DocsICP FilingConsole
官方文档
首页

Release notes for PyTorch images

更新时间：
复制 MD 格式
产品详情
我的收藏

PyTorch is an open-source machine learning framework that supports tensor operations and deep learning on CPUs and GPUs. PyTorch images provide pre-configured deep learning environments, so you can start training and inference workloads without manually installing dependencies.

Version overview

PyTorch version Image type CUDA / runtime Python Release date
2.9.0 GPU CUDA 13.0.2 3.12.7 December 2025
2.7.1.8 GPU CUDA 12.8.1 3.12.7 July 2025
2.5.1.6 AMD GPU ROCm 6.2.4 3.10.13 January 2025
2.5.1.6 GPU CUDA 12.1.1 3.10.13 February 2025
2.4.0 GPU CUDA 12.1.1 3.10.13 November 2024
2.3.0.3 CPU / GPU — / CUDA 12.1.1 / CUDA 11.8.0 3.10.13 July 2024
2.3.0 CPU / GPU — / CUDA 12.1.1 / CUDA 11.8.0 3.10.13 June 2024
2.2.0.1 CPU / GPU — / CUDA 12.1.1 / CUDA 11.8.0 3.10.13 March 2024
2.1.0 DCU DTK 24.04.1 3.10.13 August 2024
2.0.1 GPU CUDA 12.1.1 3.10.13 February 2024

PyTorch 2.9.0

Images

Image type Registry address
GPU ac2-registry.cn-hangzhou.cr.aliyuncs.com/ac2/pytorch:2.9.0.10-cuda13.0.2-py312-alinux3.2104

Image content

ac2-registry.cn-hangzhou.cr.aliyuncs.com/ac2/pytorch:2.9.0.10-cuda13.0.2-py312-alinux3.2104

CUDA libraries

  • cuda-cudart-13-0: 13.0.96

  • cuda-compat-13-0: 580.95.05

  • cuda-libraries-13-0: 13.0.2

  • cuda-opencl-13-0: 13.0.85

  • cuda-nvrtc-13-0: 13.0.88

  • cuda-nvtx-13-0: 13.0.85

  • cuda-cupti-13-0: 13.0.85

Acceleration libraries

  • libcublas-13-0: 13.1.0.3

  • libcufft-13-0: 12.0.0.61

  • libcufile-13-0: 1.15.1.6

  • libcurand-13-0: 10.4.0.35

  • libcusolver-13-0: 12.0.4.66

  • libcusparse-13-0: 12.6.3.3

  • libcusparselt0-cuda-13: 0.8.1.1

  • libnpp-13-0: 13.0.1.2

  • libnvjitlink-13-0: 13.0.88

  • libnvjpeg-13-0: 13.0.1.86

  • libnvfatbin-13-0: 13.0.85

  • libnccl: 2.27.7

  • libcudnn9-cuda-13: 9.14.0.64

Python packages

  • Python: 3.12.7

  • torch: 2.9.0+ali.10.cu130

  • torchaudio: 2.9.0+cu130

  • torchvision: 0.24.0+cu130

  • triton: 3.5.0

  • numpy: 2.1.3

  • pillow: 10.3.0

  • sympy: 1.13.3

  • mpmath: 1.3.0

  • networkx: 3.3

  • Jinja2: 3.1.4

  • MarkupSafe: 2.1.5

  • filelock: 3.17.0

  • fsspec: 2024.6.0

  • typing_extensions: 4.12.2

  • certifi: 2024.12.14

  • charset-normalizer: 3.3.2

  • idna: 3.7

  • requests: 2.32.4

  • urllib3: 2.5.0

  • PySocks: 1.7.1

  • pip: 25.3

  • setuptools: 80.9.0

  • wheel: 0.45.0

Driver requirements

Based on CUDA runtime 13.0.2. Requires NVIDIA Driver release 580 or later and is compatible with NVIDIA Driver releases R535, R550, R565, R570, and R575.

Key features

  • PyTorch 2.9.0, with compatible versions of Torchvision and Torchaudio.

  • Python 3.12.7, located at /opt/ac2/bin.

Release history

December 2025: Released the PyTorch 2.9.0 image for CUDA 13.0.2 and Python 3.12.

PyTorch 2.7.1.8

Images

Image type Registry address
GPU ac2-registry.cn-hangzhou.cr.aliyuncs.com/ac2/pytorch:2.7.1.8-cuda12.8.1-py312-alinux3.2104

Image content

ac2-registry.cn-hangzhou.cr.aliyuncs.com/ac2/pytorch:2.7.1.8-cuda12.8.1-py312-alinux3.2104

CUDA libraries

  • cuda-cudart-12-8: 12.8.90

  • cuda-compat-12-8: 570.124.06

  • cuda-libraries-12-8: 12.8.1

  • cuda-opencl-12-8: 12.8.90

  • cuda-nvrtc-12-8: 12.8.93

  • cuda-nvtx-12-8: 12.8.90

  • cuda-cupti-12-8: 12.8.90

  • cuda-nvvm-12-8: 12.8.93

  • cuda-crt-12-8: 12.8.93

  • cuda-nvcc-12-8: 12.8.93

Acceleration libraries

  • libcublas-12-8: 12.8.4.1

  • libcufft-12-8: 11.3.3.83

  • libcufile-12-8: 1.13.1.3

  • libcurand-12-8: 10.3.9.90

  • libcusolver-12-8: 11.7.3.90

  • libcusparse-12-8: 12.5.8.93

  • libnpp-12-8: 12.3.3.100

  • libnvjitlink-12-8: 12.8.93

  • libnvjpeg-12-8: 12.3.5.92

  • libnvfatbin-12-8: 12.8.90

  • libnccl: 2.26.2

  • libcudnn9: 9.8.0.87

Python packages

  • Python: 3.12.7

  • torch: 2.7.1.8+cu128

  • torchaudio: 2.7.1.8+cu128

  • torchvision: 0.22.1.8+cu128

  • triton: 3.2.0

  • numpy: 2.1.3

  • pillow: 10.3.0

  • transformers: 4.53.1

  • accelerate: 1.2.1

  • deepspeed: 0.16.9+ali

  • huggingface_hub: 0.30.2

  • safetensors: 0.4.3

  • tokenizers: 0.21.0

  • einops: 0.8.0

  • pydantic: 2.10.4

  • pydantic_core: 2.27.2

  • sympy: 1.13.3

  • mpmath: 1.3.0

  • networkx: 3.3

  • ninja: 1.11.1.1

  • Jinja2: 3.1.4

  • MarkupSafe: 2.1.5

  • filelock: 3.15.4

  • fsspec: 2024.6.0

  • typing_extensions: 4.12.2

  • annotated-types: 0.7.0

  • certifi: 2024.12.14

  • charset-normalizer: 3.3.2

  • idna: 3.7

  • requests: 2.32.4

  • urllib3: 2.2.2

  • PySocks: 1.7.1

  • packaging: 24.0

  • psutil: 5.9.2

  • py-cpuinfo: 9.0.0

  • PyYAML: 6.0.1

  • regex: 2024.5.15

  • tqdm: 4.66.4

  • msgpack: 1.1.0

  • hjson: 3.0.2

  • pip: 24.2

  • setuptools: 80.9.0

  • wheel: 0.45.0

Driver requirements

Based on CUDA runtime 12.8.1. Requires NVIDIA Driver release 570 or later and is compatible with NVIDIA Driver releases R470, R535, R550, R560, and R565.

Key features

  • PyTorch 2.7.1.8, with compatible versions of Torchvision and Torchaudio.

  • Python 3.12.7, located at /opt/ac2/bin.

Release history

July 2025: Released the PyTorch 2.7.1.8 image for CUDA 12.8.1 and Python 3.12.

PyTorch 2.5.1.6

Images

Image type Registry address
AMD GPU ac2-registry.cn-hangzhou.cr.aliyuncs.com/ac2/pytorch:2.5.1.6-rocm6.2.4-py310-alinux3.2104-enhanced
GPU ac2-registry.cn-hangzhou.cr.aliyuncs.com/ac2/pytorch:2.5.1.6-cuda12.1.1-py310-alinux3.2104

Image content

ac2-registry.cn-hangzhou.cr.aliyuncs.com/ac2/pytorch:2.5.1.6-rocm6.2.4-py310-alinux3.2104-enhanced

ROCm libraries

  • amd-smi-lib: 24.6.3.60204

  • amdgpu-core: 6.2.60204

  • half: 1.12.0.60204

  • hip-runtime-amd: 6.2.41134.60204

  • hipblas: 2.2.0.60204

  • hipblaslt: 0.8.0.60204

  • hipcc: 1.1.1.60204

  • hipfft: 1.0.16.60204

  • hipify-clang: 18.0.0.60204

  • hiprand: 2.11.1.60204

  • hipsolver: 2.2.0.60204

  • hipsparse: 3.1.1.60204

  • hipsparselt: 0.2.1.60204

  • hiptensor: 1.3.0.60204

  • hsa-rocr: 1.14.0.60204

  • miopen-hip: 3.2.0.60204

  • omnitrace: 1.11.2.60204

  • openmp-extras-runtime: 18.62.0.60204

  • rccl: 2.20.5.60204

  • rocalution: 3.2.1.60204

  • rocblas: 4.2.4.60204

  • rocfft: 1.0.30.60204

  • rocm-cmake: 0.13.0.60204

  • rocm-dbgapi: 0.76.0.60204

  • rocm-debug-agent: 2.0.3.60204

  • rocm-device-libs: 1.0.0.60204

  • rocm-gdb: 14.2.60204

  • rocm-llvm: 18.0.0.24392.60204

  • rocm-opencl-icd-loader: 1.2.60204

  • rocm-opencl: 2.0.0.60204

  • rocm-smi-lib: 7.3.0.60204

  • rocminfo: 1.0.0.60204

  • rocprim-devel: 3.2.2.60204

  • rocprofiler-plugins: 2.0.60204.60204

  • rocprofiler-register: 0.4.0.60204

  • rocprofiler: 2.0.60204.60204

  • rocrand: 3.1.1.60204

  • rocsolver: 3.26.2.60204

  • rocsparse: 3.2.1.60204

  • roctracer: 4.1.60204.60204

  • rocwmma-devel: 1.5.0.60204

Python packages

  • Python: 3.10.13

  • torch: 2.5.1.6

  • torchaudio: 2.5.1.6

  • torchvision: 0.20.1.6

  • pytorch_triton_rocm: 3.1.0

  • numpy: 1.26.4

  • pillow: 10.3.0

  • transformers: 4.47.1

  • accelerate: 1.2.1

  • huggingface_hub: 0.27.0

  • safetensors: 0.4.3

  • tokenizers: 0.21.0

  • pydantic: 2.8.2

  • pydantic_core: 2.20.1

  • sympy: 1.13.1

  • networkx: 3.3

  • ninja: 1.11.1.1

  • Jinja2: 3.1.4

  • filelock: 3.15.4

  • fsspec: 2024.6.0

  • typing-extensions: 4.12.2

  • annotated_types: 0.7.0

  • certifi: 2024.12.14

  • charset-normalizer: 3.3.2

  • idna: 3.7

  • requests: 2.32.3

  • urllib3: 2.2.2

  • pysocks: 1.7.1

  • packaging: 24.0

  • psutil: 5.9.2

  • py_cpuinfo: 9.0.0

  • PyYAML: 6.0.1

  • regex: 2024.5.15

  • tqdm: 4.66.4

  • msgpack: 1.1.0

  • hjson: 3.0.2

ac2-registry.cn-hangzhou.cr.aliyuncs.com/ac2/pytorch:2.5.1.6-cuda12.1.1-py310-alinux3.2104

CUDA libraries

  • cuda-cudart-12-1: 12.1.1

  • cuda-libraries-12-1: 12.1.1

  • cuda-opencl-12-1: 12.1.105

  • cuda-nvrtc-12-1: 12.1.105

  • cuda-nvtx-12-1: 12.1.105

  • cuda-cupti-12-1: 12.1.1

Acceleration libraries

  • libcublas-12-1: 12.1.3.1

  • libcufft-12-1: 11.0.2.54

  • libcufile-12-1: 1.6.1.9

  • libcurand-12-1: 10.3.2.106

  • libcusolver-12-1: 11.4.5.107

  • libcuparse-12-1: 12.1.0.106

  • libnpp-12-1: 12.1.0.40

  • libnvjitlink-12-1: 12.1.105

  • libnvjpeg-12-1: 12.2.0.2

  • libnccl: 2.20.5

  • libcudnn9: 9.2.1.18

Python packages

  • Python: 3.10.13

  • torch: 2.5.1.6+cu121

  • torchaudio: 2.5.1.6+cu121

  • torchvision: 0.20.1.6+cu121

  • triton: 3.1.0

  • numpy: 1.26.4

  • pillow: 10.3.0

  • transformers: 4.47.1

  • accelerate: 1.2.1

  • deepspeed: 0.15.4+ali

  • huggingface_hub: 0.27.0

  • safetensors: 0.4.3

  • tokenizers: 0.21.0

  • pydantic: 2.8.2

  • pydantic_core: 2.20.1

  • sympy: 1.13.1

  • mpmath: 1.3.0

  • networkx: 3.3

  • ninja: 1.11.1.1

  • Jinja2: 3.1.4

  • MarkupSafe: 2.1.5

  • filelock: 3.15.4

  • fsspec: 2024.6.0

  • typing_extensions: 4.12.2

  • annotated-types: 0.7.0

  • certifi: 2024.12.14

  • charset-normalizer: 3.3.2

  • idna: 3.7

  • requests: 2.32.3

  • urllib3: 2.2.2

  • PySocks: 1.7.1

  • packaging: 24.0

  • psutil: 5.9.2

  • py-cpuinfo: 9.0.0

  • PyYAML: 6.0.1

  • regex: 2024.5.15

  • tqdm: 4.66.4

  • msgpack: 1.1.0

  • hjson: 3.0.2

  • pip: 24.0

  • setuptools: 73.0.1

  • wheel: 0.43.0

Driver requirements

  • AMD GPU image (rocm6.2.4): Contains ROCm 6.2.4. The host driver version must be 6.0.x or later. For driver compatibility details, see the user and kernel-space support matrix.

  • GPU image (cuda12.1.1): Based on CUDA runtime 12.1.1. Requires NVIDIA Driver release 525 or later and is compatible with NVIDIA Driver release 470.

Key features

  • PyTorch 2.5.1.6, with compatible versions of Torchvision and Torchaudio.

  • Python 3.10.13+gc, located at /opt/ac2/bin. This version integrates a garbage collection (GC) optimization that reduces GC pauses in AI workloads.

Release history

January 2025: Released the PyTorch 2.5.1.6 for ROCm 6.2.4 image.

February 2025: Released the PyTorch 2.5.1.6 for CUDA 12.1.1 image.

PyTorch 2.4.0

Images

Image type Registry address
GPU ac2-registry.cn-hangzhou.cr.aliyuncs.com/ac2/pytorch:2.4.0-cuda12.1.1-py310-alinux3.2104

Image content

ac2-registry.cn-hangzhou.cr.aliyuncs.com/ac2/pytorch:2.4.0-cuda12.1.1-py310-alinux3.2104

CUDA libraries

  • cuda-cudart-12-1: 12.1.1

  • cuda-libraries-12-1: 12.1.1

  • cuda-opencl-12-1: 12.1.105

  • cuda-nvrtc-12-1: 12.1.105

  • cuda-nvtx-12-1: 12.1.105

  • cuda-cupti-12-1: 12.1.1

Acceleration libraries

  • libcublas-12-1: 12.1.3.1

  • libcufft-12-1: 11.0.2.54

  • libcufile-12-1: 1.6.1.9

  • libcurand-12-1: 10.3.2.106

  • libcusolver-12-1: 11.4.5.107

  • libcuparse-12-1: 12.1.0.106

  • libnpp-12-1: 12.1.0.40

  • libnvjitlink-12-1: 12.1.105

  • libnvjpeg-12-1: 12.2.0.2

  • libnccl: 2.20.5

  • libcudnn9: 9.2.1.18

Python packages

  • Python: 3.10.13

  • torch: 2.4.0+cu121

  • torchaudio: 2.4.0+cu121

  • torchvision: 0.19.0+cu121

  • numpy: 1.26.4

  • pillow: 10.3.0

  • sympy: 1.12.1

  • mpmath: 1.3.0

  • networkx: 3.3

  • Jinja2: 3.1.4

  • filelock: 3.15.4

  • fsspec: 2024.6.0

  • typing-extensions: 4.12.2

  • certifi: 2024.6.2

  • charset-normalizer: 3.3.2

  • idna: 3.7

  • requests: 2.32.3

  • urllib3: 2.2.2

  • pysocks: 1.7.1

Driver requirements

Based on CUDA runtime 12.1.1. Requires NVIDIA Driver release 525 or later and is compatible with NVIDIA Driver release 470.

Key features

PyTorch 2.4.0, with updated versions of Torchvision and Torchaudio.

Release history

November 2024: Released the PyTorch 2.4.0 for CUDA 12.1.1 image.

PyTorch 2.3.0.3

Images

Image type Registry address
CPU ac2-registry.cn-hangzhou.cr.aliyuncs.com/ac2/pytorch:2.3.0.3-alinux3.2304-enhanced
GPU ac2-registry.cn-hangzhou.cr.aliyuncs.com/ac2/pytorch:2.3.0.3-cuda12.1.1-alinux3.2304-enhanced
GPU ac2-registry.cn-hangzhou.cr.aliyuncs.com/ac2/pytorch:2.3.0.3-cuda11.8.0-alinux3.2304-enhanced

Image content

pytorch:2.3.0.3-alinux3.2304-enhanced

  • gcc: 12.2.1

  • Python: 3.10.13

  • torch: 2.3.0.3.post1

  • torchaudio: 2.3.0.3.post1

  • torchvision: 0.18.0.3.post1

  • numpy: 1.24.2

  • Pillow: 10.1.0

  • pyarrow: 14.0.2

  • sympy: 1.11.1

  • mpmath: 1.3.0

  • networkx: 2.8.8

  • Jinja2: 3.1.2

  • MarkupSafe: 2.1.2

  • filelock: 3.9.0

  • fsspec: 2023.6.0

  • typing_extensions: 4.9.0

  • certifi: 2023.7.22

  • charset-normalizer: 3.1.0

  • idna: 3.7

  • requests: 2.31.0

  • urllib3: 1.26.18

  • PySocks: 1.7.1

  • packaging: 23.0

  • paramiko: 2.12.0

  • bcrypt: 3.2.2

  • cffi: 1.15.1

  • cryptography: 42.0.5

  • pycparser: 2.21

  • PyNaCl: 1.4.0

  • pyasn1: 0.4.8

  • coverage: 7.2.1

  • pytest: 7.3.1

  • pytest-cov: 4.0.0

  • exceptiongroup: 1.1.1

  • iniconfig: 1.1.1

  • pluggy: 1.0.0

  • tomli: 2.0.1

  • tqdm: 4.65.0

  • six: 1.16.0

  • ply: 3.11

  • lxml: 4.9.2

  • libcomps: 0.1.19

  • olefile: 0.46

  • pip: 23.3.1

  • setuptools: 65.5.1

pytorch:2.3.0.3-cuda12.1.1-alinux3.2304-enhanced

CUDA libraries

  • cuda-cudart-12-1: 12.1.1

  • cuda-compat-12-1: 530.30.02

  • cuda-libraries-12-1: 12.1.1

  • cuda-opencl-12-1: 12.1.105

  • cuda-nvrtc-12-1: 12.1.105

  • cuda-nvtx-12-1: 12.1.105

  • cuda-cupti-12-1: 12.1.105

Acceleration libraries

  • libcublas-12-1: 12.1.3.1

  • libcufft-12-1: 11.0.2.54

  • libcufile-12-1: 1.6.1.9

  • libcurand-12-1: 10.3.2.106

  • libcusolver-12-1: 11.4.5.107

  • libcuparse-12-1: 12.1.0.106

  • libnpp-12-1: 12.1.0.40

  • libnvjitlink-12-1: 12.1.105

  • libnvjpeg-12-1: 12.2.0.2

  • libnccl: 2.20.5

  • libibverbs: 44.0

  • libnl3: 3.7.0

  • librdmacm: 44.0

  • libcudnn: 8.9.3.28

Python packages

  • Python: 3.10.13

  • torch: 2.3.0.3.post1+cu121

  • torchaudio: 2.3.0.3.post1+cu121

  • torchvision: 0.18.0.3.post1+cu121

  • triton: 2.3.0

  • numpy: 1.24.2

  • pyarrow: 14.0.2

  • sympy: 1.11.1

  • mpmath: 1.3.0

  • networkx: 2.8.8

  • Jinja2: 3.1.2

  • MarkupSafe: 2.1.2

  • filelock: 3.9.0

  • fsspec: 2023.6.0

  • typing_extensions: 4.9.0

  • certifi: 2023.7.22

  • charset-normalizer: 3.1.0

  • idna: 3.7

  • requests: 2.31.0

  • urllib3: 1.26.18

  • PySocks: 1.7.1

  • packaging: 23.0

  • paramiko: 2.12.0

  • bcrypt: 3.2.2

  • cffi: 1.15.1

  • cryptography: 42.0.5

  • pycparser: 2.21

  • PyNaCl: 1.4.0

  • pyasn1: 0.4.8

  • coverage: 7.2.1

  • pytest: 7.3.1

  • pytest-cov: 4.0.0

  • exceptiongroup: 1.1.1

  • iniconfig: 1.1.1

  • pluggy: 1.0.0

  • tomli: 2.0.1

  • tqdm: 4.65.0

  • six: 1.16.0

  • ply: 3.11

  • lxml: 4.9.2

  • libcomps: 0.1.19

  • pip: 23.3.1

  • setuptools: 65.5.1

pytorch:2.3.0.3-cuda11.8.0-alinux3.2304-enhanced

CUDA libraries

  • cuda-cudart-11-8: 11.8.89

  • cuda-libraries-11-8: 11.8.0

  • cuda-nvrtc-11-8: 11.8.89

  • cuda-nvtx-11-8: 11.8.86

  • cuda-cupti-11-8: 11.8.87

Acceleration libraries

  • libcublas-11-8: 11.11.3.6

  • libcufft-11-8: 10.9.0.58

  • libcufile-11-8: 1.4.0.31

  • libcurand-11-8: 10.3.0.86

  • libcusolver-11-8: 11.4.1.48

  • libcusparse-11-8: 11.7.5.86

  • libnpp-11-8: 11.8.0.86

  • libnvjpeg-11-8: 11.8.0.86

  • libnccl-cuda-11: 2.20.5

  • libibverbs: 46.0

  • libnl3: 3.7.0

  • librdmacm: 46.0

  • libcudnn-cuda-11: 8.9.3.28

Python packages

  • Python: 3.10.13

  • torch: 2.3.0.3.post1+cu118

  • torchaudio: 2.3.0.3.post1+cu118

  • torchvision: 0.18.0.3.post1+cu118

  • triton: 2.3.0

  • numpy: 1.24.2

  • pyarrow: 14.0.2

  • sympy: 1.11.1

  • mpmath: 1.3.0

  • networkx: 2.8.8

  • Jinja2: 3.1.2

  • MarkupSafe: 2.1.2

  • filelock: 3.9.0

  • fsspec: 2023.6.0

  • typing_extensions: 4.9.0

  • certifi: 2023.7.22

  • charset-normalizer: 3.1.0

  • idna: 3.7

  • requests: 2.31.0

  • urllib3: 1.26.18

  • PySocks: 1.7.1

  • packaging: 23.0

  • paramiko: 2.12.0

  • bcrypt: 3.2.2

  • cffi: 1.15.1

  • cryptography: 42.0.5

  • pycparser: 2.21

  • PyNaCl: 1.4.0

  • pyasn1: 0.4.8

  • coverage: 7.2.1

  • pytest: 7.3.1

  • pytest-cov: 4.0.0

  • exceptiongroup: 1.1.1

  • iniconfig: 1.1.1

  • pluggy: 1.0.0

  • tomli: 2.0.1

  • tqdm: 4.65.0

  • six: 1.16.0

  • ply: 3.11

  • lxml: 4.9.2

  • libcomps: 0.1.19

  • pip: 23.3.1

  • setuptools: 65.5.1

Driver requirements

  • CPU image (alinux3.2304-enhanced): No driver requirements.

  • GPU image (cuda12.1.1): Based on CUDA runtime 12.1.1. Requires NVIDIA Driver release 525 or later and is compatible with NVIDIA Driver release 470.

  • GPU image (cuda11.8.0): Requires NVIDIA Driver release 470 or later.

Key features

  • PyTorch 2.3.0.3.post1, with updated versions of Torchvision and Torchaudio.

  • Built on PyTorch 2.3.0 with optimized Dynamo model support and performance. Provides Dynamo performance optimizations when used with DeepSpeed or Fully Sharded Data Parallel (FSDP).

Release history

July 2024: Released PyTorch 2.3.0.3 images. The release includes a CPU image, a CUDA 11.8 image, and a CUDA 12.1 image.

PyTorch 2.3.0

Images

Image type Registry address
CPU ac2-registry.cn-hangzhou.cr.aliyuncs.com/ac2/pytorch:2.3.0-alinux3.2304
GPU ac2-registry.cn-hangzhou.cr.aliyuncs.com/ac2/pytorch:2.3.0-cuda12.1.1-alinux3.2304
GPU ac2-registry.cn-hangzhou.cr.aliyuncs.com/ac2/pytorch:2.3.0-cuda11.8.0-alinux3.2304

Image content

pytorch:2.3.0-alinux3.2304

  • gcc: 12.2.1

  • Python: 3.10.13

  • torch: 2.3.0

  • torchaudio: 2.3.0

  • torchvision: 0.18.0

  • numpy: 1.24.2

  • Pillow: 10.1.0

  • pyarrow: 14.0.2

  • sympy: 1.11.1

  • mpmath: 1.3.0

  • networkx: 2.8.8

  • Jinja2: 3.1.2

  • MarkupSafe: 2.1.2

  • filelock: 3.9.0

  • fsspec: 2023.6.0

  • typing_extensions: 4.9.0

  • certifi: 2023.7.22

  • charset-normalizer: 3.1.0

  • idna: 3.7

  • requests: 2.31.0

  • urllib3: 1.26.18

  • PySocks: 1.7.1

  • packaging: 23.0

  • paramiko: 2.12.0

  • bcrypt: 3.2.2

  • cffi: 1.15.1

  • cryptography: 42.0.5

  • pycparser: 2.21

  • PyNaCl: 1.4.0

  • pyasn1: 0.4.8

  • coverage: 7.2.1

  • pytest: 7.3.1

  • pytest-cov: 4.0.0

  • exceptiongroup: 1.1.1

  • iniconfig: 1.1.1

  • pluggy: 1.0.0

  • tomli: 2.0.1

  • tqdm: 4.65.0

  • six: 1.16.0

  • ply: 3.11

  • lxml: 4.9.2

  • libcomps: 0.1.19

  • olefile: 0.46

  • pip: 23.3.1

  • setuptools: 65.5.1

pytorch:2.3.0-cuda12.1.1-alinux3.2304

CUDA libraries

  • cuda-cudart-12-1: 12.1.1

  • cuda-compat-12-1: 530.30.02

  • cuda-libraries-12-1: 12.1.1

  • cuda-opencl-12-1: 12.1.105

  • cuda-nvrtc-12-1: 12.1.105

  • cuda-nvtx-12-1: 12.1.105

  • cuda-cupti-12-1: 12.1.105

Acceleration libraries

  • libcublas-12-1: 12.1.3.1

  • libcufft-12-1: 11.0.2.54

  • libcufile-12-1: 1.6.1.9

  • libcurand-12-1: 10.3.2.106

  • libcusolver-12-1: 11.4.5.107

  • libcuparse-12-1: 12.1.0.106

  • libnpp-12-1: 12.1.0.40

  • libnvjitlink-12-1: 12.1.105

  • libnvjpeg-12-1: 12.2.0.2

  • libnccl: 2.20.5

  • libibverbs: 44.0

  • libnl3: 3.7.0

  • librdmacm: 44.0

  • libcudnn: 8.9.3.28

Python packages

  • Python: 3.10.13

  • torch: 2.3.0+cu121

  • torchaudio: 2.3.0+cu121

  • torchvision: 0.18.0+cu121

  • triton: 2.3.0

  • numpy: 1.24.2

  • pyarrow: 14.0.2

  • sympy: 1.11.1

  • mpmath: 1.3.0

  • networkx: 2.8.8

  • Jinja2: 3.1.2

  • MarkupSafe: 2.1.2

  • filelock: 3.9.0

  • fsspec: 2023.6.0

  • typing_extensions: 4.9.0

  • certifi: 2023.7.22

  • charset-normalizer: 3.1.0

  • idna: 3.7

  • requests: 2.31.0

  • urllib3: 1.26.18

  • PySocks: 1.7.1

  • packaging: 23.0

  • paramiko: 2.12.0

  • bcrypt: 3.2.2

  • cffi: 1.15.1

  • cryptography: 42.0.5

  • pycparser: 2.21

  • PyNaCl: 1.4.0

  • pyasn1: 0.4.8

  • coverage: 7.2.1

  • pytest: 7.3.1

  • pytest-cov: 4.0.0

  • exceptiongroup: 1.1.1

  • iniconfig: 1.1.1

  • pluggy: 1.0.0

  • tomli: 2.0.1

  • tqdm: 4.65.0

  • six: 1.16.0

  • ply: 3.11

  • lxml: 4.9.2

  • libcomps: 0.1.19

  • pip: 23.3.1

  • setuptools: 65.5.1

pytorch:2.3.0-cuda11.8.0-alinux3.2304

CUDA libraries

  • cuda-cudart-11-8: 11.8.89

  • cuda-libraries-11-8: 11.8.0

  • cuda-nvrtc-11-8: 11.8.89

  • cuda-nvtx-11-8: 11.8.86

  • cuda-cupti-11-8: 11.8.87

Acceleration libraries

  • libcublas-11-8: 11.11.3.6

  • libcufft-11-8: 10.9.0.58

  • libcufile-11-8: 1.4.0.31

  • libcurand-11-8: 10.3.0.86

  • libcusolver-11-8: 11.4.1.48

  • libcusparse-11-8: 11.7.5.86

  • libnpp-11-8: 11.8.0.86

  • libnvjpeg-11-8: 11.8.0.86

  • libnccl-cuda-11: 2.20.5

  • libibverbs: 46.0

  • libnl3: 3.7.0

  • librdmacm: 46.0

  • libcudnn-cuda-11: 8.9.3.28

Python packages

  • Python: 3.10.13

  • torch: 2.3.0+cu118

  • torchaudio: 2.3.0+cu118

  • torchvision: 0.18.0+cu118

  • triton: 2.3.0

  • numpy: 1.24.2

  • pyarrow: 14.0.2

  • sympy: 1.11.1

  • mpmath: 1.3.0

  • networkx: 2.8.8

  • Jinja2: 3.1.2

  • MarkupSafe: 2.1.2

  • filelock: 3.9.0

  • fsspec: 2023.6.0

  • typing_extensions: 4.9.0

  • certifi: 2023.7.22

  • charset-normalizer: 3.1.0

  • idna: 3.7

  • requests: 2.31.0

  • urllib3: 1.26.18

  • PySocks: 1.7.1

  • packaging: 23.0

  • paramiko: 2.12.0

  • bcrypt: 3.2.2

  • cffi: 1.15.1

  • cryptography: 42.0.5

  • pycparser: 2.21

  • PyNaCl: 1.4.0

  • pyasn1: 0.4.8

  • coverage: 7.2.1

  • pytest: 7.3.1

  • pytest-cov: 4.0.0

  • exceptiongroup: 1.1.1

  • iniconfig: 1.1.1

  • pluggy: 1.0.0

  • tomli: 2.0.1

  • tqdm: 4.65.0

  • six: 1.16.0

  • ply: 3.11

  • lxml: 4.9.2

  • libcomps: 0.1.19

  • pip: 23.3.1

  • setuptools: 65.5.1

Driver requirements

  • CPU image (alinux3.2304): No driver requirements.

  • GPU image (cuda12.1.1): Based on CUDA runtime 12.1.1. Requires NVIDIA Driver release 525 or later and is compatible with NVIDIA Driver release 470.

  • GPU image (cuda11.8.0): Requires NVIDIA Driver release 470 or later.

Key features

PyTorch 2.3.0, with updated versions of Torchvision and Torchaudio.

Release history

June 2024: Released PyTorch 2.3.0 images. The release includes a CPU image, a CUDA 11.8 image, and a CUDA 12.1 image.

PyTorch 2.2.0.1

Images

Image type Registry address
GPU ac2-registry.cn-hangzhou.cr.aliyuncs.com/ac2/pytorch:2.2.0.1-3.2304-cu121
GPU ac2-registry.cn-hangzhou.cr.aliyuncs.com/ac2/pytorch:2.2.0.1-3.2304-cu118
CPU ac2-registry.cn-hangzhou.cr.aliyuncs.com/ac2/pytorch:2.2.0.1-3.2304

Image content

pytorch:2.2.0.1-3.2304

  • gcc: 12.2.1

  • Python: 3.10.13

  • torch: 2.2.0.1

  • torchaudio: 2.2.0.1

  • torchvision: 0.17.0.1

  • numpy: 1.24.2

  • Pillow: 9.4.0

  • SciPy: 1.10.1

  • matplotlib: 3.7.1

  • pyarrow: 14.0.2

  • sympy: 1.11.1

  • mpmath: 1.3.0

  • networkx: 2.8.8

  • Jinja2: 3.1.2

  • MarkupSafe: 2.1.2

  • filelock: 3.9.0

  • fsspec: 2023.6.0

  • typing_extensions: 4.9.0

  • certifi: 2023.7.22

  • charset-normalizer: 3.1.0

  • idna: 3.4

  • requests: 2.31.0

  • urllib3: 1.26.18

  • PySocks: 1.7.1

  • packaging: 23.0

  • paramiko: 2.12.0

  • bcrypt: 3.2.2

  • cffi: 1.15.1

  • cryptography: 39.0.0

  • pycparser: 2.21

  • PyNaCl: 1.4.0

  • pyasn1: 0.4.8

  • coverage: 7.2.1

  • pytest: 7.3.1

  • pytest-cov: 4.0.0

  • exceptiongroup: 1.1.1

  • iniconfig: 1.1.1

  • pluggy: 1.0.0

  • tomli: 2.0.1

  • tqdm: 4.65.0

  • six: 1.16.0

  • ply: 3.11

  • lxml: 4.9.2

  • libcomps: 0.1.19

  • olefile: 0.46

  • aiodns: 3.0.0

  • aiohttp: 3.8.4

  • aiosignal: 1.3.1

  • appdirs: 1.4.4

  • async-timeout: 4.0.2

  • attrs: 22.2.0

  • Brotli: 1.0.9

  • contourpy: 1.0.7

  • cycler: 0.11.0

  • fonttools: 4.39.2

  • frozenlist: 1.3.3

  • fs: 2.4.16

  • kiwisolver: 1.4.4

  • multidict: 6.0.4

  • munkres: 1.1.2

  • pycairo: 1.23.0

  • pycares: 4.3.0

  • pyparsing: 3.0.7

  • python-dateutil: 2.8.2

  • yarl: 1.8.2

  • pip: 23.3.1

  • setuptools: 65.5.1

pytorch:2.2.0.1-3.2304-cu121

CUDA libraries

  • cuda-cudart-12-1: 12.1.1

  • cuda-compat-12-1: 530.30.02

  • cuda-libraries-12-1: 12.1.1

  • cuda-opencl-12-1: 12.1.105

  • cuda-nvrtc-12-1: 12.1.105

  • cuda-nvtx-12-1: 12.1.105

  • cuda-cupti-12-1: 12.1.105

Acceleration libraries

  • libcublas-12-1: 12.1.3.1

  • libcufft-12-1: 11.0.2.54

  • libcufile-12-1: 1.6.1.9

  • libcurand-12-1: 10.3.2.106

  • libcusolver-12-1: 11.4.5.107

  • libcuparse-12-1: 12.1.0.106

  • libnpp-12-1: 12.1.0.40

  • libnvjitlink-12-1: 12.1.105

  • libnvjpeg-12-1: 12.2.0.2

  • libnccl: 2.19.3

  • libibverbs: 44.0

  • libnl3: 3.7.0

  • librdmacm: 44.0

  • libcudnn: 8.9.3.28

Python packages

  • Python: 3.10.13

  • torch: 2.2.0.1+cu121

  • torchaudio: 2.2.0.1+cu121

  • torchvision: 0.17.0.1+cu121

  • triton: 2.2.0

  • numpy: 1.24.2

  • Pillow: 9.4.0

  • SciPy: 1.10.1

  • matplotlib: 3.7.1

  • pyarrow: 14.0.2

  • sympy: 1.11.1

  • mpmath: 1.3.0

  • networkx: 2.8.8

  • Jinja2: 3.1.2

  • MarkupSafe: 2.1.2

  • filelock: 3.9.0

  • fsspec: 2023.6.0

  • typing_extensions: 4.9.0

  • certifi: 2023.7.22

  • charset-normalizer: 3.1.0

  • idna: 3.4

  • requests: 2.31.0

  • urllib3: 1.26.18

  • PySocks: 1.7.1

  • packaging: 23.0

  • paramiko: 2.12.0

  • bcrypt: 3.2.2

  • cffi: 1.15.1

  • cryptography: 39.0.0

  • pycparser: 2.21

  • PyNaCl: 1.4.0

  • pyasn1: 0.4.8

  • coverage: 7.2.1

  • pytest: 7.3.1

  • pytest-cov: 4.0.0

  • exceptiongroup: 1.1.1

  • iniconfig: 1.1.1

  • pluggy: 1.0.0

  • tomli: 2.0.1

  • tqdm: 4.65.0

  • six: 1.16.0

  • ply: 3.11

  • lxml: 4.9.2

  • libcomps: 0.1.19

  • olefile: 0.46

  • aiodns: 3.0.0

  • aiohttp: 3.8.4

  • aiosignal: 1.3.1

  • appdirs: 1.4.4

  • async-timeout: 4.0.2

  • attrs: 22.2.0

  • Brotli: 1.0.9

  • contourpy: 1.0.7

  • cycler: 0.11.0

  • fonttools: 4.39.2

  • frozenlist: 1.3.3

  • fs: 2.4.16

  • kiwisolver: 1.4.4

  • multidict: 6.0.4

  • munkres: 1.1.2

  • pycairo: 1.23.0

  • pycares: 4.3.0

  • pyparsing: 3.0.7

  • python-dateutil: 2.8.2

  • yarl: 1.8.2

  • pip: 23.3.1

  • setuptools: 65.5.1

pytorch:2.2.0.1-3.2304-cu118

CUDA libraries

  • cuda-cudart-11-8: 11.8.89

  • cuda-libraries-11-8: 11.8.0

  • cuda-nvrtc-11-8: 11.8.89

  • cuda-nvtx-11-8: 11.8.86

  • cuda-cupti-11-8: 11.8.87

Acceleration libraries

  • libcublas-11-8: 11.11.3.6

  • libcufft-11-8: 10.9.0.58

  • libcufile-11-8: 1.4.0.31

  • libcurand-11-8: 10.3.0.86

  • libcusolver-11-8: 11.4.1.48

  • libcusparse-11-8: 11.7.5.86

  • libnpp-11-8: 11.8.0.86

  • libnvjpeg-11-8: 11.8.0.86

  • libnccl-cuda-11: 2.19.3

  • libibverbs: 46.0

  • libnl3: 3.7.0

  • librdmacm: 46.0

  • libcudnn-cuda-11: 8.9.3.28

Python packages

  • Python: 3.10.13

  • torch: 2.2.0.1+cu118

  • torchaudio: 2.2.0.1+cu118

  • torchvision: 0.17.0.1+cu118

  • triton: 2.2.0+cu118

  • numpy: 1.24.2

  • Pillow: 9.4.0

  • SciPy: 1.10.1

  • matplotlib: 3.7.1

  • pyarrow: 14.0.2

  • sympy: 1.11.1

  • mpmath: 1.3.0

  • networkx: 2.8.8

  • Jinja2: 3.1.2

  • MarkupSafe: 2.1.2

  • filelock: 3.9.0

  • fsspec: 2023.6.0

  • typing_extensions: 4.9.0

  • certifi: 2023.7.22

  • charset-normalizer: 3.1.0

  • idna: 3.4

  • requests: 2.31.0

  • urllib3: 1.26.18

  • PySocks: 1.7.1

  • packaging: 23.0

  • paramiko: 2.12.0

  • bcrypt: 3.2.2

  • cffi: 1.15.1

  • cryptography: 39.0.0

  • pycparser: 2.21

  • PyNaCl: 1.4.0

  • pyasn1: 0.4.8

  • coverage: 7.2.1

  • pytest: 7.3.1

  • pytest-cov: 4.0.0

  • exceptiongroup: 1.1.1

  • iniconfig: 1.1.1

  • pluggy: 1.0.0

  • tomli: 2.0.1

  • tqdm: 4.65.0

  • six: 1.16.0

  • ply: 3.11

  • lxml: 4.9.2

  • libcomps: 0.1.19

  • fonttools: 4.47.0

  • pip: 23.3.1

  • setuptools: 65.5.1

Driver requirements

  • CPU image (2.2.0.1-3.2304): No driver requirements.

  • GPU image (cu121): Based on CUDA runtime 12.1.1. Requires NVIDIA Driver release 525 or later and is compatible with NVIDIA Driver release 470.

  • GPU image (cu118): Requires NVIDIA Driver release 470 or later.

Key features

  • PyTorch 2.2.0.1, with updated versions of Torchvision and Torchaudio.

  • Built on PyTorch 2.2.0 with optimized Dynamo model support and performance. Provides Dynamo performance optimizations when used with DeepSpeed or Fully Sharded Data Parallel (FSDP).

Release history

March 2024: Released PyTorch 2.2.0.1 images. The release includes a CPU image, a CUDA 11.8 image, and a CUDA 12.1 image.

PyTorch 2.1.0

Images

Image type Registry address
DCU ac2-registry.cn-hangzhou.cr.aliyuncs.com/ac2/pytorch:2.1.0-dtk24.04.1-py310-alinux3.2304

Image content

pytorch:2.1.0-dtk24.04.1-py310-alinux3.2304

  • Python: 3.10.13

  • DTK: 24.04.1

  • torch: 2.1.0+das1.1.git3ac1bdd.abi1.dtk2404

  • torchaudio: 2.1.2+das1.1.git63d9a68.abi1.dtk2404.torch2.1.0

  • torchvision: 0.16.0+das1.1.git7d45932.abi1.dtk2404.torch2.1

  • triton: 2.1.0+das1.1.git4bf1007a.abi1.dtk2404.torch2.1.0

  • apex: 1.1.0+das1.1.gitf477a3a.abi1.dtk2404.torch2.1.0

  • flash-attn: 2.0.4+das1.1gitc7a8c18.abi1.dtk2404.torch2.1

  • xformers: 0.0.25+das1.1.git8ef8bc1.abi1.dtk2404.torch2.1.0

  • dropout-layer-norm: 0.1+das1.1gitc7a8c18.abi1.dtk2404.torch2.1

  • fused-dense-lib: 0.1+das1.1gitc7a8c18.abi1.dtk2404.torch2.1

  • rotary-emb: 0.1+das1.1gitc7a8c18.abi1.dtk2404.torch2.1

  • xentropy-cuda-lib: 0.1+das1.1gitc7a8c18.abi1.dtk2404.torch2.1

  • einops: 0.8.0

  • numpy: 1.24.3

  • pillow: 10.3.0

  • sympy: 1.12.1

  • mpmath: 1.3.0

  • networkx: 3.3

  • ninja: 1.11.1.1

  • Jinja2: 3.1.4

  • MarkupSafe: 2.1.5

  • filelock: 3.15.4

  • fsspec: 2024.6.0

  • typing_extensions: 4.12.2

  • certifi: 2024.6.2

  • charset-normalizer: 3.3.2

  • idna: 3.7

  • requests: 2.31.0

  • urllib3: 2.2.2

  • packaging: 24.0

  • pip: 24.0

  • setuptools: 68.2.2

Driver requirements

Requires DTK driver 24.04.

Key features

PyTorch 2.1.0, Torchaudio 2.1.2, and Torchvision 0.16.0.

Release history

August 2024: Released the PyTorch DTK image. The DTK version is 24.04.1.

PyTorch 2.0.1

Images

Image type Registry address
GPU ac2-registry.cn-hangzhou.cr.aliyuncs.com/ac2/pytorch:2.0.1-3.2304

Image content

pytorch:2.0.1-3.2304

CUDA libraries

  • cuda-cudart-12-1: 12.1.105

  • cuda-compat-12-1: 530.30.2

  • cuda-libraries-12-1: 12.1.1

  • cuda-opencl-12-1: 12.1.105

  • cuda-nvrtc-12-1: 12.1.105

  • cuda-nvtx-12-1: 12.1.105

  • cuda-cupti-12-1: 12.1.105

Acceleration libraries

  • libcublas-12-1: 12.1.3.1

  • libcufft-12-1: 11.0.2.54

  • libcufile-12-1: 1.6.1.9

  • libcurand-12-1: 10.3.2.106

  • libcusolver-12-1: 11.4.5.107

  • libcusparse-12-1: 12.1.0.106

  • libnpp-12-1: 12.1.0.40

  • libnvjitlink-12-1: 12.1.105

  • libnvjpeg-12-1: 12.2.0.2

  • libnccl: 2.18.3

  • libcudnn: 8.9.3.28

Python packages

  • Python: 3.10.13

  • torch: 2.0.1

  • torchvision: 0.15.2

  • numpy: 1.24.2

  • Pillow: 9.4.0

  • SciPy: 1.10.1

  • matplotlib: 3.7.1

  • sympy: 1.11.1

  • mpmath: 1.3.0

  • networkx: 2.8.8

  • Jinja2: 3.1.2

  • MarkupSafe: 2.1.2

  • filelock: 3.9.0

  • certifi: 2023.7.22

  • charset-normalizer: 3.1.0

  • idna: 3.4

  • requests: 2.31.0

  • urllib3: 1.26.18

  • PySocks: 1.7.1

  • packaging: 23.0

  • appdirs: 1.4.4

  • Brotli: 1.0.9

  • contourpy: 1.0.7

  • cycler: 0.11.0

  • fonttools: 4.39.2

  • fs: 2.4.16

  • kiwisolver: 1.4.4

  • libcomps: 0.1.19

  • lxml: 4.9.2

  • munkres: 1.1.2

  • olefile: 0.46

  • pycairo: 1.23.0

  • pyparsing: 3.0.7

  • python-dateutil: 2.8.2

  • six: 1.16.0

  • pip: 23.3.1

  • setuptools: 65.5.1

Driver requirements

Based on CUDA runtime 12.1.1. Requires NVIDIA Driver release 525 or later and is compatible with NVIDIA Driver release 470.

Key features

PyTorch 2.0.1 and Torchvision 0.15.2.

Release history

February 2024: Released the PyTorch 2.0.1 for CUDA 12.1 image.

Previous：Release notes for cuda-profiling imagesNext：Release notes for PyTorch-Ubuntu images