gpustack-runner 0.1.24.post2__tar.gz → 0.1.24.post3__tar.gz

{gpustack_runner-0.1.24.post2 → gpustack_runner-0.1.24.post3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: gpustack-runner
-Version: 0.1.24.post2
+Version: 0.1.24.post3
 Summary: GPUStack Runner is library for registering runnable accelerated backends and services in GPUStack.
 Project-URL: Homepage, https://github.com/gpustack/runner
 Project-URL: Bug Tracker, https://github.com/gpustack/gpustack/issues
@@ -52,17 +52,17 @@ The following table lists the supported accelerated backends and their correspon
     vllm-ascend [#3316](https://github.com/vllm-project/vllm-ascend/issues/3316)
     and [#2795](https://github.com/vllm-project/vllm-ascend/issues/2795).
 
-| CANN Version <br/> (Variant) | MindIE    | vLLM                                                       | SGLang                 |
-|------------------------------|-----------|------------------------------------------------------------|------------------------|
-| 8.5 (A3/910C)                | `2.3.0`   | `0.14.1`, `0.13.0`                                         | `0.5.8`                |
-| 8.5 (910B)                   | `2.3.0`   | `0.14.1`, `0.13.0`                                         | `0.5.8`                |
-| 8.5 (310P)                   | `2.3.0`   | `0.14.1`                                                   |                        |
-| 8.3 (A3/910C)                | `2.2.rc1` | `0.12.0`, `0.11.0`                                         | `0.5.7`, `0.5.6.post2` |
-| 8.3 (910B)                   | `2.2.rc1` | `0.12.0`, `0.11.0`                                         | `0.5.7`, `0.5.6.post2` |
-| 8.3 (310P)                   | `2.2.rc1` |                                                            |                        |
-| 8.2 (A3/910C)                | `2.1.rc2` | `0.10.2`, `0.10.1.1`                                       | `0.5.2`, `0.5.1.post3` |
-| 8.2 (910B)                   | `2.1.rc2` | `0.10.2`, `0.10.1.1`, <br/>`0.10.0`, `0.9.2`, <br/>`0.9.1` | `0.5.2`, `0.5.1.post3` |
-| 8.2 (310P)                   | `2.1.rc2` | `0.10.0`, `0.9.2`                                          |                        |
+| CANN Version <br/> (Variant) | MindIE    | vLLM                                                               | SGLang                 |
+|------------------------------|-----------|--------------------------------------------------------------------|------------------------|
+| 8.5 (A3/910C)                | `2.3.0`   | `0.14.1`, `0.13.0`                                                 | `0.5.8`                |
+| 8.5 (910B)                   | `2.3.0`   | `0.14.1`, `0.13.0`                                                 | `0.5.8`                |
+| 8.5 (310P)                   | `2.3.0`   | `0.14.1`                                                           |                        |
+| 8.3 (A3/910C)                | `2.2.rc1` | `0.12.0`, `0.11.0`                                                 | `0.5.7`, `0.5.6.post2` |
+| 8.3 (910B)                   | `2.2.rc1` | `0.12.0`, `0.11.0`                                                 | `0.5.7`, `0.5.6.post2` |
+| 8.3 (310P)                   | `2.2.rc1` |                                                                    |                        |
+| 8.2 (A3/910C)                | `2.1.rc2` | `0.10.2`, ~~`0.10.1.1`~~                                           | `0.5.2`, `0.5.1.post3` |
+| 8.2 (910B)                   | `2.1.rc2` | `0.10.2`, ~~`0.10.1.1`~~, <br/>`0.10.0`, `0.9.2`, <br/>~~`0.9.1`~~ | `0.5.2`, `0.5.1.post3` |
+| 8.2 (310P)                   | `2.1.rc2` | `0.10.0`, `0.9.2`                                                  |                        |
 
 ### Iluvatar CoreX
 
@@ -80,11 +80,11 @@ The following table lists the supported accelerated backends and their correspon
 > - CUDA 12.6/12.4 supports Compute Capabilities:
     `7.5 8.0+PTX 8.9 9.0+PTX`.
 
-| CUDA Version <br/> (Variant) | vLLM                                        | SGLang                                                    | VoxBox   |
-|------------------------------|---------------------------------------------|-----------------------------------------------------------|----------|
-| 12.9                         | `0.13.0`, `0.12.0`, <br/>`0.11.2`           | `0.5.7`, `0.5.6.post2`                                    |          |
-| 12.8                         | `0.13.0`, `0.12.0`, <br/>`0.11.2`, `0.10.2` | `0.5.7`, `0.5.6.post2`, <br/>`0.5.5.post3`, `0.5.4.post3` | `0.0.21` |
-| 12.6                         | `0.13.0`, `0.12.0`, <br/>`0.11.2`,`0.10.2`  |                                                           | `0.0.21` |
+| CUDA Version <br/> (Variant) | vLLM                                                           | SGLang                                                                      | VoxBox   |
+|------------------------------|----------------------------------------------------------------|-----------------------------------------------------------------------------|----------|
+| 12.9                         | `0.14.1`, **`0.13.0`**, <br/>`0.12.0`, `0.11.2`                | `0.5.8`, `0.5.7`, <br/>`0.5.6.post2`                                        |          |
+| 12.8                         | `0.14.1`, **`0.13.0`**, <br/>`0.12.0`, `0.11.2`, <br/>`0.10.2` | `0.5.8`, `0.5.7`, <br/>`0.5.6.post2`, `0.5.5.post3`, <br/>~~`0.5.4.post3`~~ | `0.0.21` |
+| 12.6                         | `0.14.1`, **`0.13.0`**, <br/>`0.12.0`, `0.11.2`, <br/>`0.10.2` |                                                                             | `0.0.21` |
 
 ### Hygon DTK
 
@@ -128,10 +128,10 @@ The following table lists the supported accelerated backends and their correspon
 > - ROCm 6.4 SGLang supports `gfx942` only.
 > - ROCm 7.0 SGLang supports `gfx950` only.
 
-| ROCm Version <br/> (Variant) | vLLM                                        | SGLang                                     |
-|------------------------------|---------------------------------------------|--------------------------------------------|
-| 7.0                          | `0.13.0`, `0.12.0`, <br/>`0.11.2`           | `0.5.7`, `0.5.6.post2`                     |
-| 6.4                          | `0.13.0`, `0.12.0`, <br/>`0.11.2`, `0.10.2` | `0.5.7`, `0.5.6.post2`, <br/>`0.5.5.post3` |
+| ROCm Version <br/> (Variant) | vLLM                                            | SGLang                                     |
+|------------------------------|-------------------------------------------------|--------------------------------------------|
+| 7.0                          | **`0.13.0`**, `0.12.0`, <br/>`0.11.2`           | `0.5.7`, `0.5.6.post2`                     |
+| 6.4                          | **`0.13.0`**, `0.12.0`, <br/>`0.11.2`, `0.10.2` | `0.5.7`, `0.5.6.post2`, <br/>`0.5.5.post3` |
 
 ## Directory Structure
 

{gpustack_runner-0.1.24.post2 → gpustack_runner-0.1.24.post3}/README.md

@@ -32,17 +32,17 @@ The following table lists the supported accelerated backends and their correspon
     vllm-ascend [#3316](https://github.com/vllm-project/vllm-ascend/issues/3316)
     and [#2795](https://github.com/vllm-project/vllm-ascend/issues/2795).
 
-| CANN Version <br/> (Variant) | MindIE    | vLLM                                                       | SGLang                 |
-|------------------------------|-----------|------------------------------------------------------------|------------------------|
-| 8.5 (A3/910C)                | `2.3.0`   | `0.14.1`, `0.13.0`                                         | `0.5.8`                |
-| 8.5 (910B)                   | `2.3.0`   | `0.14.1`, `0.13.0`                                         | `0.5.8`                |
-| 8.5 (310P)                   | `2.3.0`   | `0.14.1`                                                   |                        |
-| 8.3 (A3/910C)                | `2.2.rc1` | `0.12.0`, `0.11.0`                                         | `0.5.7`, `0.5.6.post2` |
-| 8.3 (910B)                   | `2.2.rc1` | `0.12.0`, `0.11.0`                                         | `0.5.7`, `0.5.6.post2` |
-| 8.3 (310P)                   | `2.2.rc1` |                                                            |                        |
-| 8.2 (A3/910C)                | `2.1.rc2` | `0.10.2`, `0.10.1.1`                                       | `0.5.2`, `0.5.1.post3` |
-| 8.2 (910B)                   | `2.1.rc2` | `0.10.2`, `0.10.1.1`, <br/>`0.10.0`, `0.9.2`, <br/>`0.9.1` | `0.5.2`, `0.5.1.post3` |
-| 8.2 (310P)                   | `2.1.rc2` | `0.10.0`, `0.9.2`                                          |                        |
+| CANN Version <br/> (Variant) | MindIE    | vLLM                                                               | SGLang                 |
+|------------------------------|-----------|--------------------------------------------------------------------|------------------------|
+| 8.5 (A3/910C)                | `2.3.0`   | `0.14.1`, `0.13.0`                                                 | `0.5.8`                |
+| 8.5 (910B)                   | `2.3.0`   | `0.14.1`, `0.13.0`                                                 | `0.5.8`                |
+| 8.5 (310P)                   | `2.3.0`   | `0.14.1`                                                           |                        |
+| 8.3 (A3/910C)                | `2.2.rc1` | `0.12.0`, `0.11.0`                                                 | `0.5.7`, `0.5.6.post2` |
+| 8.3 (910B)                   | `2.2.rc1` | `0.12.0`, `0.11.0`                                                 | `0.5.7`, `0.5.6.post2` |
+| 8.3 (310P)                   | `2.2.rc1` |                                                                    |                        |
+| 8.2 (A3/910C)                | `2.1.rc2` | `0.10.2`, ~~`0.10.1.1`~~                                           | `0.5.2`, `0.5.1.post3` |
+| 8.2 (910B)                   | `2.1.rc2` | `0.10.2`, ~~`0.10.1.1`~~, <br/>`0.10.0`, `0.9.2`, <br/>~~`0.9.1`~~ | `0.5.2`, `0.5.1.post3` |
+| 8.2 (310P)                   | `2.1.rc2` | `0.10.0`, `0.9.2`                                                  |                        |
 
 ### Iluvatar CoreX
 
@@ -60,11 +60,11 @@ The following table lists the supported accelerated backends and their correspon
 > - CUDA 12.6/12.4 supports Compute Capabilities:
     `7.5 8.0+PTX 8.9 9.0+PTX`.
 
-| CUDA Version <br/> (Variant) | vLLM                                        | SGLang                                                    | VoxBox   |
-|------------------------------|---------------------------------------------|-----------------------------------------------------------|----------|
-| 12.9                         | `0.13.0`, `0.12.0`, <br/>`0.11.2`           | `0.5.7`, `0.5.6.post2`                                    |          |
-| 12.8                         | `0.13.0`, `0.12.0`, <br/>`0.11.2`, `0.10.2` | `0.5.7`, `0.5.6.post2`, <br/>`0.5.5.post3`, `0.5.4.post3` | `0.0.21` |
-| 12.6                         | `0.13.0`, `0.12.0`, <br/>`0.11.2`,`0.10.2`  |                                                           | `0.0.21` |
+| CUDA Version <br/> (Variant) | vLLM                                                           | SGLang                                                                      | VoxBox   |
+|------------------------------|----------------------------------------------------------------|-----------------------------------------------------------------------------|----------|
+| 12.9                         | `0.14.1`, **`0.13.0`**, <br/>`0.12.0`, `0.11.2`                | `0.5.8`, `0.5.7`, <br/>`0.5.6.post2`                                        |          |
+| 12.8                         | `0.14.1`, **`0.13.0`**, <br/>`0.12.0`, `0.11.2`, <br/>`0.10.2` | `0.5.8`, `0.5.7`, <br/>`0.5.6.post2`, `0.5.5.post3`, <br/>~~`0.5.4.post3`~~ | `0.0.21` |
+| 12.6                         | `0.14.1`, **`0.13.0`**, <br/>`0.12.0`, `0.11.2`, <br/>`0.10.2` |                                                                             | `0.0.21` |
 
 ### Hygon DTK
 
@@ -108,10 +108,10 @@ The following table lists the supported accelerated backends and their correspon
 > - ROCm 6.4 SGLang supports `gfx942` only.
 > - ROCm 7.0 SGLang supports `gfx950` only.
 
-| ROCm Version <br/> (Variant) | vLLM                                        | SGLang                                     |
-|------------------------------|---------------------------------------------|--------------------------------------------|
-| 7.0                          | `0.13.0`, `0.12.0`, <br/>`0.11.2`           | `0.5.7`, `0.5.6.post2`                     |
-| 6.4                          | `0.13.0`, `0.12.0`, <br/>`0.11.2`, `0.10.2` | `0.5.7`, `0.5.6.post2`, <br/>`0.5.5.post3` |
+| ROCm Version <br/> (Variant) | vLLM                                            | SGLang                                     |
+|------------------------------|-------------------------------------------------|--------------------------------------------|
+| 7.0                          | **`0.13.0`**, `0.12.0`, <br/>`0.11.2`           | `0.5.7`, `0.5.6.post2`                     |
+| 6.4                          | **`0.13.0`**, `0.12.0`, <br/>`0.11.2`, `0.10.2` | `0.5.7`, `0.5.6.post2`, <br/>`0.5.5.post3` |
 
 ## Directory Structure
 

{gpustack_runner-0.1.24.post2 → gpustack_runner-0.1.24.post3}/gpustack_runner/_version.py

@@ -27,8 +27,8 @@ version_tuple: VERSION_TUPLE
 __commit_id__: COMMIT_ID
 commit_id: COMMIT_ID
 
-__version__ = version = '0.1.24.post2'
-__version_tuple__ = version_tuple = (0, 1, 24, 'post2')
+__version__ = version = '0.1.24.post3'
+__version_tuple__ = version_tuple = (0, 1, 24, 'post3')
 try:
     from ._version_appendix import git_commit
     __commit_id__ = commit_id = git_commit

gpustack_runner-0.1.24.post3/gpustack_runner/_version_appendix.py

@@ -0,0 +1 @@
+git_commit = "dc41ed2"

{gpustack_runner-0.1.24.post2 → gpustack_runner-0.1.24.post3}/gpustack_runner/runner.py.json

@@ -261,7 +261,7 @@
     "service_version": "0.10.1.1",
     "platform": "linux/amd64",
     "docker_image": "gpustack/runner:cann8.2-a3-vllm0.10.1.1",
-    "deprecated": false
+    "deprecated": true
   },
   {
     "backend": "cann",
@@ -272,7 +272,7 @@
     "service_version": "0.10.1.1",
     "platform": "linux/arm64",
     "docker_image": "gpustack/runner:cann8.2-a3-vllm0.10.1.1",
-    "deprecated": false
+    "deprecated": true
   },
   {
     "backend": "cann",
@@ -558,7 +558,7 @@
     "service_version": "0.10.1.1",
     "platform": "linux/amd64",
     "docker_image": "gpustack/runner:cann8.2-910b-vllm0.10.1.1",
-    "deprecated": false
+    "deprecated": true
   },
   {
     "backend": "cann",
@@ -569,7 +569,7 @@
     "service_version": "0.10.1.1",
     "platform": "linux/arm64",
     "docker_image": "gpustack/runner:cann8.2-910b-vllm0.10.1.1",
-    "deprecated": false
+    "deprecated": true
   },
   {
     "backend": "cann",
@@ -624,7 +624,7 @@
     "service_version": "0.9.1",
     "platform": "linux/amd64",
     "docker_image": "gpustack/runner:cann8.2-910b-vllm0.9.1",
-    "deprecated": false
+    "deprecated": true
   },
   {
     "backend": "cann",
@@ -635,7 +635,7 @@
     "service_version": "0.9.1",
     "platform": "linux/arm64",
     "docker_image": "gpustack/runner:cann8.2-910b-vllm0.9.1",
-    "deprecated": false
+    "deprecated": true
   },
   {
     "backend": "cann",
@@ -802,6 +802,28 @@
     "docker_image": "gpustack/runner:corex4.2-vllm0.8.3",
     "deprecated": false
   },
+  {
+    "backend": "cuda",
+    "backend_version": "12.9",
+    "original_backend_version": "12.9.1",
+    "backend_variant": "",
+    "service": "sglang",
+    "service_version": "0.5.8",
+    "platform": "linux/amd64",
+    "docker_image": "gpustack/runner:cuda12.9-sglang0.5.8",
+    "deprecated": false
+  },
+  {
+    "backend": "cuda",
+    "backend_version": "12.9",
+    "original_backend_version": "12.9.1",
+    "backend_variant": "",
+    "service": "sglang",
+    "service_version": "0.5.8",
+    "platform": "linux/arm64",
+    "docker_image": "gpustack/runner:cuda12.9-sglang0.5.8",
+    "deprecated": false
+  },
   {
     "backend": "cuda",
     "backend_version": "12.9",
@@ -846,6 +868,28 @@
     "docker_image": "gpustack/runner:cuda12.9-sglang0.5.6.post2",
     "deprecated": false
   },
+  {
+    "backend": "cuda",
+    "backend_version": "12.9",
+    "original_backend_version": "12.9.1",
+    "backend_variant": "",
+    "service": "vllm",
+    "service_version": "0.14.1",
+    "platform": "linux/amd64",
+    "docker_image": "gpustack/runner:cuda12.9-vllm0.14.1",
+    "deprecated": false
+  },
+  {
+    "backend": "cuda",
+    "backend_version": "12.9",
+    "original_backend_version": "12.9.1",
+    "backend_variant": "",
+    "service": "vllm",
+    "service_version": "0.14.1",
+    "platform": "linux/arm64",
+    "docker_image": "gpustack/runner:cuda12.9-vllm0.14.1",
+    "deprecated": false
+  },
   {
     "backend": "cuda",
     "backend_version": "12.9",
@@ -912,6 +956,28 @@
     "docker_image": "gpustack/runner:cuda12.9-vllm0.11.2",
     "deprecated": false
   },
+  {
+    "backend": "cuda",
+    "backend_version": "12.8",
+    "original_backend_version": "12.8.1",
+    "backend_variant": "",
+    "service": "sglang",
+    "service_version": "0.5.8",
+    "platform": "linux/amd64",
+    "docker_image": "gpustack/runner:cuda12.8-sglang0.5.8",
+    "deprecated": false
+  },
+  {
+    "backend": "cuda",
+    "backend_version": "12.8",
+    "original_backend_version": "12.8.1",
+    "backend_variant": "",
+    "service": "sglang",
+    "service_version": "0.5.8",
+    "platform": "linux/arm64",
+    "docker_image": "gpustack/runner:cuda12.8-sglang0.5.8",
+    "deprecated": false
+  },
   {
     "backend": "cuda",
     "backend_version": "12.8",
@@ -998,7 +1064,7 @@
     "service_version": "0.5.4.post3",
     "platform": "linux/amd64",
     "docker_image": "gpustack/runner:cuda12.8-sglang0.5.4.post3",
-    "deprecated": false
+    "deprecated": true
   },
   {
     "backend": "cuda",
@@ -1009,6 +1075,28 @@
     "service_version": "0.5.4.post3",
     "platform": "linux/arm64",
     "docker_image": "gpustack/runner:cuda12.8-sglang0.5.4.post3",
+    "deprecated": true
+  },
+  {
+    "backend": "cuda",
+    "backend_version": "12.8",
+    "original_backend_version": "12.8.1",
+    "backend_variant": "",
+    "service": "vllm",
+    "service_version": "0.14.1",
+    "platform": "linux/amd64",
+    "docker_image": "gpustack/runner:cuda12.8-vllm0.14.1",
+    "deprecated": false
+  },
+  {
+    "backend": "cuda",
+    "backend_version": "12.8",
+    "original_backend_version": "12.8.1",
+    "backend_variant": "",
+    "service": "vllm",
+    "service_version": "0.14.1",
+    "platform": "linux/arm64",
+    "docker_image": "gpustack/runner:cuda12.8-vllm0.14.1",
     "deprecated": false
   },
   {
@@ -1209,6 +1297,28 @@
     "docker_image": "gpustack/runner:cuda12.8-voxbox0.0.20",
     "deprecated": true
   },
+  {
+    "backend": "cuda",
+    "backend_version": "12.6",
+    "original_backend_version": "12.6.3",
+    "backend_variant": "",
+    "service": "vllm",
+    "service_version": "0.14.1",
+    "platform": "linux/amd64",
+    "docker_image": "gpustack/runner:cuda12.6-vllm0.14.1",
+    "deprecated": false
+  },
+  {
+    "backend": "cuda",
+    "backend_version": "12.6",
+    "original_backend_version": "12.6.3",
+    "backend_variant": "",
+    "service": "vllm",
+    "service_version": "0.14.1",
+    "platform": "linux/arm64",
+    "docker_image": "gpustack/runner:cuda12.6-vllm0.14.1",
+    "deprecated": false
+  },
   {
     "backend": "cuda",
     "backend_version": "12.6",

gpustack_runner-0.1.24.post3/pack/.post_operation/20260129_sglang_reinstall_kernel/cann/Dockerfile

@@ -0,0 +1,74 @@
+ARG CMAKE_MAX_JOBS
+ARG CANN_VERSION=8.3
+ARG CANN_ARCHS=910b
+ARG SGLANG_VERSION=0.12.0
+ARG SGLANG_KERNEL_VERSION=20251206
+
+FROM gpustack/runner:cann${CANN_VERSION}-${CANN_ARCHS}-sglang${SGLANG_VERSION} AS sglang
+SHELL ["/bin/bash", "-eo", "pipefail", "-c"]
+
+ARG TARGETPLATFORM
+ARG TARGETOS
+ARG TARGETARCH
+
+## Reinstall SGLang Kernel
+
+ARG CMAKE_MAX_JOBS
+ARG SGLANG_VERSION
+ARG SGLANG_KERNEL_VERSION
+
+ENV SGLANG_VERSION=${SGLANG_VERSION} \
+    SGLANG_KERNEL_VERSION=${SGLANG_KERNEL_VERSION}
+
+RUN <<EOF
+    # SGLang
+
+    CMAKE_MAX_JOBS="${CMAKE_MAX_JOBS}"
+    if [[ -z "${CMAKE_MAX_JOBS}" ]]; then
+        CMAKE_MAX_JOBS="$(( $(nproc) / 2 ))"
+    fi
+    if (( $(echo "${CMAKE_MAX_JOBS} > 8" | bc -l) )); then
+        CMAKE_MAX_JOBS="8"
+    fi
+    export MAX_JOBS="${CMAKE_MAX_JOBS}"
+    export COMPILE_CUSTOM_KERNELS=1
+    export LD_LIBRARY_PATH="${LD_LIBRARY_PATH}:${CANN_HOME}/ascend-toolkit/latest/$(uname -i)-linux/devlib"
+    export LD_LIBRARY_PATH="${CANN_HOME}/ascend-toolkit/latest/runtime/lib64/stub:${LD_LIBRARY_PATH}"
+    source ${CANN_HOME}/ascend-toolkit/set_env.sh
+    echo "Building SGLang with the following environment variables:"
+    env
+
+    # Install Dependencies
+    cat <<EOT >/tmp/requirements.txt
+attrs==25.4.0
+decorator==5.2.1
+psutil==7.1.3
+pyyaml==6.0.3
+triton-ascend==3.2.0
+EOT
+    uv pip install \
+        -r /tmp/requirements.txt
+
+    # Build and Install SGLang Kernel
+    git -C /tmp clone --recursive --shallow-submodules \
+        --depth 1 --branch ${SGLANG_KERNEL_VERSION} --single-branch \
+        https://github.com/sgl-project/sgl-kernel-npu.git sgl-kernel-npu
+    unset ASCEND_HOME_PATH
+    pushd /tmp/sgl-kernel-npu \
+        && ./build.sh \
+        && tree -hs /tmp/sgl-kernel-npu/output \
+        && uv pip install /tmp/sgl-kernel-npu/output/deep_ep*.whl /tmp/sgl-kernel-npu/output/sgl_kernel_npu*.whl
+
+    # Postprocess SGLang Kernel (DeepEP)
+    cd "$(pip show deep-ep | awk '/^Location:/ {print $2}')" && ln -sf deep_ep/deep_ep_cpp*.so
+
+    # Cleanup
+    rm -rf /var/tmp/* \
+        && rm -rf /tmp/* \
+        && ccache --clear --clean
+EOF
+
+## Entrypoint
+
+WORKDIR /
+ENTRYPOINT [ "tini", "--" ]

gpustack_runner-0.1.24.post3/pack/.post_operation/20260129_sglang_reinstall_kernel/matrix.yaml

@@ -0,0 +1,28 @@
+rules:
+
+  #
+  # Ascend CANN
+  #
+
+  ## Packed Ascend CANN 8.3, using CANN Kernel for A3.
+  ##
+  - backend: "cann"
+    services:
+      - "sglang"
+    platforms:
+      - "linux/arm64"
+    args:
+      - "CANN_VERSION=8.3"
+      - "CANN_ARCHS=a3"
+      - "SGLANG_VERSION=0.5.7"
+  ## Packed Ascend CANN 8.3, using CANN Kernel for 910B.
+  ##
+  - backend: "cann"
+    services:
+      - "sglang"
+    platforms:
+      - "linux/arm64"
+    args:
+      - "CANN_VERSION=8.3"
+      - "CANN_ARCHS=910b"
+      - "SGLANG_VERSION=0.5.7"

gpustack_runner-0.1.24.post3/pack/.post_operation/20260129_vllm_patch_dp/cuda/Dockerfile

@@ -0,0 +1,25 @@
+ARG CMAKE_MAX_JOBS
+ARG CUDA_VERSION=12.8
+ARG VLLM_VERSION=0.13.0
+
+FROM gpustack/runner:cuda${CUDA_VERSION}-vllm${VLLM_VERSION} AS vllm
+SHELL ["/bin/bash", "-eo", "pipefail", "-c"]
+
+ARG TARGETPLATFORM
+ARG TARGETOS
+ARG TARGETARCH
+
+## Patch
+
+RUN --mount=type=bind,target=/workspace,rw <<EOF
+    # Patch
+
+    tree -hs /workspace/patches
+    pushd $(pip show vllm | grep Location: | cut -d" " -f 2) \
+        && patch -p1 < /workspace/patches/vllm_*.patch
+EOF
+
+## Entrypoint
+
+WORKDIR /
+ENTRYPOINT [ "tini", "--" ]

gpustack_runner-0.1.24.post3/pack/.post_operation/20260129_vllm_patch_dp/cuda/patches/vllm_001_wrong_dp_ray.patch

@@ -0,0 +1,41 @@
+diff --git a/vllm/utils/network_utils.py b/vllm/utils/network_utils.py
+index 7d01533cb..311ed44df 100644
+--- a/vllm/utils/network_utils.py
++++ b/vllm/utils/network_utils.py
+@@ -147,6 +147,9 @@ def get_open_zmq_inproc_path() -> str:
+     return f"inproc://{uuid4()}"
+ 
+ 
++_next_port: int | None = None
++
++
+ def get_open_port() -> int:
+     """
+     Get an open port for the vLLM process to listen on.
+@@ -163,7 +166,7 @@ def get_open_port() -> int:
+             candidate_port = _get_open_port()
+             if candidate_port not in reserved_port_range:
+                 return candidate_port
+-    return _get_open_port()
++    return _get_open_port(_next_port)
+ 
+ 
+ def get_open_ports_list(count: int = 5) -> list[int]:
+@@ -174,13 +177,15 @@ def get_open_ports_list(count: int = 5) -> list[int]:
+     return list(ports)
+ 
+ 
+-def _get_open_port() -> int:
+-    port = envs.VLLM_PORT
++def _get_open_port(start: int | None = None) -> int:
++    port = start or envs.VLLM_PORT
+     if port is not None:
++        global _next_port
+         while True:
+             try:
+                 with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
+                     s.bind(("", port))
++                    _next_port = port + 1
+                     return port
+             except OSError:
+                 port += 1  # Increment port number if already in use

gpustack_runner-0.1.24.post3/pack/.post_operation/20260129_vllm_patch_dp/matrix.yaml

@@ -0,0 +1,55 @@
+rules:
+
+  #
+  # NVIDIA CUDA
+  #
+
+  ## Packed NVIDIA CUDA 12.9.
+  ##
+  - backend: "cuda"
+    services:
+      - "vllm"
+    args:
+      - "CUDA_VERSION=12.9"
+      - "VLLM_VERSION=0.13.0"
+  ## Packed NVIDIA CUDA 12.8.
+  ##
+  - backend: "cuda"
+    services:
+      - "vllm"
+    args:
+      - "CUDA_VERSION=12.8"
+      - "VLLM_VERSION=0.13.0"
+  ## Packed NVIDIA CUDA 12.6.
+  ##
+  - backend: "cuda"
+    services:
+      - "vllm"
+    args:
+      - "CUDA_VERSION=12.6"
+      - "VLLM_VERSION=0.13.0"
+
+  #
+  # AMD ROCm
+  #
+
+  ## Packed AMD ROCm 7.0.
+  ##
+  - backend: "rocm"
+    services:
+      - "vllm"
+    platforms:
+      - "linux/amd64"
+    args:
+      - "ROCM_VERSION=7.0"
+      - "VLLM_VERSION=0.13.0"
+  ## Packed AMD ROCm 6.4.
+  ##
+  - backend: "rocm"
+    services:
+      - "vllm"
+    platforms:
+      - "linux/amd64"
+    args:
+      - "ROCM_VERSION=6.4"
+      - "VLLM_VERSION=0.13.0"

gpustack_runner-0.1.24.post3/pack/.post_operation/20260129_vllm_patch_dp/rocm/Dockerfile

@@ -0,0 +1,25 @@
+ARG CMAKE_MAX_JOBS
+ARG ROCM_VERSION=6.4
+ARG VLLM_VERSION=0.13.0
+
+FROM gpustack/runner:rocm${ROCM_VERSION}-vllm${VLLM_VERSION} AS vllm
+SHELL ["/bin/bash", "-eo", "pipefail", "-c"]
+
+ARG TARGETPLATFORM
+ARG TARGETOS
+ARG TARGETARCH
+
+## Patch
+
+RUN --mount=type=bind,target=/workspace,rw <<EOF
+    # Patch
+
+    tree -hs /workspace/patches
+    pushd $(pip show vllm | grep Location: | cut -d" " -f 2) \
+        && patch -p1 < /workspace/patches/vllm_*.patch
+EOF
+
+## Entrypoint
+
+WORKDIR /
+ENTRYPOINT [ "tini", "--" ]

gpustack_runner-0.1.24.post3/pack/.post_operation/20260129_vllm_patch_dp/rocm/patches/vllm_001_wrong_dp_ray.patch

@@ -0,0 +1,41 @@
+diff --git a/vllm/utils/network_utils.py b/vllm/utils/network_utils.py
+index 7d01533cb..311ed44df 100644
+--- a/vllm/utils/network_utils.py
++++ b/vllm/utils/network_utils.py
+@@ -147,6 +147,9 @@ def get_open_zmq_inproc_path() -> str:
+     return f"inproc://{uuid4()}"
+ 
+ 
++_next_port: int | None = None
++
++
+ def get_open_port() -> int:
+     """
+     Get an open port for the vLLM process to listen on.
+@@ -163,7 +166,7 @@ def get_open_port() -> int:
+             candidate_port = _get_open_port()
+             if candidate_port not in reserved_port_range:
+                 return candidate_port
+-    return _get_open_port()
++    return _get_open_port(_next_port)
+ 
+ 
+ def get_open_ports_list(count: int = 5) -> list[int]:
+@@ -174,13 +177,15 @@ def get_open_ports_list(count: int = 5) -> list[int]:
+     return list(ports)
+ 
+ 
+-def _get_open_port() -> int:
+-    port = envs.VLLM_PORT
++def _get_open_port(start: int | None = None) -> int:
++    port = start or envs.VLLM_PORT
+     if port is not None:
++        global _next_port
+         while True:
+             try:
+                 with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
+                     s.bind(("", port))
++                    _next_port = port + 1
+                     return port
+             except OSError:
+                 port += 1  # Increment port number if already in use

{gpustack_runner-0.1.24.post2 → gpustack_runner-0.1.24.post3}/pack/.post_operation/README.md

@@ -33,3 +33,5 @@ We leverage the matrix expansion feature of GPUStack Runner to achieve this, and
 - [x] 2025-12-19: Install `petit-kernel` package for vLLM 0.12.0/0.11.2 and SGLang 0.5.6.post2/0.5.5.post3 of ROcm released images.
 - [x] 2025-12-24: Apply ATB config patches to MindIE 2.2.rc1 for CANN released images.
 - [ ] 2026-01-05: Install `vllm-omni` packages for vLLM 0.12.0 of CUDA/ROCm/CANN released images.
+- [x] 2026-01-29: Apply DP deployment patches to vLLM 0.13.0 for CUDA/ROCm released images.
+- [x] 2026-01-29: Reinstall SGLang Kernel for SGLang 0.5.7 of CANN released images.