Skip to content

Commit bd631b4

Browse files
thxCodethxCode
committed
refactor: polish cuda/rocm pack
Signed-off-by: thxCode <thxcode0824@gmail.com>
1 parent d93aa8f commit bd631b4

File tree

8 files changed

+303
-6
lines changed

8 files changed

+303
-6
lines changed

‎pack/.post_operation/20251219_rocm_install_petit_kernel/matrix.yaml‎

Lines changed: 52 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,52 @@
1+
rules:
2+
3+
#
4+
# AMD ROCm
5+
#
6+
7+
#
8+
# AMD ROCm
9+
#
10+
11+
## Packed AMD ROCm 7.0.
12+
##
13+
- backend: "rocm"
14+
services:
15+
- "vllm"
16+
- "sglang"
17+
platforms:
18+
- "linux/amd64"
19+
args:
20+
- "ROCM_VERSION=7.0"
21+
- "VLLM_VERSION=0.12.0"
22+
- "SGLANG_VERSION=0.5.6.post2"
23+
- backend: "rocm"
24+
services:
25+
- "vllm"
26+
platforms:
27+
- "linux/amd64"
28+
args:
29+
- "ROCM_VERSION=7.0"
30+
- "VLLM_VERSION=0.11.2"
31+
## Packed AMD ROCm 6.4.
32+
##
33+
- backend: "rocm"
34+
services:
35+
- "vllm"
36+
- "sglang"
37+
platforms:
38+
- "linux/amd64"
39+
args:
40+
- "ROCM_VERSION=6.4"
41+
- "VLLM_VERSION=0.12.0"
42+
- "SGLANG_VERSION=0.5.6.post2"
43+
- backend: "rocm"
44+
services:
45+
- "vllm"
46+
- "sglang"
47+
platforms:
48+
- "linux/amd64"
49+
args:
50+
- "ROCM_VERSION=6.4"
51+
- "VLLM_VERSION=0.11.2"
52+
- "SGLANG_VERSION=0.5.5.post3"

‎pack/.post_operation/20251219_rocm_install_petit_kernel/rocm/Dockerfile‎

Lines changed: 61 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,61 @@
1+
ARG CMAKE_MAX_JOBS
2+
ARG ROCM_VERSION=6.4
3+
ARG VLLM_VERSION=0.12.0
4+
ARG SGLANG_VERSION=0.5.6.post2
5+
6+
FROM gpustack/runner:rocm${ROCM_VERSION}-vllm${VLLM_VERSION} AS vllm
7+
SHELL ["/bin/bash", "-eo", "pipefail", "-c"]
8+
9+
ARG TARGETPLATFORM
10+
ARG TARGETOS
11+
ARG TARGETARCH
12+
13+
## Petit Kernel
14+
15+
RUN <<EOF
16+
# Petit Kernel
17+
18+
# Install
19+
uv pip install petit-kernel
20+
21+
# Review
22+
uv pip tree
23+
24+
# Cleanup
25+
rm -rf /var/tmp/* \
26+
&& rm -rf /tmp/*
27+
EOF
28+
29+
## Entrypoint
30+
31+
WORKDIR /
32+
ENTRYPOINT [ "tini", "--" ]
33+
34+
35+
FROM gpustack/runner:rocm${ROCM_VERSION}-sglang${SGLANG_VERSION} AS sglang
36+
SHELL ["/bin/bash", "-eo", "pipefail", "-c"]
37+
38+
ARG TARGETPLATFORM
39+
ARG TARGETOS
40+
ARG TARGETARCH
41+
42+
## Petit Kernel
43+
44+
RUN <<EOF
45+
# Petit Kernel
46+
47+
# Install
48+
uv pip install petit-kernel
49+
50+
# Review
51+
uv pip tree
52+
53+
# Cleanup
54+
rm -rf /var/tmp/* \
55+
&& rm -rf /tmp/*
56+
EOF
57+
58+
## Entrypoint
59+
60+
WORKDIR /
61+
ENTRYPOINT [ "tini", "--" ]

‎pack/.post_operation/20251219_vllm_install_audio_extra/cuda/Dockerfile‎

Lines changed: 38 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,38 @@
1+
ARG CMAKE_MAX_JOBS
2+
ARG CUDA_VERSION=12.8
3+
ARG VLLM_VERSION=0.12.0
4+
5+
FROM gpustack/runner:cuda${CUDA_VERSION}-vllm${VLLM_VERSION} AS vllm
6+
SHELL ["/bin/bash", "-eo", "pipefail", "-c"]
7+
8+
ARG TARGETPLATFORM
9+
ARG TARGETOS
10+
ARG TARGETARCH
11+
12+
## Audio Extras
13+
14+
RUN <<EOF
15+
# Audio Extras
16+
17+
# Install
18+
cat <<EOT >/tmp/requirements.txt
19+
# audio extras
20+
librosa
21+
soundfile
22+
mistral_common[audio]
23+
EOT
24+
uv pip install \
25+
-r /tmp/requirements.txt
26+
27+
# Review
28+
uv pip tree
29+
30+
# Cleanup
31+
rm -rf /var/tmp/* \
32+
&& rm -rf /tmp/*
33+
EOF
34+
35+
## Entrypoint
36+
37+
WORKDIR /
38+
ENTRYPOINT [ "tini", "--" ]

‎pack/.post_operation/20251219_vllm_install_audio_extra/matrix.yaml‎

Lines changed: 89 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,89 @@
1+
rules:
2+
3+
#
4+
# NVIDIA CUDA
5+
#
6+
7+
## Packed NVIDIA CUDA 12.9.
8+
##
9+
- backend: "cuda"
10+
services:
11+
- "vllm"
12+
args:
13+
- "CUDA_VERSION=12.9"
14+
- "VLLM_VERSION=0.12.0"
15+
- backend: "cuda"
16+
services:
17+
- "vllm"
18+
args:
19+
- "CUDA_VERSION=12.9"
20+
- "VLLM_VERSION=0.11.2"
21+
## Packed NVIDIA CUDA 12.8.
22+
##
23+
- backend: "cuda"
24+
services:
25+
- "vllm"
26+
args:
27+
- "CUDA_VERSION=12.8"
28+
- "VLLM_VERSION=0.12.0"
29+
- backend: "cuda"
30+
services:
31+
- "vllm"
32+
args:
33+
- "CUDA_VERSION=12.8"
34+
- "VLLM_VERSION=0.11.2"
35+
## Packed NVIDIA CUDA 12.6.
36+
##
37+
- backend: "cuda"
38+
services:
39+
- "vllm"
40+
args:
41+
- "CUDA_VERSION=12.6"
42+
- "VLLM_VERSION=0.12.0"
43+
- backend: "cuda"
44+
services:
45+
- "vllm"
46+
args:
47+
- "CUDA_VERSION=12.6"
48+
- "VLLM_VERSION=0.11.2"
49+
50+
#
51+
# AMD ROCm
52+
#
53+
54+
## Packed AMD ROCm 7.0.
55+
##
56+
- backend: "rocm"
57+
services:
58+
- "vllm"
59+
platforms:
60+
- "linux/amd64"
61+
args:
62+
- "ROCM_VERSION=7.0"
63+
- "VLLM_VERSION=0.12.0"
64+
- backend: "rocm"
65+
services:
66+
- "vllm"
67+
platforms:
68+
- "linux/amd64"
69+
args:
70+
- "ROCM_VERSION=7.0"
71+
- "VLLM_VERSION=0.11.2"
72+
## Packed AMD ROCm 6.4.
73+
##
74+
- backend: "rocm"
75+
services:
76+
- "vllm"
77+
platforms:
78+
- "linux/amd64"
79+
args:
80+
- "ROCM_VERSION=6.4"
81+
- "VLLM_VERSION=0.12.0"
82+
- backend: "rocm"
83+
services:
84+
- "vllm"
85+
platforms:
86+
- "linux/amd64"
87+
args:
88+
- "ROCM_VERSION=6.4"
89+
- "VLLM_VERSION=0.11.2"

‎pack/.post_operation/20251219_vllm_install_audio_extra/rocm/Dockerfile‎

Lines changed: 38 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,38 @@
1+
ARG CMAKE_MAX_JOBS
2+
ARG ROCM_VERSION=6.4
3+
ARG VLLM_VERSION=0.12.0
4+
5+
FROM gpustack/runner:rocm${ROCM_VERSION}-vllm${VLLM_VERSION} AS vllm
6+
SHELL ["/bin/bash", "-eo", "pipefail", "-c"]
7+
8+
ARG TARGETPLATFORM
9+
ARG TARGETOS
10+
ARG TARGETARCH
11+
12+
## Audio Extras
13+
14+
RUN <<EOF
15+
# Audio Extras
16+
17+
# Install
18+
cat <<EOT >/tmp/requirements.txt
19+
# audio extras
20+
librosa
21+
soundfile
22+
mistral_common[audio]
23+
EOT
24+
uv pip install \
25+
-r /tmp/requirements.txt
26+
27+
# Review
28+
uv pip tree
29+
30+
# Cleanup
31+
rm -rf /var/tmp/* \
32+
&& rm -rf /tmp/*
33+
EOF
34+
35+
## Entrypoint
36+
37+
WORKDIR /
38+
ENTRYPOINT [ "tini", "--" ]

‎pack/.post_operation/README.md‎

Lines changed: 8 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -19,13 +19,15 @@ We leverage the matrix expansion feature of GPUStack Runner to achieve this, and
1919
- [x] 2025-11-05: Polish NVIDIA HPC-X configuration for CUDA released images.
2020
- [x] 2025-11-06: Install EP kernel for CUDA released images.
2121
- [x] 2025-11-07: Reinstall `lmcache` package for vLLM 0.11.0 CUDA released images.
22-
- [x] 2025-11-10: Install diffusion extension package for SGLang 0.5.5 CUDA released images.
23-
- [x] 2025-11-12: Install FlashAttention package for SGLang 0.5.5 CUDA released images.
24-
- [x] 2025-11-25: Install Posix IPC package for MindIE 2.2.rc1 CANN released images.
22+
- [x] 2025-11-10: Install `sglang[diffusion]` package for SGLang 0.5.5 CUDA released images.
23+
- [x] 2025-11-12: Install `FlashAttention` package for SGLang 0.5.5 CUDA released images.
24+
- [x] 2025-11-25: Install `Posix IPC` package for MindIE 2.2.rc1 CANN released images.
2525
- [x] 2025-12-01: Apply Qwen2.5 VL patches to vLLM 0.11.2 for CUDA released images.
26-
- [x] 2025-12-09: Install AV package for MindIE 2.2.rc1 and 2.1.rc2 CANN released images.
27-
- [x] 2025-12-13: Apply MiniCPM Qwen2 V2 patches to MindIE 2.2.rc1 and 2.1.rc2 for CANN released images.
26+
- [x] 2025-12-09: Install `AV` package for MindIE 2.2.rc1/2.1.rc2 CANN released images.
27+
- [x] 2025-12-13: Apply MiniCPM Qwen2 V2 patches to MindIE 2.2.rc1/2.1.rc2 for CANN released images.
2828
- [x] 2025-12-13: Apply server args patches to SGLang 0.5.6.post2 for CUDA released images.
2929
- [x] 2025-12-14: Apply several patches to vLLM 0.12.0 and SGLang 0.5.6.post2 for CUDA released images.
3030
- [x] 2025-12-15: Apply several patches to vLLM 0.11.0 and SGLang 0.5.6.post2 for CANN released images.
31-
- [ ] 2025-12-16: Uninstall `runai-model-streamer` packages from SGLang 0.5.6.post2 for CUDA released images.
31+
- [x] 2025-12-16: Uninstall `runai-model-streamer` packages from SGLang 0.5.6.post2 for CUDA released images.
32+
- [x] 2025-12-19: Install `vLLM[audio]` packages for vLLM 0.12.0/0.11.2 of CUDA/ROCm released images.
33+
- [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.

‎pack/cuda/Dockerfile‎

Lines changed: 8 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1400,14 +1400,22 @@ RUN <<EOF
14001400
BITSANDBYTES_VERSION="0.42.0"
14011401
fi
14021402
cat <<EOT >/tmp/requirements.txt
1403+
# dockerfile extras
14031404
accelerate
14041405
hf_transfer
14051406
modelscope
14061407
bitsandbytes>=${BITSANDBYTES_VERSION}
14071408
timm>=1.0.17
14081409
boto3
1410+
1411+
# kvconnectors extras
14091412
nixl>=0.7.1
14101413
mooncake-transfer-engine==${VLLM_MOONCAKE_VERSION}
1414+
1415+
# audio extras
1416+
librosa
1417+
soundfile
1418+
mistral_common[audio]
14111419
EOT
14121420
uv pip install \
14131421
-r /tmp/requirements.txt

‎pack/rocm/Dockerfile‎

Lines changed: 9 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1074,6 +1074,7 @@ RUN <<EOF
10741074

10751075
# Install
10761076
cat <<EOT >/tmp/requirements.txt
1077+
# dockerfile extras
10771078
accelerate
10781079
hf_transfer
10791080
modelscope
@@ -1083,6 +1084,14 @@ awscrt==0.28.2
10831084
cufile-python==0.2.0
10841085
nvtx==0.2.13
10851086
sortedcontainers==2.4.0
1087+
1088+
# audio extras
1089+
librosa
1090+
soundfile
1091+
mistral_common[audio]
1092+
1093+
# petit extras
1094+
petit-kernel
10861095
EOT
10871096
uv pip install \
10881097
-r /tmp/requirements.txt

0 commit comments

Comments
 (0)