rocm/sgl-dev:v0.5.22-rocm101-mi45x-20261017

Manifest digest

sha256:c5af4d3a36ac0f6c03a4597239f005241fcd52894b24fd8333376e9055458edd

OS/ARCH

linux/amd64

Compressed size

11.58 GB

Last pushed

27 days by henryx

Type

Image

Manifest digest

sha256:c5af4d3a36ac0f6c03a4597239f005241fcd52894b24fd8333376e9055458edd

Image Layers

Image Layers

1ARG RELEASE0 B

2ARG LAUNCHPAD_BUILD_ARCH0 B

3LABEL org.opencontainers.image.version=24.040 B

4ADD file ... in / 28.39 MB

5CMD ["/bin/bash"]0 B

6ARG GPU_ARCH=gfx1250-rocm10100 B

7ARG ROCM1000_SDK_VERSION=10.0.00 B

8ARG ROCM1000_TORCH_VERSION=2.11.00 B

9ARG ROCM1000_TORCHVISION_VERSION=0.26.00 B

10ARG ROCM1000_TORCHAUDIO_VERSION=2.11.00 B

11ARG ROCM1000_TRITON_VERSION=3.8.0+git4cff872c0 B

12ARG ROCM1010_SDK_VERSION=10.1.00 B

13ARG ROCM1010_TORCH_VERSION=2.12.00 B

14ARG ROCM1010_TORCHVISION_VERSION=0.27.00 B

15ARG ROCM1010_TORCHAUDIO_VERSION=2.11.00 B

16ARG ROCM1010_TRITON_VERSION=3.8.0+git669b31ac0 B

17ARG ROCM_INDEX_URL=https://stable.repo.amd.com/rocm/whl-next/0 B

18ARG ROCM_DEVICE_ARCH_LIST=gfx942 gfx950 gfx12500 B

19RUN |13 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.0151.05 MB

20ENV VIRTUAL_ENV=/opt/venv0 B

21RUN |13 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.04 MB

22ENV PATH=/opt/venv/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin0 B

23RUN |13 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.06 MB

24RUN |13 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.02.52 GB

25RUN |13 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.01.79 GB

26RUN |13 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.032 B

27ENV ROCM_HOME=/opt/venv/lib/python3.12/site-packages/_rocm_sdk_devel0 B

28ENV ROCM_PATH=/opt/venv/lib/python3.12/site-packages/_rocm_sdk_devel0 B

29ENV CPATH=/opt/venv/lib/python3.12/site-packages/_rocm_sdk_devel/include0 B

30ENV LIBRARY_PATH=/opt/venv/lib/python3.12/site-packages/_rocm_sdk_devel/lib0 B

31ENV LD_LIBRARY_PATH=/opt/venv/lib/python3.12/site-packages/_rocm_sdk_devel/lib0 B

32RUN |13 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.01.22 KB

33RUN |13 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.0167 B

34RUN |13 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.0167 B

35ENV BUILD_VLLM=00 B

36ENV BUILD_TRITON=00 B

37ENV BUILD_LLVM=00 B

38ENV BUILD_AITER_ALL=10 B

39ENV BUILD_MOONCAKE=10 B

40ENV AITER_COMMIT_DEFAULT=2bb549102ba2d9e51da8ef21107361827075d8b60 B

41ENV HIPBLASLT_TENSILE_LIBPATH=/opt/venv/lib/python3.12/site-packages/_rocm_sdk_libraries/lib/hipblaslt/library/gfx12500 B

42ENV PIP_CONSTRAINT=/etc/sglang/constraints/torch-rocm.txt0 B

43RUN |13 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.0174 B

44ARG GPU_ARCH=gfx1250-rocm10100 B

45RUN |13 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.01.3 KB

46ENV GPU_ARCH=gfx1250-rocm10100 B

47ENV GPU_ARCH_LIST=gfx12500 B

48ENV PYTORCH_ROCM_ARCH=gfx942;gfx950;gfx12500 B

49ARG SGL_REPO=https://github.com/sgl-project/sglang.git0 B

50ARG SGL_DEFAULT=main0 B

51ARG SGL_BRANCH=025655068ca9d3c571200d0210221416445de6190 B

52ARG BRANCH_TYPE=remote0 B

53ARG SETUPTOOLS_SCM_PRETEND_VERSION=0.5.22.dev20261017+g025655068c0 B

54ARG TRITON_REPO=https://github.com/triton-lang/triton.git0 B

55ENV TRITON_COMMIT=0 B

56ARG TRITON_INDEX_URL=https://pypi.amd.com/triton/release/rocm-7.2.0/simple/0 B

57ARG TRITON_VERSION=3.7.0+amd.rocm7.2.0.git890024100 B

58ARG TRITON_KERNELS_VERSION=1.0.0+amd.rocm7.2.0.git890024100 B

59ARG TORCH_ROCM_INDEX_URL=https://download.pytorch.org/whl/rocm7.20 B

60ARG TORCH_ROCM_VERSION=2.11.0+rocm7.20 B

61ARG TORCHVISION_ROCM_VERSION=0.26.0+rocm7.20 B

62ARG TORCHAUDIO_ROCM_VERSION=2.11.0+rocm7.20 B

63ARG AITER_REPO=https://github.com/ROCm/aiter.git0 B

64ARG AITER_COMMIT=0 B

65ENV AITER_COMMIT=2bb549102ba2d9e51da8ef21107361827075d8b60 B

66ARG LLVM_REPO=https://github.com/jrbyrnes/llvm-project.git0 B

67ARG LLVM_BRANCH=MainOpSelV20 B

68ARG LLVM_COMMIT=6520ace8227ffe2728148d5f3b9872a870b0a5600 B

69ARG MOONCAKE_REPO=https://github.com/kvcache-ai/Mooncake.git0 B

70ARG MOONCAKE_COMMIT=01d1eb2a7ec37fd5e20a88573e9b4956e7846e9a0 B

71ARG TILELANG_REPO=https://github.com/tile-ai/tilelang.git0 B

72ARG TILELANG_COMMIT=a55a82302bf7f3c5af635b5c9146f728185cc9000 B

73ARG FHT_REPO=https://github.com/jeffdaily/fast-hadamard-transform.git0 B

74ARG FHT_BRANCH=rocm0 B

75ARG FHT_COMMIT=46efb7d776d38638fc39f3c803eaee3dd7016bd10 B

76ARG ENABLE_SGL_AMD_MONO=10 B

77ARG SGL_AMD_MONO_REPO=https://github.com/sgl-project/sgl-amd-mono.git0 B

78ARG SGL_AMD_MONO_COMMIT=815b104f483d180027054a8f44bad0c2a3f895fd0 B

79ARG ENABLE_MORI=00 B

80ARG NIC_BACKEND=none0 B

81ARG MORI_REPO=https://github.com/ROCm/mori.git0 B

82ARG MORI_COMMIT=f39fa765e897c95995957d0518908cc982a88fc20 B

83ARG ENABLE_NIXL=10 B

84ARG UCX_REPO=https://github.com/openucx/ucx.git0 B

85ARG UCX_BRANCH=v1.19.x0 B

86ARG NIXL_REPO=https://github.com/ai-dynamo/nixl.git0 B

87ARG NIXL_COMMIT=c28061f9782e099f975bcc79198b7b5a1a36cc400 B

88ARG AINIC_VERSION=1.117.5-a-380 B

89ARG UBUNTU_CODENAME=jammy0 B

90ARG UBUNTU_MIRROR=https://archive.ubuntu.com0 B

91USER root0 B

92RUN |53 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.01.34 KB

93RUN |53 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.032 B

94RUN |53 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.039.27 KB

95RUN |53 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.032 B

96RUN |53 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.01.17 MB

97RUN |53 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.032 B

98RUN |53 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.0224 B

99WORKDIR /sgl-workspace103 B

100RUN |53 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.032 B

101ENV SETUPTOOLS_SCM_PRETEND_VERSION=0 B

102ENV AITER_USE_SYSTEM_TRITON=10 B

103RUN |53 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.032 B

104RUN |53 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.0732.74 MB

105

COPY /src/python/sglang/kernels/ops/attention/dsv4/asm/gfx950/mla_v4/mla_a8w8_qh64_qseqlen1_gqaratio64_nm.co /sgl-workspace/aiter/hsa/gfx950/mla_v4/ # buildkit

16.91 KB

106RUN |53 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.0765.45 KB

107RUN |53 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.01.82 KB

108ENV PATH=/opt/venv/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/local/go/bin0 B

109RUN |53 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.01.76 GB

110ARG BUILD_TYPE=all0 B

111ARG SETUPTOOLS_SCM_PRETEND_VERSION=0.5.22.dev20261017+g025655068c0 B

112RUN |54 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.014.66 MB

113ENV PATH=/root/.cargo/bin:/opt/venv/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/local/go/bin0 B

114RUN |54 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.0295.11 MB

115ENV CARGO_BUILD_JOBS=40 B

116RUN |54 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.032 B

117COPY /src /tmp/local_src # buildkit472.39 MB

118RUN |54 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.0487.08 MB

119RUN |54 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.0200 B

120RUN |54 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.0822.09 MB

121RUN |54 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.0179 B

122RUN |54 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.0129 B

123RUN |54 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.032 B

124RUN |54 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.01.34 GB

125ENV DEBIAN_FRONTEND=noninteractive0 B

126ENV LIBGL_ALWAYS_INDIRECT=10 B

127RUN |54 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.0236 B

128RUN |54 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.01006.9 MB

129RUN |54 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.019.71 MB

130RUN |54 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.032 B

131RUN |54 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.06.86 MB

132RUN |54 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.032 B

133RUN |54 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.0216.37 MB

134ARG TORCH_ROCM_FILE=torch-2.9.1+rocm7.2.0.lw.git7e1940d4-cp310-cp310-linux_x86_64.whl0 B

135RUN |55 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.0926 B

136RUN |55 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.032 B

137RUN |55 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.032 B

138RUN |55 GPU_ARCH=gfx1250-rocm1010 ROCM1000_SDK_VERSION=10.0.0 ROCM1000_TORCH_VERSION=2.11.032 B

139ENV SGLANG_DISABLE_CUDNN_CHECK=10 B

140ENV HIP_FORCE_DEV_KERNARG=10 B

141ENV HSA_NO_SCRATCH_RECLAIM=10 B

142ENV SGLANG_ALLOW_OVERWRITE_LONGER_CONTEXT_LEN=10 B

143ENV SGLANG_INT4_WEIGHT=00 B

144ENV SGLANG_MOE_PADDING=10 B

145ENV SGLANG_ROCM_DISABLE_LINEARQUANT=00 B

146ENV SGLANG_ROCM_FUSED_DECODE_MLA=10 B

147ENV SGLANG_SET_CPU_AFFINITY=10 B

148ENV SGLANG_USE_AITER=10 B

149ENV SGLANG_USE_ROCM700A=10 B

150ENV NCCL_MIN_NCHANNELS=1120 B

151ENV ROCM_QUICK_REDUCE_QUANTIZATION=INT80 B

152ENV TORCHINDUCTOR_MAX_AUTOTUNE=10 B

153ENV TORCHINDUCTOR_MAX_AUTOTUNE_POINTWISE=10 B

154CMD ["/bin/bash"]0 B

Command

 ARG RELEASE