorch/Triton dependencies.
Triton and AITER Dependency Management: Updated the Triton repository and commit hash. Introduced hot patches to address specific dependency issues with PyTorch/Triton and AITER when building with ROCm 7.2, ensuring compatibility and correct functionality.
CI Script Simplification: The CI script amd_ci_start_container.sh was updated to remove the conditional selection of Dockerfiles, now always pointing to the unified rocm.Dockerfile.

🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console.

Changelog

docker/rocm.Dockerfile
- Updated Docker build command comments to reflect v0.5.8.post1 and added rocm720 specific build examples.
- Introduced new ARG declarations for BASE_IMAGE_942_ROCM720 and BASE_IMAGE_950_ROCM720.
- Added new FROM stages (gfx942-rocm720 and gfx950-rocm720) to define base images and environment variables for ROCm 7.2.
- Updated TRITON_REPO to https://github.com/triton-lang/triton.git and TRITON_COMMIT to a new hash.
- Implemented conditional installation of the AMD SMI Python package for rocm720 builds.
- Added psutil and pybind11 as required dependencies for AITER.
- Introduced a conditional hot patch for AITER's pa_mqa_logits.py for ROCm 7.2 builds.
- Appended export PYTHONPATH=/sgl-workspace/aiter:${PYTHONPATH} to /etc/bash.bashrc.
- Removed the previous Triton build section, replacing it with a later hot patch.
- Added vim to the list of system packages installed.
- Added tabulate to the list of Python tools installed.
- Set ENV PYTORCH_ROCM_ARCH=gfx942;gfx950.
- Included a Python script for a hot patch to modify the torch-ROCm wheel, relaxing its Triton dependency for ROCm 7.2.
- Added a new hot patch section for Triton, building it from source for ROCm 7.2.
- Added ENV SGLANG_DISABLE_CUDNN_CHECK=1.
docker/rocm720.Dockerfile
- Removed the entire Dockerfile, as its functionality has been integrated into rocm.Dockerfile.
scripts/ci/amd/amd_ci_start_container.sh
- Removed conditional logic for selecting rocm720.Dockerfile or rocm.Dockerfile, now always using rocm.Dockerfile.

Ignored Files

Ignored by pattern: .github/workflows/** (3)
- .github/workflows/pr-test-amd-rocm720.yml
- .github/workflows/release-docker-amd-rocm720-nightly.yml
- .github/workflows/release-docker-amd.yml

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	`/gemini review`	Performs a code review for the current pull request in its current state.
Pull Request Summary	`/gemini summary`	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	`/gemini help`	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

gemini-code-assist

Code Review

The pull request successfully unifies rocm720.Dockerfile into rocm.Dockerfile, streamlining the Dockerfile management for ROCm builds. The changes involve updating base images, adding ROCm 7.2 specific patches for AMD SMI and AITER, and modifying the CI script to use the unified Dockerfile. The accuracy tests passed for the specified models, indicating functional correctness. The removal of rocm720.Dockerfile and the consolidation of logic are positive steps towards better maintainability.

docker/rocm.Dockerfile

gemini-code-assist · 2026-02-24T01:56:59Z

docker/rocm.Dockerfile

+        fi \
+     && echo "export PYTHONPATH=/sgl-workspace/aiter:\${PYTHONPATH}" >> /etc/bash.bashrc


The PYTHONPATH export should ideally be placed in a more persistent configuration file like /etc/profile or a custom script that sources it, rather than /etc/bash.bashrc. While /etc/bash.bashrc works for interactive shells, it might not be sourced in non-interactive contexts (e.g., when running scripts or services), leading to unexpected behavior. Alternatively, if this is only for the build environment, it could be set as an ENV variable in the Dockerfile itself, or passed as a build argument to the Python setup.py command if applicable.

&& echo "export PYTHONPATH=/sgl-workspace/aiter:${PYTHONPATH}" >> /etc/profile

docker/rocm.Dockerfile

HaiShaw · 2026-02-26T00:23:29Z

/tag-and-rerun-ci

HaiShaw

Let's keep using release-docker-amd-nightly.yml vs. release-docker-amd-rocm720-nightly.yml

docker/rocm.Dockerfile

HaiShaw · 2026-02-26T00:46:03Z

docker/rocm.Dockerfile

@@ -117,17 +174,6 @@ RUN cd aiter \
          sh -c "GPU_ARCHS=$GPU_ARCH_LIST python setup.py develop"; \
        fi



Lost && echo "export PYTHONPATH=/sgl-workspace/aiter:\${PYTHONPATH}" >> /etc/bash.bashrc in case ROCm 7.2
Please be careful at merging two dockerfiles

We (YC, BingXu) discussed yesterday that the line wasn't needed.

This is needed so sys.path on ROCm 7.2 is set properly to include /sgl-workspace/aiter

Sorry but can you elaborate? If it is the case, why doesn't ROCm 7.0 image need it in the first place?
CC. @yctseng0211 @bingxche

HaiShaw · 2026-02-26T00:46:29Z

docker/rocm.Dockerfile

-     && cd python \
-     && python setup.py install; \
-    fi
-


Should keep this for ROCm 7.0

It was just move to bottom, not removed.

docker/rocm.Dockerfile

HaiShaw · 2026-02-26T00:55:39Z

docker/rocm.Dockerfile

+     && pip install -r python/requirements.txt \
+     && pip install -e .; \
+    fi
+


L426-L509, pls keep it identical as in ROCm 7.2 dockerfile, and apply it only to ROCm7.2 build

No. I believe BUILD_TRITON is already a better flag for this block and serves its purpose correctly.

Currently ROCm 7.2 images gives BUILD_TRITON=1 while ROCm 7.0 images give BUILD_TRITON=0. The flag has a finer granularity that can serve us well. For example, if we don't want to update gfx942-rocm720 for some reason, then it can keep as is while we may have fixed the triton dependency issue for gfx950-rocm720.

akao-amd · 2026-02-26T01:27:12Z

@HaiShaw Thanks for the review. Fixed two positions (vllm removal and cancel prompt for pip install) accordingly. For the rest I commented without modification.

bingxche · 2026-02-27T08:12:08Z

Nightly Release successfully
rocm700: https://github.com/sgl-project/sglang/actions/runs/22471253280/job/65088525837
rocm720: https://github.com/sgl-project/sglang/actions/runs/22471255572/job/65088532429

Co-authored-by: bingxche <Bingxu.Chen@amd.com>

akao-amd requested review from Fridge003, HaiShaw, Kangyan-Zhou, ishandhanani, ispobock, merrymercy and yctseng0211 as code owners February 24, 2026 01:53

github-actions bot added the amd label Feb 24, 2026

gemini-code-assist bot reviewed Feb 24, 2026

View reviewed changes

akao-amd force-pushed the amd_merge_720_dockerfile branch from 8a57843 to 7a49043 Compare February 25, 2026 07:28

github-actions bot added the run-ci label Feb 26, 2026

HaiShaw requested changes Feb 26, 2026

View reviewed changes

akao-amd requested a review from bingxche as a code owner February 26, 2026 01:24

akao-amd force-pushed the amd_merge_720_dockerfile branch from a27cd12 to 525cbc6 Compare February 26, 2026 05:29

akao-amd added 2 commits February 26, 2026 13:29

[AMD] Merge Dockerfiles for ROCm

08f4f11

Refine according to review

a5cd2f1

akao-amd force-pushed the amd_merge_720_dockerfile branch from 525cbc6 to a5cd2f1 Compare February 26, 2026 05:29

minor fix

3d445b1

bingxche temporarily deployed to prod February 27, 2026 03:16 — with GitHub Actions Inactive

HaiShaw approved these changes Feb 27, 2026

View reviewed changes

HaiShaw merged commit 9b2fbf7 into sgl-project:main Feb 27, 2026
54 of 63 checks passed

magicYang1573 pushed a commit to magicYang1573/sglang that referenced this pull request Mar 9, 2026

[AMD] Merge Dockerfiles for ROCm (sgl-project#19203)

e1ecd69

Co-authored-by: bingxche <Bingxu.Chen@amd.com>

sammysun0711 pushed a commit to sammysun0711/sglang that referenced this pull request Mar 20, 2026

[AMD] Merge Dockerfiles for ROCm (sgl-project#19203)

2cdf24b

Co-authored-by: bingxche <Bingxu.Chen@amd.com>

Wangzheee pushed a commit to Wangzheee/sglang that referenced this pull request Mar 21, 2026

[AMD] Merge Dockerfiles for ROCm (sgl-project#19203)

e1fab2c

Co-authored-by: bingxche <Bingxu.Chen@amd.com>

JustinTong0323 pushed a commit to JustinTong0323/sglang that referenced this pull request Apr 7, 2026

[AMD] Merge Dockerfiles for ROCm (sgl-project#19203)

e2fc480

Co-authored-by: bingxche <Bingxu.Chen@amd.com>

vvagaytsev mentioned this pull request Apr 8, 2026

[AMD] Fix aiter import failure in ROCm Docker images #22363

Open

5 tasks

		fi \
		&& echo "export PYTHONPATH=/sgl-workspace/aiter:\${PYTHONPATH}" >> /etc/bash.bashrc

		@@ -117,17 +174,6 @@ RUN cd aiter \
		sh -c "GPU_ARCHS=$GPU_ARCH_LIST python setup.py develop"; \
		fi

Conversation

akao-amd commented Feb 24, 2026

Motivation

Modifications

Accuracy Tests

Benchmarking and Profiling

Checklist

Review Process

Uh oh!

gemini-code-assist bot commented Feb 24, 2026

Summary of Changes

Highlights

Footnotes

Uh oh!

gemini-code-assist bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

Uh oh!

gemini-code-assist bot Feb 24, 2026

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

HaiShaw commented Feb 26, 2026

Uh oh!

HaiShaw left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

akao-amd commented Feb 26, 2026

Uh oh!

bingxche commented Feb 27, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants