PyTorch 2.7 RC1 produced for pytorch, audio, vision

Release Announcements
1

Following the announcement of branch cut completion, the core binaries are now available for download and testing on pytorch test channel. Here is the 2.7 release issue tracker for cherry pick submissions.

REMINDER OF KEY DATES
Milestones M1 through M3 are complete and the next milestone is M4 at month end.

  • M4: Release branch finalized, Announce final launch date, Feature classifications published (week of 3/31/25) - Final RC is produced.
  • M4.1: Tutorial drafts submission deadline (4/9/25)
  • M5: External-Facing Content Finalized (4/18/25)
  • M6: Release Day (4/23/25)

DOWNLOAD INSTRUCTIONS FOR TESTING RC1

PIP CPU

  • Windows/Linux/MacOS:
    pip3 install torch==2.7.0 torchvision torchaudio --index-url https://download.pytorch.org/whl/test/cpu

PIP CUDA 11.8, 12.6, 12.8
Windows/Linux:

  • pip3 install torch==2.7.0 torchvision torchaudio --index-url https://download.pytorch.org/whl/test/cu118
  • pip3 install torch==2.7.0 torchvision torchaudio --index-url https://download.pytorch.org/whl/test/cu126
  • pip3 install torch==2.7.0 torchvision torchaudio --index-url https://download.pytorch.org/whl/test/cu128

PIP ROCM 6.2.4
pip3 install torch==2.7.0 torchvision torchaudio --index-url https://download.pytorch.org/whl/test/rocm6.2.4

PIP ROCM 6.3
pip3 install torch==2.7.0 torchvision torchaudio --index-url https://download.pytorch.org/whl/test/rocm6.3

PIP XPU

  • Linux/Windows:
    pip3 install torch==2.7.0 torchvision torchaudio --index-url https://download.pytorch.org/whl/test/xpu

PIP Linux Aarch6
pip3 install torch==2.7.0 torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/test/cpu

PIP Linux CUDA 12.8 Aarch64
pip3 install torch==2.7.0 torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/test/cu128

Libtorch

CPU Linux:

CPU Windows:

CPU MacOS:

CUDA 11.8, 12.6, 12.8 Linux

(cxx11 ABI):

Windows CUDA 11.8, 12.6, 12.8

(Release version):

(Debug version):

Docker images CUDA 11.8, 12.6, 12.8
Devel:

  • docker pull ghcr.io/pytorch/pytorch-test:2.7.0-cuda11.8-cudnn9-devel
  • docker pull ghcr.io/pytorch/pytorch-test:2.7.0-cuda12.6-cudnn9-devel
  • docker pull [ghcr.io/pytorch/pytorch-test:2.7.0-cuda12.8-cudnn9-devel](http://ghcr.io/pytorch/pytorch-test:2.7.0-cuda12.8-cudnn9-devel)

Runtime:

  • docker pull ghcr.io/pytorch/pytorch-test:2.7.0-cuda11.8-cudnn9-runtime
  • docker pull ghcr.io/pytorch/pytorch-test:2.7.0-cuda12.6-cudnn9-runtime
  • docker pull ghcr.io/pytorch/pytorch-test:2.7.0-cuda12.8-cudnn9-runtime

If there are any questions or concerns, please create an issue here.

Cheers,

Team PyTorch

2

It seems like the format needs to be a little bit refined by putting PIP Linux Aarch6 and PIP Linux CUDA 12.8 Aarch64 to the same level as other pip.

2 Likes
3

Thanks Eikan! Updated.

4

On my training codebase, I receive bug likely caused by a refcount error in a C extension error after an indeterminate number of steps, when using 2.7.0+cu126 or 2.7.0+cu128.

This does not happen on torch 2.6 or 2.8.0.dev20250403+cu128.

5

Hi there. Please create an issue here as soon as you can so this can be looked into. Please provide all details and logs if possible on the issue.
Thanks.

6

I have a questions.
Cuda aarch64 is based on SBSA or ARM64-tegra?