feat(pack): add Enflame GCU (燧原 S60) backend support by huqingjie0118-droid · Pull Request #248 · gpustack/runner

huqingjie0118-droid · 2026-07-15T08:55:21Z

feat(pack): add Enflame GCU (燧原 S60) backend support

Summary

Add Dockerfile support for Enflame GCU (燧原 S60) to the GPUStack runner pack system.

Changes

New directory: pack/gcu/
- Dockerfile — Combined build with vanilla → runtime → vllm stages
- Dockerfile.vllm — Per-service vLLM Dockerfile (runtime + vllm targets)
Modified: pack/matrix.yaml — Added gcu backend rule

Version Information

Component	Version
TopsRider (燧原加速框架)	3.7.107 (i3x 3.6+)
vLLM-GCU	0.11.0
Python	3.10
OS	Ubuntu 22.04
Hardware	S60 GCU

Vanilla Image

registry-egc.enflame-tech.com/artifacts/vllm_gcu:v0.11.0-TR3.7.107-ubuntu2204

This image is provided by Enflame and includes:

Enflame GCU driver and TopsRider SDK
torch_gcu (PyTorch for GCU)
vllm-gcu (vLLM with GCU backend)
triton_gcu

Expected Image Tags

After the copy.yml workflow runs and the Dockerfiles are built:

Stage	Tag
Vanilla	`gpustack/runner:gcu3.7.107-python3.10-vanilla`
Runtime	`gpustack/runner:gcu3.7.107-python3.10`
vLLM	`gpustack/runner:gcu3.7.107-vllm0.11.0-python3.10`

copy.yml Parameters

For the GitHub Action copy.yml, the suggested inputs:

src-registry: registry-egc.enflame-tech.com
src-image: artifacts/vllm_gcu:v0.11.0-TR3.7.107-ubuntu2204
dst-image-tag: gcu3.7.107-python3.10

Build & Test

To build the runtime image locally:

docker build --progress=plain --platform=linux/amd64 \
  --file=pack/gcu/Dockerfile \
  --tag=gpustack/runner:gcu3.7.107-python3.10 \
  --target=runtime \
  pack/gcu

To build the vLLM image:

docker build --progress=plain --platform=linux/amd64 \
  --file=pack/gcu/Dockerfile \
  --tag=gpustack/runner:gcu3.7.107-vllm0.11.0-python3.10 \
  --target=vllm \
  pack/gcu

Notes

Only vLLM backend is supported (SGLang support not yet available for GCU)
The vanilla image is Ubuntu 22.04 based
Device visibility env var: GCU_VISIBLE_DEVICES
Uses RAY_EXPERIMENTAL_NOSET_GCU_VISIBLE_DEVICES=1 in entrypoint

📋 Version Information for GPUStack Team`n`nPlease copy the following vanilla image via the `copy.yml` workflow:`n`n- Source: `registry-egc.enflame-tech.com/artifacts/vllm_gcu:v0.11.0-TR3.7.107-ubuntu2204` `- Destination tag:` gcu3.7.107-python3.10-vanilla``n`n## Version Details`n`n| Component | Version |`n|-----------|---------|`n| TopsRider (燧原加速框架) | 3.7.107 |`n| vLLM-GCU | 0.11.0 |`n| Python | 3.10 |`n| OS | Ubuntu 22.04 |`n| Hardware | S60 GCU |`n`n## Registry Access`n`nThe source registry `registry-egc.enflame-tech.com` requires authentication.`nPlease contact support@enflame-tech.com for registry credentials if needed.`n`n---`n`nNote: Issues are disabled on this repo, so posting here.

gemini-code-assist

Code Review

This pull request introduces support for the Enflame GCU (燧原 S60) backend to the GPUStack runner pack system, adding Dockerfiles for runtime and vLLM stages, documentation, and matrix configuration. The reviewer provided several actionable suggestions to improve the Dockerfiles and repository structure. Key feedback includes fixing a critical issue in pack/gcu/Dockerfile where the vllm stage was built directly from the vanilla image instead of the runtime stage (which would cause missing dependencies like tini and uv), removing an accidentally committed PR_TEMPLATE.md file, renaming ISSUE_TEMPLATE.md to README.md for better documentation, adding retry flags to curl commands, using python3 -m pip instead of pip, ensuring consistent removal of pip.conf, and cleaning up /var/lib/apt/lists/* to reduce Docker image sizes.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

gemini-code-assist · 2026-07-15T08:57:15Z

+# Stage: vllm
+# ============================================================
+
+FROM ${VLLM_BASE_IMAGE} AS vllm


The vllm stage should be built on top of the runtime stage rather than VLLM_BASE_IMAGE (which defaults to the raw VANILLA_IMAGE). If built directly from the vanilla image, the final image will lack the essential tools installed in the runtime stage (such as tini, which is used as the entrypoint, and uv, which is used in the postprocess step). This will cause the build to fail at the uv pip tree step, or fail at runtime due to a missing tini executable.

FROM runtime AS vllm

gemini-code-assist · 2026-07-15T08:57:15Z

@@ -0,0 +1,86 @@
+# feat(pack): add Enflame GCU (燧原 S60) backend support


It appears that PR_TEMPLATE.md was accidentally committed to the root of the repository. This file contains the specific description for this Pull Request rather than a generic template. You should delete this file from the repository.

gemini-code-assist · 2026-07-15T08:57:15Z

@@ -0,0 +1,85 @@
+# Enflame GCU (燧原 S60) Backend Support Request


This file contains valuable documentation about the Enflame GCU backend, including supported models, build instructions, and expected tags. However, naming it ISSUE_TEMPLATE.md inside the pack/gcu/ directory is non-standard and confusing. It should be renamed to README.md to serve as the developer documentation for this backend.

gemini-code-assist · 2026-07-15T08:57:15Z

+    # Update alternatives
+    if [[ -f /etc/alternatives/python3 ]]; then update-alternatives --remove-all python3; fi; update-alternatives --install /usr/bin/python3 python3 /usr/bin/python${PYTHON_VERSION} 1
+    if [[ -f /etc/alternatives/python ]]; then update-alternatives --remove-all python; fi; update-alternatives --install /usr/bin/python python /usr/bin/python${PYTHON_VERSION} 1
+    curl -sS "https://bootstrap.pypa.io/get-pip.py" | python${PYTHON_VERSION}


Add retry flags to the curl command when downloading get-pip.py to prevent transient network issues from failing the build, matching the robust curl options used elsewhere in the Dockerfile.

curl -sS --retry 3 --retry-connrefused -fL "https://bootstrap.pypa.io/get-pip.py" | python${PYTHON_VERSION}

gemini-code-assist · 2026-07-15T08:57:16Z

+
+    if [[ -f /etc/alternatives/python3 ]]; then update-alternatives --remove-all python3; fi; update-alternatives --install /usr/bin/python3 python3 /usr/bin/python${PYTHON_VERSION} 1
+    if [[ -f /etc/alternatives/python ]]; then update-alternatives --remove-all python; fi; update-alternatives --install /usr/bin/python python /usr/bin/python${PYTHON_VERSION} 1
+    curl -sS "https://bootstrap.pypa.io/get-pip.py" | python${PYTHON_VERSION}


Add retry flags to the curl command when downloading get-pip.py to prevent transient network issues from failing the build.

curl -sS --retry 3 --retry-connrefused -fL "https://bootstrap.pypa.io/get-pip.py" | python${PYTHON_VERSION}

gemini-code-assist · 2026-07-15T08:57:16Z

+pipx<=1.8.0
+uv<=0.9.11
+EOT
+    pip install -r /tmp/requirements.txt


It is safer and more robust to use python3 -m pip instead of calling pip directly, ensuring packages are installed into the correct Python environment.

python3 -m pip install -r /tmp/requirements.txt

gemini-code-assist · 2026-07-15T08:57:16Z

+pipx<=1.8.0
+uv<=0.9.11
+EOT
+    pip install -r /tmp/requirements.txt


It is safer and more robust to use python3 -m pip instead of calling pip directly, ensuring packages are installed into the correct Python environment.

python3 -m pip install -r /tmp/requirements.txt

gemini-code-assist · 2026-07-15T08:57:16Z

+RUN <<EOF
+    # Buildkit
+
+    cat <<EOT >/tmp/requirements.txt


To ensure consistency with Dockerfile.vllm and prevent any pre-configured pip mirrors or settings in the base image from interfering with the build, remove any existing pip.conf files before installing requirements.

RUN <<EOF # Buildkit rm -rf /root/.pip/pip.conf /etc/pip.conf cat <<EOT >/tmp/requirements.txt

gemini-code-assist · 2026-07-15T08:57:16Z

+    # Cleanup
+    rm -rf /var/tmp/* \
+        && rm -rf /tmp/* \
+        && rm -rf /var/cache/apt


To minimize the Docker image size, it is highly recommended to clean up the apt package lists in /var/lib/apt/lists/* in addition to /var/cache/apt. Please apply this cleanup to all apt-get blocks in both Dockerfile and Dockerfile.vllm.

# Cleanup rm -rf /var/tmp/* \ && rm -rf /tmp/* \ && rm -rf /var/cache/apt \ && rm -rf /var/lib/apt/lists/*

huqingjieClark added 2 commits July 15, 2026 15:38

feat(pack): add Enflame GCU backend support

714e2a3

- Add pack/gcu/Dockerfile: combined vanilla/runtime/vllm stages - Add pack/gcu/Dockerfile.vllm: per-service vLLM Dockerfile - Update pack/matrix.yaml: add gcu backend rule - Version: TopsRider 3.7.107, vLLM 0.11.0, Python 3.10, Ubuntu 22.04

docs: add PR template for GCU backend

ec69451

gemini-code-assist Bot reviewed Jul 15, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

feat(pack): add Enflame GCU (燧原 S60) backend support#248

feat(pack): add Enflame GCU (燧原 S60) backend support#248
huqingjie0118-droid wants to merge 2 commits into
gpustack:mainfrom
huqingjie0118-droid:feat/add-gcu-backend

huqingjie0118-droid commented Jul 15, 2026

Uh oh!

huqingjie0118-droid commented Jul 15, 2026

Uh oh!

gemini-code-assist Bot left a comment

Uh oh!

gemini-code-assist Bot Jul 15, 2026

Uh oh!

gemini-code-assist Bot Jul 15, 2026

Uh oh!

gemini-code-assist Bot Jul 15, 2026

Uh oh!

gemini-code-assist Bot Jul 15, 2026

Uh oh!

gemini-code-assist Bot Jul 15, 2026

Uh oh!

gemini-code-assist Bot Jul 15, 2026

Uh oh!

gemini-code-assist Bot Jul 15, 2026

Uh oh!

gemini-code-assist Bot Jul 15, 2026

Uh oh!

gemini-code-assist Bot Jul 15, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

		@@ -0,0 +1,86 @@
		# feat(pack): add Enflame GCU (燧原 S60) backend support

		@@ -0,0 +1,85 @@
		# Enflame GCU (燧原 S60) Backend Support Request

Uh oh!

Conversation

huqingjie0118-droid commented Jul 15, 2026

feat(pack): add Enflame GCU (燧原 S60) backend support

Summary

Changes

Version Information

Vanilla Image

Expected Image Tags

copy.yml Parameters

Build & Test

Notes

Related

Uh oh!

huqingjie0118-droid commented Jul 15, 2026

Uh oh!

gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

gemini-code-assist Bot Jul 15, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot Jul 15, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot Jul 15, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot Jul 15, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot Jul 15, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot Jul 15, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot Jul 15, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot Jul 15, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot Jul 15, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants