gpustack-runner 0.1.24.post3__tar.gz → 0.1.25__tar.gz

{gpustack_runner-0.1.24.post3 → gpustack_runner-0.1.25}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: gpustack-runner
-Version: 0.1.24.post3
+Version: 0.1.25
 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`                          | `0.5.2`, `0.5.1.post3` |
+| 8.2 (910B)                   | `2.1.rc2` | `0.10.2`, `0.10.0`,  <br/>`0.9.2` | `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                         | `0.14.1`, **`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` |
+| CUDA Version <br/> (Variant) | vLLM                                                                 | SGLang                                              | VoxBox   |
+|------------------------------|----------------------------------------------------------------------|-----------------------------------------------------|----------|
+| 12.9                         | `0.15.0`, `0.14.1`, <br/>`0.13.0`, `0.12.0`, <br/>`0.11.2`           | `0.5.8`, `0.5.7`, <br/>`0.5.6.post2`                |          |
+| 12.8                         | `0.15.0`, `0.14.1`, <br/>`0.13.0`, `0.12.0`, <br/>`0.11.2`, `0.10.2` | `0.5.8`, `0.5.7`, <br/>`0.5.6.post2`, `0.5.5.post3` | `0.0.21` |
+| 12.6                         | `0.15.0`, `0.14.1`, <br/>`0.13.0`, `0.12.0`, <br/>`0.11.2`, `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                          | `0.15.0`, `0.14.1`, <br/>`0.13.0`, `0.12.0`, <br/>`0.11.2`           | `0.5.8`, `0.5.7`, <br/>`0.5.6.post2`                |
+| 6.4                          | `0.15.0`, `0.14.1`, <br/>`0.13.0`, `0.12.0`, <br/>`0.11.2`, `0.10.2` | `0.5.8`, `0.5.7`, <br/>`0.5.6.post2`, `0.5.5.post3` |
 
 ## Directory Structure
 

{gpustack_runner-0.1.24.post3 → gpustack_runner-0.1.25}/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`                          | `0.5.2`, `0.5.1.post3` |
+| 8.2 (910B)                   | `2.1.rc2` | `0.10.2`, `0.10.0`,  <br/>`0.9.2` | `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                         | `0.14.1`, **`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` |
+| CUDA Version <br/> (Variant) | vLLM                                                                 | SGLang                                              | VoxBox   |
+|------------------------------|----------------------------------------------------------------------|-----------------------------------------------------|----------|
+| 12.9                         | `0.15.0`, `0.14.1`, <br/>`0.13.0`, `0.12.0`, <br/>`0.11.2`           | `0.5.8`, `0.5.7`, <br/>`0.5.6.post2`                |          |
+| 12.8                         | `0.15.0`, `0.14.1`, <br/>`0.13.0`, `0.12.0`, <br/>`0.11.2`, `0.10.2` | `0.5.8`, `0.5.7`, <br/>`0.5.6.post2`, `0.5.5.post3` | `0.0.21` |
+| 12.6                         | `0.15.0`, `0.14.1`, <br/>`0.13.0`, `0.12.0`, <br/>`0.11.2`, `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                          | `0.15.0`, `0.14.1`, <br/>`0.13.0`, `0.12.0`, <br/>`0.11.2`           | `0.5.8`, `0.5.7`, <br/>`0.5.6.post2`                |
+| 6.4                          | `0.15.0`, `0.14.1`, <br/>`0.13.0`, `0.12.0`, <br/>`0.11.2`, `0.10.2` | `0.5.8`, `0.5.7`, <br/>`0.5.6.post2`, `0.5.5.post3` |
 
 ## Directory Structure
 

{gpustack_runner-0.1.24.post3 → gpustack_runner-0.1.25}/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.post3'
-__version_tuple__ = version_tuple = (0, 1, 24, 'post3')
+__version__ = version = '0.1.25'
+__version_tuple__ = version_tuple = (0, 1, 25)
 try:
     from ._version_appendix import git_commit
     __commit_id__ = commit_id = git_commit

gpustack_runner-0.1.25/gpustack_runner/_version_appendix.py

@@ -0,0 +1 @@
+git_commit = "b005327"

{gpustack_runner-0.1.24.post3 → gpustack_runner-0.1.25}/gpustack_runner/runner.py.json

@@ -868,6 +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": "0.15.0",
+    "platform": "linux/amd64",
+    "docker_image": "gpustack/runner:cuda12.9-vllm0.15.0",
+    "deprecated": false
+  },
+  {
+    "backend": "cuda",
+    "backend_version": "12.9",
+    "original_backend_version": "12.9.1",
+    "backend_variant": "",
+    "service": "vllm",
+    "service_version": "0.15.0",
+    "platform": "linux/arm64",
+    "docker_image": "gpustack/runner:cuda12.9-vllm0.15.0",
+    "deprecated": false
+  },
   {
     "backend": "cuda",
     "backend_version": "12.9",
@@ -1077,6 +1099,28 @@
     "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": "0.15.0",
+    "platform": "linux/amd64",
+    "docker_image": "gpustack/runner:cuda12.8-vllm0.15.0",
+    "deprecated": false
+  },
+  {
+    "backend": "cuda",
+    "backend_version": "12.8",
+    "original_backend_version": "12.8.1",
+    "backend_variant": "",
+    "service": "vllm",
+    "service_version": "0.15.0",
+    "platform": "linux/arm64",
+    "docker_image": "gpustack/runner:cuda12.8-vllm0.15.0",
+    "deprecated": false
+  },
   {
     "backend": "cuda",
     "backend_version": "12.8",
@@ -1297,6 +1341,28 @@
     "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": "0.15.0",
+    "platform": "linux/amd64",
+    "docker_image": "gpustack/runner:cuda12.6-vllm0.15.0",
+    "deprecated": false
+  },
+  {
+    "backend": "cuda",
+    "backend_version": "12.6",
+    "original_backend_version": "12.6.3",
+    "backend_variant": "",
+    "service": "vllm",
+    "service_version": "0.15.0",
+    "platform": "linux/arm64",
+    "docker_image": "gpustack/runner:cuda12.6-vllm0.15.0",
+    "deprecated": false
+  },
   {
     "backend": "cuda",
     "backend_version": "12.6",
@@ -1748,6 +1814,17 @@
     "docker_image": "gpustack/runner:musa4.1-vllm0.9.2",
     "deprecated": false
   },
+  {
+    "backend": "rocm",
+    "backend_version": "7.0",
+    "original_backend_version": "7.0.2",
+    "backend_variant": "",
+    "service": "sglang",
+    "service_version": "0.5.8",
+    "platform": "linux/amd64",
+    "docker_image": "gpustack/runner:rocm7.0-sglang0.5.8",
+    "deprecated": false
+  },
   {
     "backend": "rocm",
     "backend_version": "7.0",
@@ -1770,6 +1847,28 @@
     "docker_image": "gpustack/runner:rocm7.0-sglang0.5.6.post2",
     "deprecated": false
   },
+  {
+    "backend": "rocm",
+    "backend_version": "7.0",
+    "original_backend_version": "7.0.2",
+    "backend_variant": "",
+    "service": "vllm",
+    "service_version": "0.15.0",
+    "platform": "linux/amd64",
+    "docker_image": "gpustack/runner:rocm7.0-vllm0.15.0",
+    "deprecated": false
+  },
+  {
+    "backend": "rocm",
+    "backend_version": "7.0",
+    "original_backend_version": "7.0.2",
+    "backend_variant": "",
+    "service": "vllm",
+    "service_version": "0.14.1",
+    "platform": "linux/amd64",
+    "docker_image": "gpustack/runner:rocm7.0-vllm0.14.1",
+    "deprecated": false
+  },
   {
     "backend": "rocm",
     "backend_version": "7.0",
@@ -1814,6 +1913,17 @@
     "docker_image": "gpustack/runner:rocm7.0-vllm0.11.0",
     "deprecated": true
   },
+  {
+    "backend": "rocm",
+    "backend_version": "6.4",
+    "original_backend_version": "6.4.4",
+    "backend_variant": "",
+    "service": "sglang",
+    "service_version": "0.5.8",
+    "platform": "linux/amd64",
+    "docker_image": "gpustack/runner:rocm6.4-sglang0.5.8",
+    "deprecated": false
+  },
   {
     "backend": "rocm",
     "backend_version": "6.4",
@@ -1847,6 +1957,28 @@
     "docker_image": "gpustack/runner:rocm6.4-sglang0.5.5.post3",
     "deprecated": false
   },
+  {
+    "backend": "rocm",
+    "backend_version": "6.4",
+    "original_backend_version": "6.4.4",
+    "backend_variant": "",
+    "service": "vllm",
+    "service_version": "0.15.0",
+    "platform": "linux/amd64",
+    "docker_image": "gpustack/runner:rocm6.4-vllm0.15.0",
+    "deprecated": false
+  },
+  {
+    "backend": "rocm",
+    "backend_version": "6.4",
+    "original_backend_version": "6.4.4",
+    "backend_variant": "",
+    "service": "vllm",
+    "service_version": "0.14.1",
+    "platform": "linux/amd64",
+    "docker_image": "gpustack/runner:rocm6.4-vllm0.14.1",
+    "deprecated": false
+  },
   {
     "backend": "rocm",
     "backend_version": "6.4",

gpustack_runner-0.1.25/pack/.post_operation/20260203_cuda_several_patches/cuda/Dockerfile

@@ -0,0 +1,77 @@
+ARG CMAKE_MAX_JOBS
+ARG CUDA_VERSION=12.8
+ARG VLLM_VERSION=0.14.1
+ARG SGLANG_VERSION=0.5.8
+
+FROM gpustack/runner:cuda${CUDA_VERSION}-vllm${VLLM_VERSION} AS vllm
+SHELL ["/bin/bash", "-eo", "pipefail", "-c"]
+
+ARG TARGETPLATFORM
+ARG TARGETOS
+ARG TARGETARCH
+
+## Update CuDNN and NCCL packages
+
+RUN <<EOF
+    # Update CuDNN and NCCL packages
+
+    IFS="." read -r CUDA_MAJOR CUDA_MINOR CUDA_PATCH <<< "${VLLM_TORCH_CUDA_VERSION}"
+
+    # Install
+    cat <<EOT >/tmp/requirements.txt
+nvidia-cudnn-cu${CUDA_MAJOR}>=9.16.0.29
+nvidia-cudnn-frontend>=1.17.0
+nvidia-nccl-cu${CUDA_MAJOR}>=2.28.3
+EOT
+    uv pip install \
+        -r /tmp/requirements.txt
+
+    # Review
+    uv pip tree
+
+    # Cleanup
+    rm -rf /var/tmp/* \
+        && rm -rf /tmp/*
+EOF
+
+## Entrypoint
+
+WORKDIR /
+ENTRYPOINT [ "tini", "--" ]
+
+
+FROM gpustack/runner:cuda${CUDA_VERSION}-sglang${SGLANG_VERSION} AS sglang
+SHELL ["/bin/bash", "-eo", "pipefail", "-c"]
+
+ARG TARGETPLATFORM
+ARG TARGETOS
+ARG TARGETARCH
+
+## Update CuDNN and NCCL packages
+
+RUN <<EOF
+    # Update CuDNN and NCCL packages
+
+    IFS="." read -r CUDA_MAJOR CUDA_MINOR CUDA_PATCH <<< "${VLLM_TORCH_CUDA_VERSION}"
+
+    # Install
+    cat <<EOT >/tmp/requirements.txt
+nvidia-cudnn-cu${CUDA_MAJOR}>=9.16.0.29
+nvidia-cudnn-frontend>=1.17.0
+nvidia-nccl-cu${CUDA_MAJOR}>=2.28.3
+EOT
+    uv pip install \
+        -r /tmp/requirements.txt
+
+    # Review
+    uv pip tree
+
+    # Cleanup
+    rm -rf /var/tmp/* \
+        && rm -rf /tmp/*
+EOF
+
+## Entrypoint
+
+WORKDIR /
+ENTRYPOINT [ "tini", "--" ]

gpustack_runner-0.1.25/pack/.post_operation/20260203_cuda_several_patches/matrix.yaml

@@ -0,0 +1,22 @@
+rules:
+
+  #
+  # NVIDIA CUDA
+  #
+
+  ## Packed NVIDIA CUDA 12.9.
+  ##
+  - backend: "cuda"
+    services:
+      - "vllm"
+    args:
+      - "CUDA_VERSION=12.9"
+      - "VLLM_VERSION=0.15.0"
+  - backend: "cuda"
+    services:
+      - "vllm"
+      - "sglang"
+    args:
+      - "CUDA_VERSION=12.9"
+      - "VLLM_VERSION=0.14.1"
+      - "SGLANG_VERSION=0.5.8"

gpustack_runner-0.1.25/pack/.post_operation/20260203_sglang_disable_cudnn_check/cuda/Dockerfile

@@ -0,0 +1,17 @@
+ARG CMAKE_MAX_JOBS
+ARG CUDA_VERSION=12.8
+ARG SGLANG_VERSION=0.5.8
+
+FROM gpustack/runner:cuda${CUDA_VERSION}-sglang${SGLANG_VERSION} AS sglang
+SHELL ["/bin/bash", "-eo", "pipefail", "-c"]
+
+ARG TARGETPLATFORM
+ARG TARGETOS
+ARG TARGETARCH
+
+## Entrypoint
+
+ENV SGLANG_DISABLE_CUDNN_CHECK=1
+
+WORKDIR /
+ENTRYPOINT [ "tini", "--" ]

gpustack_runner-0.1.25/pack/.post_operation/20260203_sglang_disable_cudnn_check/matrix.yaml

@@ -0,0 +1,56 @@
+rules:
+
+  #
+  # NVIDIA CUDA
+  #
+
+  ## Packed NVIDIA CUDA 12.9.
+  ##
+  - backend: "cuda"
+    services:
+      - "sglang"
+    args:
+      - "CUDA_VERSION=12.9"
+      - "VLLM_VERSION=0.14.1"
+      - "SGLANG_VERSION=0.5.8"
+
+  #
+  # AMD ROCm
+  #
+
+  ## Packed ROCm 7.0.
+  ##
+  - backend: "rocm"
+    services:
+      - "sglang"
+    platforms:
+      - "linux/amd64"
+    args:
+      - "ROCM_VERSION=7.0"
+      - "SGLANG_VERSION=0.5.8"
+  - backend: "rocm"
+    services:
+      - "sglang"
+    platforms:
+      - "linux/amd64"
+    args:
+      - "ROCM_VERSION=7.0"
+      - "SGLANG_VERSION=0.5.7"
+  ## Packed ROCm 6.4.
+  ##
+  - backend: "rocm"
+    services:
+      - "sglang"
+    platforms:
+      - "linux/amd64"
+    args:
+      - "ROCM_VERSION=6.4"
+      - "SGLANG_VERSION=0.5.8"
+  - backend: "rocm"
+    services:
+      - "sglang"
+    platforms:
+      - "linux/amd64"
+    args:
+      - "ROCM_VERSION=6.4"
+      - "SGLANG_VERSION=0.5.7"

gpustack_runner-0.1.25/pack/.post_operation/20260203_sglang_disable_cudnn_check/rocm/Dockerfile

@@ -0,0 +1,17 @@
+ARG CMAKE_MAX_JOBS
+ARG ROCM_VERSION=7.0
+ARG SGLANG_VERSION=0.5.8
+
+FROM gpustack/runner:rocm${ROCM_VERSION}-sglang${SGLANG_VERSION} AS sglang
+SHELL ["/bin/bash", "-eo", "pipefail", "-c"]
+
+ARG TARGETPLATFORM
+ARG TARGETOS
+ARG TARGETARCH
+
+## Entrypoint
+
+ENV SGLANG_DISABLE_CUDNN_CHECK=1
+
+WORKDIR /
+ENTRYPOINT [ "tini", "--" ]

{gpustack_runner-0.1.24.post3 → gpustack_runner-0.1.25}/pack/.post_operation/README.md

@@ -35,3 +35,5 @@ We leverage the matrix expansion feature of GPUStack Runner to achieve this, and
 - [ ] 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.
+- [x] 2026-02-03: Apply several patches to vLLM 0.14.1/0.15.0 and SGLang 0.5.8 for CUDA 12.9 released images.
+- [x] 2026-02-03: Patch SGLang 0.5.8/0.5.7 of CUDA/ROCm released images to disable CuDNN version check.

{gpustack_runner-0.1.24.post3 → gpustack_runner-0.1.25}/pack/cann/Dockerfile

@@ -59,7 +59,7 @@ ARG VLLM_VERSION=0.14.1
 ARG VLLM_ASCEND_VERSION=0.14.0rc1
 ARG VLLM_TORCH_VERSION=2.9.0
 ARG VLLM_MOONCAKE_VERSION=0.3.7.post2
-ARG VLLM_OMNI_COMMIT=e8aa32b
+ARG VLLM_OMNI_COMMIT=de2cac9
 ARG SGLANG_BASE_IMAGE=gpustack/runner:cann${CANN_VERSION}-${CANN_ARCHS}-python${PYTHON_VERSION}
 ARG SGLANG_VERSION=0.5.8
 ARG SGLANG_TORCH_VERSION=2.8.0
@@ -865,6 +865,15 @@ RUN --mount=type=bind,from=vllm-build-omni,source=/,target=/omni,rw <<EOF
     uv pip install --no-build-isolation \
         /omni/workspace/*.whl
 
+    # Dependencies
+    uv pip uninstall onnxruntime || true
+    cat <<EOT >/tmp/requirements.txt
+onnxruntime-cann
+sox
+EOT
+    uv pip install \
+        -r /tmp/requirements.txt
+
     # Cleanup
     rm -rf /var/tmp/* \
         && rm -rf /tmp/*
@@ -956,7 +965,11 @@ RUN --mount=type=bind,target=/workspace,rw <<EOF
 
     tree -hs /workspace/patches
     pushd $(pip show vllm | grep Location: | cut -d" " -f 2) \
-        && patch -p1 < /workspace/patches/vllm_*.patch
+        && patch -p1 < /workspace/patches/vllm/*.patch
+    if pip show vllm_omni > /dev/null 2>&1; then \
+        pushd $(pip show vllm_omni | grep Location: | cut -d" " -f 2) \
+            && patch -p1 < /workspace/patches/vllm_omni/*.patch; \
+    fi
 EOF
 
 ## Entrypoint

gpustack_runner-0.1.25/pack/cann/patches/vllm_omni/001_wrong_patch.patch

@@ -0,0 +1,13 @@
+diff --git a/vllm_omni/patch.py b/vllm_omni/patch.py
+index 687ff51..6b67924 100644
+--- a/vllm_omni/patch.py
++++ b/vllm_omni/patch.py
+@@ -19,6 +19,8 @@ for module_name, module in sys.modules.items():
+     # only do patch on module of vllm, pass others
+     if "vllm" not in module_name:
+         continue
++    if "--omni" not in sys.argv:
++        continue
+     if hasattr(module, "EngineCoreOutput") and module.EngineCoreOutput == _OriginalEngineCoreOutput:
+         module.EngineCoreOutput = OmniEngineCoreOutput
+     if hasattr(module, "EngineCoreOutputs") and module.EngineCoreOutputs == _OriginalEngineCoreOutputs: